Perspective-Based Microblog Summarization ^†

Abstract

Social media allows people to express and share a variety of experiences, opinions, beliefs, interpretations, or viewpoints on a single topic. Summarizing a collection of social media posts (microblogs) on one topic may be challenging and can result in an incoherent summary due to multiple perspectives from different users. We introduce a novel approach to microblog summarization, the Multiple-View Summarization Framework (MVSF), designed to efficiently generate multiple summaries from the same social media dataset depending on chosen perspectives and deliver personalized and fine-grained summaries. The MVSF leverages component-of-perspective computing, which can recognize the perspectives expressed in microblogs, such as sentiments, political orientations, or unreliable opinions (fake news). The perspective computing can filter social media data to summarize them according to specific user-selected perspectives. For the summarization methods, our framework implements three extractive summarization methods: Entity-based, Social Signal-based, and Triple-based. We conduct comparative evaluations of MVSF summarizations against state-of-the-art summarization models, including BertSum, SBert, T5, and Bart-Large-CNN, by using a gold-standard BBC news dataset and Rouge scores. Furthermore, we utilize a dataset of 18,047 tweets about COVID-19 vaccines to demonstrate the applications of MVSF. Our contributions include the innovative approach of using user perspectives in summarization methods as a unified framework, capable of generating multiple summaries that reflect different perspectives, in contrast to prior approaches of generating one-size-fits-all summaries for one dataset. The practical implication of MVSF is that it offers users diverse perspectives from social media data. Our prototype web application is also implemented using ChatGPT to show the feasibility of our approach.

1. Introduction

In 2023, Twitter (now “X”) had around 450 million monthly active users, and 6000 tweets were posted on average every second [1,2]. When extrapolating the counts, there are 360,000 tweets posted every minute, 518 million tweets a day, and 189 billion tweets a year. Due to the large volume, it is infeasible for humans to review all relevant tweets if they want to investigate a certain topic or event. A number of machine learning and deep learning algorithms for social media summarization have been proposed, e.g., [3,4,5,6,7,8,9,10] using reinforcement learning along with attention layers and a deep learning-based model, based on using a recurrent neural network (RNN) and a Bi-LSTM (bidirectional long short-term memory) network. Others have used K-means clustering, and a Twitter Online Word Graph Summarizer for a set of related tweets.

However, social media posts can talk about the same topic or content but reflect different user perspectives, such as negative or positive sentiments, emotions, biases, political views, distorted opinions (so-called fake vs. real news), etc. Thus, providing a single summary of social posts may be misleading or not represent different perspectives of users, such as their emotions, opinions, and attitudes toward content.

In addition, despite the potential that social media platforms have to democratize access to diverse social and political perspectives [11], the meteoric rise of these platforms has further fueled the creation and ossification of “echo chambers” [12,13,14]. Echo chambers are environments where the opinions, political leanings, or beliefs of users about a topic are reinforced due to repetitive interactions with peers with similar tendencies and attitudes [15]. The echo chamber effect usually arises concerning controversial topics, e.g., gun control, vaccinations, abortion, school prayer, etc. [15]. Spending time in communities of like-minded individuals not only raises individuals’ exposure to pro-attitudinal messages but also decreases their exposure to counter-attitudinal information. This leads to the issue of echo chambers, where citizens do not see or hear different topics or ideas. This issue limits their capacity to reach common ground on political issues [16].

A number of methods have been researched to overcome echo chamber effects, for example, increasing the diversity of information sources [17]. Receiving information from different sources will help users better understand a topic from different perspectives, develop more in-depth opinions, and make better decisions [18,19]. However, users often restrict their personal exposure to only like-minded individuals or platforms [20]. According to [21], the primary cause of polarization and extremism is the loss of truly public platforms/forums. Therefore, recreating a new environment where people are more likely to encounter opposing perspectives and opinions could counteract this tendency, as having trust in “the media” is likely to increase confidence in seeking and receiving information from diverse sources [22]. It is plausible that the greater a person’s media trust, the more likely they are to explore different views from these sources, thereby becoming less susceptible to echo chambers [20].

In this paper, we propose the perspective-based summarization approach to address the challenges posed by the overwhelming volume and echo chamber issues in social media. This approach is designed to summarize and compare the subtle nuances between different viewpoints, e.g., the dissemination and reception of fake and real news. By employing this comparative analysis, our framework aims to highlight the discrepancies between authentic and manipulated content, which are important in our big data world. The proposed summarization framework, the Multiple-View Summarization Framework (MVSF), generates different perspective-based summaries from the same set of social media posts based on user-centered views, opinions, or emotions by flexibly combining those to match different desired perspectives expressed in social media posts. Providing summaries with different perspectives of the same social media content may help individuals understand the contrary opinions and diverse perspectives on the same topic.

In our MVSF, a “perspective” refers to a distinct viewpoint or angle interpreted or analyzed. This encompasses a range of elements such as sentiments (both positive and negative), political views, and the authenticity of information (distinguishing between fake and real news). The framework identifies these perspectives through a combination of natural language-processing techniques, including sentiment analysis, and fact-checking models. Each perspective is integrated into the summarization process by filtering content that aligns with the user-defined perspectives. This approach allows for personalized summaries that reflect user interests. Therefore, the approach enhances the relevance and engagement of the summary.

With the MVSF, users can obtain summaries reflecting specific interests and viewpoints expressed on social media. For example, a positive summary on Pfizer vaccines in COVID-19 related posts, or a summary with only negative sentiment regarding the same vaccines. These can help policymakers realize how the vaccines are perceived by different groups. The summaries can also allow further content analyses, to gain an understanding of positivity vs. negativity or who the influential figures are in these different perspective summaries.

To address the issue of echo chambers, our MVSF is designed to expose users to a broad range of viewpoints. By aggregating and summarizing diverse perspectives, the MVSF ensures that summaries include viewpoints from across the ideological spectrum. Thus, the MVSF provides users with a balanced view of the topic. This is achieved through algorithms including identifying and quantifying sentiment biases and diversifying other perspectives included in the summaries. Such an approach enriches the user’s understanding as well as fosters more informed discussions. The integration of diverse perspectives is important in combating the polarization and misleading information prevalent on social media.

Our MVSF has two components, summarization and perspective computing components. The summarization component summarizes the social media data using one of three summarization methods: entity-based (EbS), triple-based (TbS), and social signal-based (SbS). The perspectives computing component is intended to detect specific user perspectives expressed in social media posts, such as negative sentiments. The combination of two components enables a wide range of fine-grained summaries, such as a summary of only negative sentiment views whose focus is the topical entity. As another example, it can combine a triple-based summary with positive sentiments.

The summarization methods use the semantic analysis of each social media post in the form of <subject, predicate, object> triples [23]. We show that the semantic analysis-based summarization outperforms state-of-the-art extractive and abstractive summarization models from the literature, as indicated by better Rouge F-1 scores [24] of around 14%.

The perspectives can be combined to convey summaries with multiple perspectives. For example, when a journalist writes a story on social media posts on a politician’s policy, they can focus on one perspective, such as a summary of negative opinions about the policy, or a summary of user posts that are negative and fake opinions about the policy. In this way they can convey summaries that contain one or multiple perspectives. By employing MVSF’s various methods and perspectives, users can navigate widespread social media content, gaining deeper insights and enhancing information extraction. The framework’s adaptability underpins its significance in providing comprehensive and user-centric summarization solutions. The work in this paper contributes in the following ways:

• We have developed three summarization methods for social media posts: entity-based, triple-based and social feature-focused summarization.
• We have developed the Multiple-View Summarization Framework (MVSF), which is capable of generating summaries based on multiple perspectives, such as political orientations, sentiments, or fake opinions, combining them with any of the three summarization methods. The framework provides greater flexibility of choosing summarization methods as well as combining different perspectives, generating fine-grained and personalized summaries tailored to the end-user’s preferences.
• Theoretical implications are the integration of multiple perspectives and extractive methods within a unified framework, in contrast to one-size-fits-all summarization. This enables comprehensive information extraction from social media content and constitutes an extended paradigm for summarization research. On the practical side, we have developed summarization algorithms for the MVSF to allow users to tailor summaries to their needs.
• Through extensive performance experiments, we compared the summaries generated by our entity-based and triple-based methods, both independently and in conjunction. When benchmarking the results against prominent summarization models from the literature, such as Bart-large-CNN 5, Text-To-Text-Transfer-Transformer (T5) [25], BertSum [26], and SBert [27], our summaries achieved a performance improvement of 14% in terms of Rouge scores.
• Our framework was effectively applied to an X/Twitter dataset of 18,047 COVID-19 vaccine-related tweets, demonstrating the flexibility of presenting different summaries of this topic.
• We leverage our MVSF to present a comparative analysis of fake and real news, to enhance its utility in discerning truth and misleading information distributed across social media platforms, which is increasingly important in today’s world.
• We have developed a user-friendly web prototype (http://ai4sg.njit.edu/ai4sg/Summarize, accessed on 1 September 2023) leveraging the power of OpenAI ChatGPT (text-davinci-003) [28] for text summarization. This application works within the MVSF, while not suffering from the hardware-intensive resource use of our algorithms. The application allows users to input text and customize summarization preferences based on the MVSF’s methods and perspectives, serving as an accessible perspective-based summarization tool.

These contributions collectively underscore the significance and practical applicability of the MVSF in addressing the challenges of efficiently summarizing extensive social media content, while catering to diverse user needs. Our summarization approach is novel as opposed to a one-size-fits-all summarization algorithm.

This paper is organized as follows. In Section 2, we provide a synopsis of related work. We present our summarization framework in Section 3. In Section 4, Section 5 and Section 6, we detail each component of the MVSF, including knowledge triple extraction and processing, perspective analysis and detection models, and our summarization methods, respectively. The evaluation of our summarization method is presented in Section 7. We present the application of our summarization framework to a large dataset, the results, and our findings in Section 8. We compare and present the findings of summaries of fake news and real news in Section 9. In Section 10, we present our web prototype application, which leverages the power of OpenAI ChatGPT [28] to perform text summarization within our MVSF. Section 11 contain the Discussion, Conclusions, and Future Work.

2. Related Work

In this section, we introduce general social media summarization approaches along with different applications, as well as echo chamber effects on social media.

2.1. General Summarization Approaches

Researchers distinguish between abstractive and extractive summarization methods [29]. Abstractive summarization generates a summary by capturing the prominent ideas of the source text. The summaries contain new sentences that do not exist in the original text. The abstractive summarization methods in the literature include Bart-large-CNN [30], Text-To-Text Transfer Transformer (T5) [25], FactSumm [31], FAIRSEQ [32], PEGASUS [33], XNLG [34], ChatGPT [28], GPT-2 [35], etc. On the other hand, extractive summarization selects a subset of the sentences that are able to best represent the original document. The extractive summarization models in the literature include BertSum [26], SBert [27], RankSum [36], HAHSum [37], NeRoBERTa [38], DebateSum [39], MemSum [40], Gensim [41], etc. While abstractive summarization is beneficial in situations where high rates of compression are required [42], microblogs are the antithesis to long documents. Abstractive systems usually perform best in limited domains, since they require outside knowledge sources. The abstractive approaches might not work so well with microblog posts, as they are unstructured and diverse in their subject matter. Furthermore, the summaries generated by abstractive models usually face factual inconsistency problems [43]. There is also a huge inference speed gap between the abstractive and extractive summarization methods [44]. Extractive techniques are known to scale better in highly diverse domains [45]. It is also beneficial to work with tweet fragments rather than entire tweets [46]. Therefore, in this paper, we work on extractive summarization along with this tweet post fragment approach.

To specifically summarize large content, the LOCOST model [47] is introduced as a recent advancement in handling extensive texts. LOCOST is designed to manage documents exceeding 600,000 tokens. This capability makes it suitable for summarizing entire books or comprehensive documents longer than the typical input length limitations of models.

Joshi et al. [36] proposed a method based on LDA (Latent Dirichlet Allocation) [48] topic modeling and word embedding [49] for the extractive summarization of single documents. Their work, based on CNN Daily Mail datasets [50], achieved a state-of-the-art performance at the time. LDA and word embedding are effective when sentences are coherent; however, when sentences are not coherent, which often happens in social media, extra contextual information is needed [51]. One of our summary methods identifies contextual information and bridges this gap.

Our MVSF integrates user perspectives such as sentiments, political biases, and authenticity, which these traditional models often overlook. This personalized approach allows the MVSF to produce summaries that reflect the nuanced views of social media users, enhancing the relevance and comprehensiveness of summaries.

2.2. Microblogging Summarization Approaches

Many algorithms for social media summarization have been proposed, e.g., [3,4,5,6,7,8,9,10]. Yadav et al. [3] used reinforcement learning along with attention layers and a deep learning-based model, by using a recurrent neural network (RNN) and a Bi-LSTM (bidirectional long short-term memory) network for summarization tasks. Their work achieved a state-of-the-art performance when evaluated with BLEU and ROUGE scores. Modhe et al. [4] made use of TF-IDF encoding to obtain both single-post and multi-post summaries of Twitter activity, based on the rankings of words and a user-defined threshold. Geng et al. [5] proposed a microblogging cluster stream (Microblog Cluster Vectors) and a ranking method, by using K-means clustering [52] and query-lexical-rank to generate a query-focused extractive summaries of Twitter data. Dutta et al. [7] performed the first systematic analysis of approaches towards summarizing Twitter posts during disasters. They found that different algorithms applied to the same input yielded summaries with significant differences, which is superficially similar to our results. However, we achieve different summaries intentionally and as part of a unified framework. Olariu et al. [8] introduced a Twitter Online Word Graph Summarizer, which was the first online abstractive summarization algorithm for tweets. In their experiments, for a set of related tweets, they generated a high-quality summary. However, when applied to unrelated tweets, the initially generated summary lacked meaning. This happened because event-related signals (in their case, bigrams) stand out when analyzing similar tweets. As a better solution, they built a word graph from trigrams to solve the issue.

Sharifi et al. [9] chose an extractive approach, since it is more appropriate for the structure and diversity of microblogs. They first applied Phrase Reinforcement [53], which generates summaries by looking for the most commonly occurring phrases. Then, they processed the results of the Phrase Reinforcement approach with their proposed Hybrid TF-IDF algorithm, where the TF (term frequency) component is computed from the entire collection of posts, whereas the IDF (inverse document frequency) component is computed from a single post. Inouye [10] proposed a clustering-based algorithm and a threshold-based Hybrid TF-IDF algorithm. The first step performs clustering by combining the K-Means++ algorithm [54] with the bisecting K-means algorithm [52] to cluster posts into subtopics, and then each cluster [55] is summarized individually. The second step uses the modified Hybrid TF-IDF summarization algorithm [9], so that it can produce multiple post summaries. Originally [9], the algorithm only selected the best summarizing topic sentence, but later, Inouye [10] modified the method to select the top four highest weighted posts.

2.3. Other Summarization Approaches

Gunaratna et al. [56] selected related features among entities while maintaining the diversity and saliency of features within entity sets. By selecting (i) inter-entity facts that are similar and (ii) intra-entity facts that are important and diverse, the approach summarizes facts about a collection of entities. Entity summarization has been categorized into extractive and non-extractive methods. Among extractive methods, the summarization is further divided into single-entity and multi-entity categories. For single-entity categories, Gunaratna et al. [57] proposed FACES to incorporate diversity in summarization. For multi-entity categories, FACES-E [58] uses focus term detection and aligns these focus terms with ontology classes and entities present in knowledge graphs [59]. FACES-E showed the usefulness of type-computed literals in creating comprehensive entity summaries. For the non-extractive categories, REMES was introduced [56] to maximize the relatedness of facts between entity summaries and importance and diversity of facts within each entity summary. One of our methods performs entity-based summarization.

Amplayo et al. [60] presented an abstractive opinion summarization model that generates input aspect-based summaries for a set of product or hotel reviews. Their aspect-based summaries use transformer models, and for the generation of summaries, users need to input a specific aspect code or keywords. Li and Chaturvedi [61] presented the Rationale-based Opinion Summarization system (RATION), which generates summaries of user reviews by extracting the representative opinions as well as one or more corresponding rationales as supporting details for opinions. They use transformer-based opinion extraction and Gibbs sampling to sample a user-specified number of sentences as rationales by approximating this joint probability distribution.

Unlike most of the approaches in the literature, which focus on a single aspect, our framework focuses on separate or combined perspectives expressed in social media posts, and it ranks the extracted information based on respective criteria to obtain fine-grained extractive summaries.

3. Multiple-View Summarization Framework

The MVSF is designed to address the complex nature of social media data, allowing for customized summaries based on multiple perspectives. In the MVSF, the terms “views” and “perspectives” are used synonymously to describe the different angles of content analyzed and summarized. Figure 1 illustrates the general architecture of our framework, which consists of the following components:

Data Cleaning and Semantic Triple Extraction Component: The collected data undergo standard preprocessing, which includes the removal of non-ASCII characters, redundant spaces, and URLs. To capture the essence of social media posts, such as described events or activities [62], we implement semantic triple extraction. This process extracts <subject (S), predicate (P), object (O)> triples from each post. This structures the content by identifying key entities, actions, and relationships. These triples serve as the foundational data, and they are used to generate perspective-based summaries.

View/Perspective Component: This component analyzes the extracted text to identify multiple perspectives, such as sentiments or potential misinformation. Using advanced NLP techniques, it performs sentiment analysis to gauge public emotions and perceptions [63] and applies fake news detection algorithms to identify and flag potentially misleading information. The outcomes of this component are tagged with metadata, which categorize each post according to detected sentiments, biases, or factual accuracies.

Summarization Component: After perspectives are identified, the summarization component processes the enriched data to generate summaries. It employs three distinct methods: Entity-based (EbS) summarization focuses on key individuals, organizations, or locations mentioned in the content. Triple-based (TbS) summarization groups posts related to specific events or statements. Social signal-based (SbS) summarization leverages social media interaction metrics such as follower counts and retweets to highlight content with popularity. This component integrates the perspectives from the previous component that match the user-defined criteria for summarization and reflect the desired perspectives.

Output Component: The final summaries are generated based on the users’ selection of summarization methods and perspectives. Users have the flexibility to select from simple sentiment-driven summaries to complex composites that merge various perspectives, such as combining sentiment analysis results with fake news detection. For instance, a user might request a summary focusing solely on negative sentiments about a policy, or a combined summary addressing both negative sentiments and misleading information.

Users can utilize the MVSF to generate multiple summaries from the same social media posts, reflecting different perspectives or a combination of perspectives. This offers multiple views of the discourse surrounding a topic. This capability also highlights the framework’s adaptability and the potential to enhance decision-making processes in various analytical scenarios.

4. Semantic Triple Extraction and Cleaning

The Semantic Triple Extraction and Cleaning Component (in Figure 1) is a fundamental and important component in the MVSF. Semantic triple extraction transforms unstructured social media text data into structured knowledge triple representations. Social media content usually contains irregular sentence structures, hashtags, URLs, and abbreviations [64]. Thus, it is challenging to perform analysis directly on social media content. To address this issue, we extract meaningful insights by capturing relationships between key entities in the form of <subject; predicate; object> (SPO) triples. The triples capture events and actions expressed in the text, and they serve as the foundation for generating diverse summaries in our framework. For instance, a post like “The CDC announced new vaccine guidelines” would yield a triple <CDC; announced; new vaccine guidelines>, which showcases the key information in the sentence.

We used the Stanford Open Information Extraction (Open IE) model [65] to extract the triples. The model first breaks down sentences into shorter logical clauses. The model then identifies the <subject; predicate; object> for each clause, and it predicts whether an edge in the dependency graph should form a new independent clause [66]. For example, the sentence “The CDC and WHO announced that vaccines will be distributed by March” can be broken into two triples: <CDC; announced; vaccines distribution> and <WHO; announced; vaccines distribution>. Each extracted triple is then converted to lowercase to achieve a unified text representation, ensuring consistency across the dataset.

5. Perspective Analytics for Summarization

The Perspective Detection Component in our framework is used to identify different perspectives expressed in each social media post. Perspective detection can use different methods. For instance, a negative perspective can be identified by sentiment analysis, and a fake content view can be detected by a machine learning model to classify a post as fake or not. These perspectives are essential to provide unique insights, such as a summary with a particular sentiment, or a summary of a particular political bias. In this section, we provide the perspective analysis approaches we used. With the proliferation of AI models available with APIs, our framework assumes that the appropriate perspective detection model is available to be utilized for the summarization.

5.1. Sentiment Perspective Analysis

To distinguish between different emotions expressed from the text, we performed sentiment analysis using the Stanford Sentiment Analyzer [67]. The choice of that analyzer was based on human evaluation [63] among the Stanford Sentiment Analyzer, VADER [68], and TextBlob [69]. In addition, based on the literature, the Stanford Sentiment Analyzer has the best accuracy among the three, at 80.7% [67], while VADER has an accuracy of 76.8% and TextBlob has an 68.8% accuracy [70]. Thus, we chose the Stanford Sentiment Analyzer.

The Stanford Sentiment Analyzer labels each input phrase either as “(Very) Negative”, “Neutral”, or “(Very) Positive”. With sentiment labels, we can choose to focus only on the <S, P, O> triples with a specific sentiment, e.g., triples expressing negative sentiments towards mandatory COVID-19 policies.

5.2. Fake News or Real News Perspectives

Social media platforms have been flooded with misinformation, so-called fake news, which confuses citizens, causes conflicts, and drowns out authentic information [71]. Therefore, summarizing large fake news item sets can produce a readable summary. This will enable concerned users to counteract the spread of misleading information. That can be achieved by training a machine learning model to classify social media posts and collecting those labeled as fake news.

5.3. Political Bias Perspectives

To systematically discern the political biases in a posted text, we can employ a machine learning model trained on data labeled either as ‘left’ or ‘right’ political viewpoints or biases. This binary classification approach, inspired by the approaches in [72], allows the detection of political orientations within content. By integrating this method into our framework, we enhance the granularity of our summarization, ensuring a comprehensive representation of the political landscape within the data.

In addition to these, one can consider contextual perspectives such as temporal or location-specific perspectives for summaries of the public reactions to restrictive government COVID-19 policies, e.g., right after their introduction and again after six months. This could allow us to evaluate the acceptance or rejection of a policy, which would allow government agencies to fine-tune the policy. This view can be derived from the time stamps of social media posts. A summary based on specific locations, or at different administrative levels, such as country, state/province, etc., can capture location-specific perspectives [73], which can be compared with summaries from different locations to determine differences.

5.4. Composition of Multiple Perspectives

Our MVSF allows users to select and combine multiple perspectives to generate customized summaries. This showcases its adaptability to the complex nature of social media data.

User-Driven Composition: In the MVSF, the composition of perspectives is entirely user-driven, allowing for a customized summarization. Users can choose and combine different perspectives to generate a summary, i.e., a fake news-only view with negative sentiments. The composite perspectives are achieved through aggregation:

c_v = compose (v₁, v₂, …, v_n)

where n is the number of perspectives and v_i is a single view or perspective. This process emphasizes the framework’s flexibility, as it accommodates a combination of perspectives without internal weighting or prioritization.

Handling of Perspective Outputs: Each perspective operates independently and analyzes social media content to produce insights from its specific viewpoint. When combined, the outputs are presented collectively, without automated conflict resolution. This allows users to view and interpret composite summaries that incorporate diverse analytical angles, therefore enabling informed decision-making.

Independence from Summarization Methods: While perspectives are combinable, the summarization methods within the MVSF operate independently. For example, the entity-based method focuses on key entities, the triple-based method organizes data around specific events, and the social signal-based method highlights content with significant social engagement. These methods are designed to address different aspects of the data, and they are used individually or combined with selected perspectives.

6. Microblogging Summarization Methods

6.1. Entity-Based Summarization (EbS)

In this method (EbS), the entities define the primary subjects and objects for obtaining summaries. Unlike in other research where the user provides the entities of interest (e.g., “Joe Biden”), we discover the most prominent entities mentioned in the social media posts by first finding the triple verbs that contain the same semantic events, and by identifying the salient entities in triple subjects and objects.

• Identify the same events in triples using WordNet: Important entities can appear in semantically similar events that are expressed by different predicates/actions. Thus, we need to identify predicates expressing similar or identical meanings with different verbs, such as “offer”, “pass”, and “transfer”, or different forms of the same verb, e.g., “provide”, and “provided”, all of which express a similar meaning of “giving something to someone”. Groups of semantically similar verbs can be subsumed by one root verb, using the appropriate synset from WordNet [74]. A synset is a set of one or more synonyms. WordNet organizes synset into generalization hierarchies. A verb in a hierarchy that is the most general is referred to as a root verb. For triples using verbs expressing similar/identical meanings with different words or tense forms, we replaced the verbs as follows. To disambiguate the meaning of a predicate and find the closest synsets, we compared two methods and used a human evaluator to determine the better one. The first method is the Lesk algorithm [75]. Given a verb and the triple where it occurs, Lesk returns a synset that represents its context meaning. However, Lesk often failed at finding the correct synset. Among a set of 80 randomly picked triples, only 37 verb synsets were correctly identified, according to the human evaluator. The second method was that we selected “v.01” (primary meaning returned by WordNet) as the verb synset. Although this is a simple approach, it produced a better result (78/80) according to the human review. Therefore, we used the “v.01” meaning for each verb to find the root synset. For each synset vx.v.01 of the verb vx, we followed its hypernym (superclass) chain upward until reaching its root synset. We identified the frequent events by selecting triples whose root verbs occur more often than a threshold θ. Given a triple set TS = {ts₁, ts₂, …} where ts_i = <s, p, o>, find ts_j = <s, root(p), o>. We select triples that have frequent root verbs root(p), i.e., frequency(root(p)) > θ. We present the program to identify the synset hierarchy in

  Table 1. Pseudo code to identify the root verb synset by using WordNet.

  highestSuperclass = []

  for v in tripleVerb:

  currentV = v + ‘.v.01’ # we apply v.01 (most natural) synset to all triple verbs

  while (len(wordNet.synset(currentV).hypernyms()) == 1):

  currentV = wordNet.synset(currentV).hypernyms()[0] # replace current synset

  with its direct superclass synset

  highestSuperclass.append(currentV)

  , and an example synset hierarchy in Figure 2, where all verbs are summarized by “give.”
• Identifying Salient Entities: The second task in this method is to identify important entities to focus on. Each triple includes two entities, a subject and an object. As noted, this method is based on the entities (subject/object) that occur most often in the triple sets. To identify the frequencies of meaningful words, for each entity, we removed numbers, stop words, and words with fewer than three characters and performed lemmatization. If a preprocessed entity became an empty string, we removed its triple from the triple set. If a preprocessed subject consisted of more than two words, we removed all words except for the last two (which are likely to contain “most of the meaning”).

Each subject is labeled with its frequency in the triple set. If there is more than one word in a subject, we obtain the subject score based on the word with the higher frequency. For example, if there are 77 mentions of “COVID” and 100 of “vaccine,” the subject score of “COVID” is 77, of “vaccine” is 100, and of “COVID vaccine” is also 100. The same method is applied to objects for computing object scores. The preprocessing and score calculation of triple entities is shown in

Table 2. Pseudo code for identifying top entities.

def preprocess (entity):

entity = remove stop words and numbers from entity

if (length of entity) < three characters:

entity = empty string

entity = lemmatization (entity)

return entity

subjectWord = [ ] // a list that collects all words in triple subjects

objectWord = [ ] // a list that collects all words in triple objects

for each subject s, object o, triple t in triple set ST: // (s, o) is the (subject, object) of triple t in ST,

s = preprocess (s)

o = preprocess (o)

if s == empty string or o == empty string:

remove t from ST

s = s.split(“ ”)[:-2] // if s have more than two words, remove all except for the last two words

subjectWord.extend (s.split(“ ”))

objectWord.extend (o.split(“ ”))

for each word x in subjectWord:

label x with an integer score of its frequency in subjectWord

for each word y in objectWord:

label y with an integer score of its frequency in objectWord

for each subject s, object o, triple t in triple set ST: // (s, o) is the (subject, object) of triple t in ST

Subject Score of s = highest score of the word in s

Object Score of o = highest score of the word in o

.

To identify the summary with the best accuracy, we experimented with assigning different weights (α) to the subjects and (1 − α) to the objects. We select the top-scoring m_E triples and the corresponding original sentences. The parameter m_E controls the length of the summary.

$$\mathit{T}\mathit{r}\mathit{i}\mathit{p}\mathit{l}\mathit{e}\mathit{S}\mathit{c}\mathit{o}\mathit{r}\mathit{e}=\alpha \times \left(Subject Score\right)+\left(1-\alpha \right)\times \left(Object Score\right)$$

(1)

where α ∈ [0.0, 1.0].

6.2. Triple-Based Summarization (TbS)

In this method, we capture the important contextual meaning (the whole statement instead of entities) by using triple sentence representations. To create sentence representations, we use BERT [76] sentence embeddings. BERT reads the entire input sequence at one time, which allows BERT to learn the contextual information of each word based on its neighboring (left and right) words. We used an autoencoder [77] to learn a 32-dimensional vector representation to ensure that we achieved a lower-dimensional representation [51]. Instead of focusing on specific entities, we produced a summary based on important statement information in triples. We used the distance measures to the centroid of the triple vectors to select the salient triples for summarization.

For each triple t, its vector representation v is

$$v_t=[v_{t1},v_{t2},v_{t3},\dots ,v_{t32}]$$

(2)

The centroid vector $v_c$ of all triple vectors is calculated:

$$v_c={\displaystyle \frac{1}{n}}\left({\sum }_{t=1}^n{v}_t\right)=[v_{c1},v_{c2},v_{c3},\dots ,v_{c32}]$$

(3)

where n is the triple count.

The Euclidean distance d from a triple t to the centroid c is

$$d_{tc}=sqrt\left({\sum }_{i=1}^{32}{\left(v_{ti}-v_{ci}\right)}^2\right)$$

(4)

We selected triples with the shortest distances to the centroid c as top summarization triples. The original sentences corresponding to the selected triples were recovered to form the summary.

6.3. Social Signal-Based Summarization (SbS)

In this method (SbS), we exploit the social signals (i.e., a tweet’s retweet count, and its poster’s follower count) to identify the salience of tweets. A tweet is more important if it is (1) posted by a user with many followers, and (2) retweeted many times, according to [62], where the salience score of a tweet is the multiplication of the retweet count, user follower count, and readability. Our goal is to identify summaries with social prominence on social media. Thus, we modified the formula and defined the salience score of a tweet:

$$\mathit{S}\mathit{a}\mathit{l}\mathit{i}\mathit{e}\mathit{n}\mathit{c}\mathit{e}\mathit{S}\mathit{c}\mathit{o}\mathit{r}\mathit{e}=\left\{\begin{array}{c}follower+retweet\times 707, follower>0\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}0, follower=0\end{array}\right.$$

(5)

where 707 is the average number of followers of a Twitter user [9]. When a post is retweeted, there will be on average 707 people who will see it. We rank triples based on the scores of their original tweets and select the m_SF top-scoring triples. The corresponding original sentences of the triples are selected to form the summary.

We also acknowledge the limitations of using social signals such as followers and retweets to represent the importance of content. A study “The Million Tweets Fallacy” [78] indicates that engagement metrics such as retweets and followers may not always correlate with the content relevance or quality. These metrics often reflect visibility more than value, leading to potential biases in assessment [79].

7. Evaluation of Microblogging Summarization Methods

To evaluate our summarization approaches, we compared them with extractive and abstractive summarization models from the literature. The extractive models are BertSum [26] and SBert [27], variants of BERT used in our method (TbS). Abstractive summarization models are Bart-large-CNN [30] and T5 [25], which according to [63], generate better summaries than TextRank [41] and GPT-2 [35].

In our pursuit of evaluation, we encountered a challenge: we could not locate a social media dataset with a gold-standard summary. Nonetheless, we could perform an evaluation by leveraging an alternative dataset—the BBC news items set [80]. Originally designed as a benchmark for machine learning research, we repurposed this dataset to evaluate the performance of summaries in our specific task. The dataset comprises 2225 documents, each sourced from the BBC news website, spanning five diverse topical areas, namely business, entertainment, sport, politics, and technology. This dataset serves as a trustworthy foundation for our assessment of summary generation by providing human extractive summaries for each document.

We selected 20 news documents from the business category. We summarized them using our entity-based (EbS) and triple-based summarization (TbS) approaches, with and without a sentiment perspective (S). Because this dataset lacks social signals, we excluded the SbS method from our evaluation process. Each of our generated summaries adheres to a consistent length of approximately 300 words, aligning with both our chosen methods and the published results.

We employed Rouge scores [24] to quantify performance, which measure the overlap between the generated summaries and human-crafted reference summaries. The evaluation focused on 1-g (unigram overlap), 2-g (bigram overlap), and LCS (Longest Common Subsequence). Our summaries based on an entity and triple with or without a perspective outperformed those by the benchmarking models (

Table 3. Rouge scores of summarization models.

Rouge Score	1-g	2-g	LCS
Method (EbS), α = 0.7	0.465	0.310	0.444
Method (TbS)	0.557	0.427	0.551
TbS with View (S)	0.58	0.416	0.572
EbS with View(S), α = 0.7	0.459	0.299	0.433
BertSum	0.406	0.174	0.394
SBert	0.444	0.152	0.428
bart-large-cnn	0.402	0.150	0.394
T5	0.304	0.104	0.304

). The triple-based summaries with a sentiment view, i.e., TbS + S performed best with the highest score for 1-g and LCS, while the method TbS performed best for 2-g. This result shows that our summaries are more consistent with the gold-standard summaries than existing models. The inclusion of negative sentiment analysis in our TbS model helps with focusing on emotionally charged and impactful phrases, which are often important in news reporting. This focus aligns the summaries more closely with the nuances and tones of the source documents.

8. Application of MVSF to COVID-19 Vaccine Tweets

This section demonstrates the application of the MVSF to a dataset of COVID-19 vaccine-related tweets. Our objective is to illustrate how the MVSF captures diverse perspectives, providing insights on critical health issues like vaccinations.

8.1. Data

We used 18,047 tweets [81] about widely used COVID-19 vaccines worldwide, Pfizer/BioNTech, Sinopharm, Sinovac, Moderna, AstraZeneca (AZ), Covaxin, and Sputnik V. The tweets were posted between December 2020 and November 2021 (this dataset may not fully represent all tweets concerning COVID-19 vaccines). The dataset’s columns include tweet ID, tweet content text, timestamp, user ID, retweets, likes, hashtags, and user locations.

8.2. Preprocessing

We removed non-ASCII codes, URLs, and redundant spaces. We assigned each post an integer post index. We then performed sentence tokenization [82] to split posts into sentences, because a tweet may contain multiple sentences. It is beneficial to work with post fragments (sentences) rather than entire posts [46]. We assigned to each sentence an integer sentence index, as this sentence tokenization is used for tracing back to the corresponding original sentences from the selected knowledge triples.

Table 4. A post with two sentences in the dataset.

Post Index	Post Text	Sentence Index	Sentence Text
484	A shipment of Sputnik V vaccine arrived in Vietnam. The handover ceremony took place at Noi Bai airport.	0	A shipment of Sputnik V vaccine arrived in Vietnam.
		1	The handover ceremony took place at Noi Bai airport.

shows an example of a post with two sentences. We obtained 28,242 sentences from 18,047 posts.

8.3. Triple Extraction and Cleaning Process

Using triple extraction methods (OpenIE) [65] from the Stanford CoreNLP package [83], we extracted 101,432 triples and linked them to their sentences and their original posts.

We further filtered out triples with auxiliary verbs that carry little meaning, e.g., “be” or “have”. This reduced the triples from 101,432 to 67,049. In a number of cases, there were overlapping triples from the same sentence, but of different lengths. For triples from the same post and sentence, and using the same verb, we eliminated all but the one triple, retaining the most information (words). The final 16,270 triples were used as the input for each summarization method.

8.4. Experimental Results

This subsection presents the application of the MVSF to COVID-19 vaccine-related tweets, demonstrating how the framework generates summaries from social media data. The following subsections illustrate the summarization results both with and without integrating specific perspectives, showcasing the flexibility of the MVSF in capturing different viewpoints.

8.4.1. Summaries Without Perspectives

We use all the 16,270 triples (from Section 8.3) to generate a summary for each method (EbS, TbS, SbS), respectively.

Figure 3 shows a knowledge graph [59] of the summary triples using entity-based summarization (EbS). We used Gephi, Version 0.10.1 [84] as our visualization tool. The size of each node is proportional to its degree, with a size range (10, 20) implemented in Gephi. The edge color expresses the sentiment expressed by the triple, with red (=1) for “Negative”, yellow (=2) for “Neutral”, and green (=3) for “Positive.” The edge thickness represents the triple’s rank. The thicker an edge, the higher its rank in the summary set (i.e., higher (triple/salience) score, or shorter distance to the centroid). For the summary generation, we selected the top-scoring m_e triples from the triple set, with their corresponding original sentences, such that the total number of words in the summary was approximately 300. The details of the summary triples are in

Table 5. Summary triples of method (EbS), their triple scores, sentiments, and vaccines, in rank order.

Triples	Sentiment	Triple Score	Vaccine
j&j vaccine is essentially first shot of sputnik v vaccine	1	1646	sputnik
new recombinant COVID-19 vaccine developed by national vaccine	3	1646	sinopharm
COVID-19 vaccine of batch is international vaccine campaign co-led by world health organization	1	1646	sinopharm
vaccine be listed on whos list of approved vaccines	2	1646	covaxin
covaxin shows vaccine efficacy of 81% in phase 3 trial	1	1153.2	covaxin
covaxin will from june 21 will most expensive of three vaccines	1	1153.2	covaxin
moderna seeks regulatory approval for its COVID vaccine in india	1	1129.4	moderna
moderna will fourth vaccine used for vaccination drive in india	1	1129.4	moderna
sputnikv joins indias vaccine utsav	2	1028.6	sputnik
india will have its third vaccine	2	1002.7	sputnik
china has approved emergency use of sinovac biotech’s COVID-19 vaccine in people aged	1	999.2	sinovac
other countries are also relying heavily on pfizer vaccine	2	994.3	pfizer
bharat biotech had submitted eoi for its vaccine andwho informed in document	2	978.2	covaxin
people develop reaction about week after moderna vaccine	2	974	moderna
bharatbiotech support worldwide amid COVID-19 surge coronavirus vaccine india bharatbiotech	1	962.8	covaxin
fda warning to moderna COVID-19 vaccine shots	1	939.7	moderna pfizer

. The corresponding original sentences of the selected summary triples are in

Table 6. Entity-based summary (in temporal order).

Other countries in the Middle East, including Saudi Arabia, Qatar, Kuwait, and Oman, are also relying heavily on the Pfizer vaccine, developed by the US company PfizerVaccine. This is also another validation of Sputnik V pioneering technology as J&J vaccine is essentially a first shot of Sputnik V vaccine (Ad 26—human adenoviral vector 26). Covaxin shows vaccine efficacy of 81% in phase 3 trial. About a week after the Moderna vaccine, some people develop a reaction. The new recombinant COVID-19 vaccine, developed by the National Vaccine &Serum Institute, a R&D center of Sinopharms bioscience subsidiary the China National Biotec Group, got approval from the National Medical Products Administration on April 9. SputnikV joins Indias vaccine utsav. Once DCGI approves SputnikV, India will have its third vaccine after Covishield and Covaxin. Bharat Biotech announces COVAXIN capacity expansion to support vaccination campaigns in India & worldwide amid COVID-19 surge Coronavirus Vaccine India BharatBiotech. The first batch of COVID-19 vaccine supplied by Chinas Sinopharm to COVAX, the international vaccine campaign co-led by the World Health Organization, was officially rolled off the production line on Tuesday. China has approved emergency use of Sinovac Biotechs COVID-19 vaccine in people aged between 3 and 17. From June 21 Covaxin will be the most expensive of the three vaccines which will be available in private hospitals. The vaccine is yet to be listed on the WHOs list of approved vaccines. FDA adds warning to Pfizer, Moderna COVID-19 vaccine shots to indicate the rare risk of heart inflammation after its use Pfizer Moderna COVID-19 Vaccine US FDA. Moderna seeks regulatory approval for its COVID vaccine in India. Moderna will be the fourth vaccine to be used for the vaccination drive in India. Bharat Biotech had submitted EOI (Expression of Interest) on April 19 for its vaccine and WHO informed in a document that the assessment status for Covaxin is ongoing

in temporal order of the original tweets.

Similarly, for methods (TbS) and (SbS), the goal was again to retain the m_t and m_S top-ranking triples that could be used to generate readable summaries of a length of approximately 300 words, respectively. Visualizations of method (TbS), method (SbS), and the selected triples are omitted, but we show the summary results in

Table 7. Triple-based summarization (in temporal order).

The Food and Drug Administration on Wednesday gave its approval for Sinovac use on the elderly after considering the recommendation of the experts and the current situation of high COVID-19 transmission and limited available vaccines. The Moderna COVID-19 jab is now available at 11 of 38 vaccination centres in Singapore, while the rest are offering the PfizerBioNTech product. The CoronaVac vaccine developed by the Chinese biopharmaceutical company Sinovac Biotech has effectively reduced the risk of COVID-19 symptoms in medical workers by 94%, showed a study by the Indonesian Health Ministry. On the picture Deepak Sapra, Global Head of Custom Pharma Services at drreddys Laboratories is getting a shot of Sputnik V in Hyderabad. unless you had J&J) Moderna Pfizer As of today154,199,664 Americans are fully vaccinated, according to the CDC! The first validation samples taken from the produced batch will be shipped to the Gamaleya Center for quality control. The WorldHealthOrganizations pandemic programme plans to ship 100 million doses of the Sinovac and Sinopharm COVID-19 shots by the end of next month, mostly to Africa and Asia, in its first delivery of Chinese vaccines, a WHO document shows. Your queries answered Deadline for booster dose for Sinopharm announced, if you received vaccine over six months PfizerBiontech COVID-19. USA doesnt recognise Indian vaccine COVAXIN PM Modi has gone to US and possibly he took doses of COVAXIN Whether all Indians are allowed to visit US with COVAXIN doses? days after PM Modis vaccine diplomatic push, Covaxin gets WHO nod; propaganda by anti-govt voices falls flat. Sinopharm approved for travel to UK From Nov. 22, the Sinopharm vaccine will be added to the UKs list of approved vaccines for inbound travel, benefiting more fully vaccinated people travelling from to Sinopharm is the leading vaccine administered in SriLanka.

and

Table 8. Social signal-based summary (in temporal order).

India gets third coronavirus vaccine as Russias SputnikV is cleared for emergency use COVIDVaccine. Dr Reddys administers first dose of the SputnikV vaccine in Hyderabad. The second consignment of SputnikV arrives in Hyderabad, Telangana. There are plans to introduce single-dose vaccine soon in India-Sputnik Lite. JustIn. A PIL has been moved in Delhi High Court challenging Centres notification which has accorded permission to conduct the Phase II/III clinical trial of Covaxin in the age group 2 to 18 years to its manufacturer Bharat Biotech. Bharat Biotech Amid Travel Fears NDTVs Shonakshi Chakravarty reports Read more. Malta firm wants to supply 60 million doses of SputnikV to Haryana, state government writes to Centre. Assam Covaxin Shortage Grows Into Crisis, Some Miss Second Dose Deadline. Made in India Covaxin is the third costliest vaccine globally. The appointment slots can be booked via CoWIN portal, according to the hospital administration. Delhis Madhukar Rainbow Childrens Hospital will start administering Russias COVID-19 vaccine SputnikV, tentatively by June 20. children are undergoing trials for the vaccine across the country. After Covaxin, Zydus Cadilla is the second indigenously produced vaccine for children currently under trial in India. This will also benefit people travelling abroad for education, jobs or business. I request for your kind intervention so that an early approval is received for Covaxin from WHO. Moderna approved for emergency use, 4th vaccine okayed by India COVID-19 Vaccine. Accept Covishield, Covaxin Or Face Mandatory Quarantine, India Tells EU. COVAXIN effective against DeltaPlus variant of COVID-19, says Indian Council of Medical Research study. DCGI gives nod to study mixing of Covishield and Covaxin. Subject Expert Committee recommends Covaxin for kids aged between 2 and 18 NDTVs Meher Pandey reports. Covaxin Cleared By UK, Relief For Indian Students And Tourists.

, respectively.

As shown in the summary results, three different methods yielded completely different triple sets, and therefore produced different summaries. Each corresponding original sentence in our summaries is unique, meaning there is no common sentence among our methods. As desired, this shows that the three methods generate different results, focusing either on entities, triple statements, or social features.

8.4.2. Summarization with Sentiment Perspectives

In this subsection, we generate summaries based on negative and positive sentiments (view S) as follows. The different view compositions show the power of our framework, which provides summaries with different perspectives from a single post set. We performed sentiment analysis on each of the data items using the Stanford Sentiment Analyzer [67]. This analyzer uses a fine-grained analysis based on both words and labeled phrasal parse trees to train a Recursive Neural Tensor Network (RNTN) model. The RNTN model computed the sentiments based on (1) the sentiment values of each word, and (2) the sentiment of the parse-tree structure composed of the sentiment values of words and sub-phrases. Each post will be labeled either as negative, very negative, neutral, positive, or very positive.

We present the (1) S view + social signal-based method (S + SbS) (

Table 9. Composite view summaries of positive (left) and negative (right) views with SbS.

The 80-year-old three-time World Cup champion called it an unforgettable day and urged discipline to preserve lives. The COVID-19 vaccine developed by Chinas Sinopharm has been approved for emergency use in the Maldives, the Maldives Food and Drug Administration announced at a press conference Monday afternoon. Azerbaijan on Thursday received a batch of Sinovacs COVID-19 vaccines that it directly purchased from China. Serbian President Aleksandar Vucic avucic received a dose of Chinese Sinopharm COVID-19 vaccine on Tuesday, encouraging more people to join the immunization, according to local media. Delhi high court refuses to stay Covaxin trial among children (RichaBanka reports) I request for your kind intervention so that an early approval is received for Covaxin from WHO The U.S. drug regulator on Friday added a warning to the literature that accompanies Moderna and Pfizer-BioNTech COVID-19 vaccine shots, indicating the rare risk of heart inflammation after its use. Accept Covishield, Covaxin Or Face Mandatory Quarantine, India Tells EU The Union health ministry cited a large-scale, real-life study conducted by the ICMR and said that two doses of COVIDvaccines, irrespective of Covishield and Covaxin, were successful to extend 95% protection from death. Covaxin receives certificate of Good Manufacturing Practice from Hungarian authorities. ICMR study Watch for details We want to ensure equitable access of the vaccine to every Indian citizen, and the expansion of Covaxin production facilities by Bharat Biotech will take us closer to this goal. Dr Sumit Ray, Holy Family Hospital, on delay in Covaxin approval by the World Health Organisation. Bharat Biotechs Covaxin, the vaccine against COVID-19, has received the recommendation of a SEC for use in children between the ages 2 to 18 PM NarendraModi jis visionary decision to back our scientists & researchers is now a perfect Diwali Gift from to the World. This video fits the last almost 2 years into 2 min.

Covaxin 81% Effective, Works Against UK Variant, Claims Bharat Biotech Read more: India to get its third COVID-19 vaccine; Subject Expert Committee recommends Russias SputnikV vaccine for Emergency Use Authorisation NDTVs Sukirti Dwivedi with the latest updates India gets third coronavirus vaccine as Russias SputnikV is cleared for emergency use COVIDVaccine Dr Reddys administers first dose of the SputnikV vaccine in Hyderabad There are plans to introduce single-dose vaccine soon in India-Sputnik Lite. JustIn. A PIL has been moved in Delhi High Court challenging Centres notification which has accorded permission to conduct the Phase II/III clinical trial of Covaxin in the age group 2 to 18 years to its manufacturer Bharat Biotech Malta firm wants to supply 60 million doses of SputnikV to Haryana, state government writes to Centre. After Malta Firm Offers 6 Crore Sputnik Jabs, Haryana Seeks Centres Help NDTVs Mohammad Ghazali reports SputnikV Assam Covaxin Shortage Grows Into Crisis, Some Miss Second Dose Deadline VaccinationDrive COVIDVaccine Assam Covaxin Shortage Grows Into Crisis, Some Miss Second Dose Deadline Made in India Covaxin is the third costliest vaccine globally Delhis Madhukar Rainbow Childrens Hospital will start administering Russias COVID-19 vaccine SputnikV, tentatively by June 20. The appointment slots can be booked via CoWIN portal, according to the hospital administration. After Covaxin, Zydus Cadilla is the second indigenously produced vaccine for children currently under trial in India. Moderna approved for emergency use, 4th vaccine okayed by India COVID-19Vaccine COVAXIN effective against DeltaPlus variant of COVID-19, says Indian Council of Medical Research study Union Minister of Health and Family Welfare, Mansukh Mandaviya launches the first commercial batch of Bharat Biotechs Covaxin manufactured in Gujarats Ankleshwar. Subject Expert Committee recommends Covaxin for kids aged between 2 and 18 NDTVs Meher Pandey reports The Lancet peer-review confirms the efficacy analysis of Bharat Biotechs Covaxin. As per phase-three clinical trials data, Covaxin demonstrates 77.8% efficacy against symptomatic COVID-19.

), (2) S view + triple-based method (S + TbS) (

Table 10. Composite view summaries of positive (left) and negative (right) views with TbS.

Only the PfizerBioNtech vaccine has received EUL approval, so far. Sinovac vaccine works on UK, South African variants—Brazil institute SLnews SriLanka Sinovac US President JoeBiden announces that US will share US-authorized vaccines doses of Pfizer, Moderna and Johnson & Johnson, as they become available, with the rest of the world. India cant afford to fail because India will lead the fightback by mass production of vaccines for the developing world. Haryana government has received an expression of interest from an international pharmaceutical company headquartered in Malta to provide up to 60 million doses of SputnikV vaccine. NSTworld Israel will receive in return the doses that Pfizer is to send to the PalestinianAuthority. The state has reportedly received a fresh consignment of 693,210 doses of Covaxin and Covishield vaccines from the central govt. Top sources tell CNNnews18 Authorisation of Covaxin internationally will happen by end of Aug WHO rep Meeting Health Minister today COVID-19 Vaccination Drive to Resume In Mumbai From Tomorrow As BMC Receives Fresh Stock of Vaccines COVID-19 COVID-19 A new study by the Centers for Disease Control and Prevention revealed that Modernas COVID-19 vaccine is somewhat more effective than those offered by Pfizer and Johnson & Johnsons vaccines Following the arrival of more Sinopharm vaccines to SriLanka, doses will be released to the North to begin administering on the 20–29 age group. Subject Expert Committee has given a recommendation to Official sources to ANI NSTnation Individuals who have been vaccinated with Sinovac are allowed to perform the umrah in SaudiArabia on the condition that they get a third dose of the vaccine. The last-minute addition comes less than one week before the US launches its new travel system, granting entry to travellers who have received a vaccine that has been approved by the FDA or WHO US Covaxin Travel

But with the general pool eligible for the vaccine expanding 2.5 times to 345 million from April 1, Covaxin will need to step up to service demand MintPlainFacts rashmi kundu As you can see, the UK could conceivably restrict AstraZeneca vaccine to older people and lean on Moderna, Novavax and $JNJ jabs for the younger crowd. My COVID vaccine volunteering experience Vaccinated GetVaccinated COVID Moderna Pfizer Maryland Central Govt recently gave BharatBiotech permission to test Covaxin on 2–18 yrs age group, marking it an important milestone in vaccine development. Indonesia aims to be regional vaccine making hub. According to researchers data, both Pfizer and Moderna vaccines remain highly effective at preventing severe illness and death, even amid surging DeltaVariant cases and a booster is not required. The U.S. Food and Drug Administration on Thursday authorized a booster dose of COVID-19 vaccines from Pfizer Inc and Moderna Inc for people with compromised immune systems US FDA COVID-19 COVID-19Vaccine CDCgov now recommends COVID-19 BoosterShots for all eligible Americans. Egypt to roll out a vaccination campaign with local produced additional 5 million doses Chinese Sinovac vaccines in Sept, Consultant to Minister of Health and Population for Research Noha Assem stated in a TV interview. Your queries answered Deadline for booster dose for Sinopharm announced, if you received vaccine over six months PfizerBiontech COVID-19 BharatBiotech releases a statement after the Subject Expert Committee on COVID-19 recommends the emergency use of Covaxin for children 2–18 years of age If the US FDA signs off on Modernas booster, the U.S. Centers for Disease Control and Prevention will make specific recommendations on who should get the shots. Indias COVID-19 vaccine Covaxin will be added to the UK governments approved list of vaccines for international travellers from 22 November. BharatBiotechs Covaxin now recognized by HongKong COVID-19

), and (3) S view + entity-based method (S + EbS) (

Table 11. Composite view summaries of positive (left) and negative (right) views with EbS.

Peru launched the first stage of its national vaccination campaign against the novel coronavirus disease on Feb. 9, using vaccines developed by Chinese company Sinopharm to immunize healthcare workers. Argentina on Sunday approved the COVID-19 vaccine developed by Chinese company Sinopharm for emergency use. PM narendramodi was administered a shot of Covaxin, the vaccine fully Researched & Developed as well as Made in India by BharatBiotech to mark phase 2 of COVID-19 inoculation campaign. CDSCO expert panel recommends moving Covaxin out of clinical trial mode Covaxin The COVID-19 vaccine developed by Chinas Sinopharm has been approved for emergency use in the Maldives, the Maldives Food and Drug Administration announced at a press conference Monday afternoon. Mauritius arms itself with COVAXIN, Indias indigenously developed vaccine, in its fight against COVID-19 India stands strong with Mauritius in these tough times Consignment to arrive tomorrow A time-tested and enduring partnership IndiaMauritius The new recombinant COVID-19 vaccine, developed by the National Vaccine &Serum Institute, a R&D center of Sinopharms bioscience subsidiary the China National Biotec Group, got approval from the National Medical Products Administration on Apr. 9. Covaxin, developed completely in India, can effectively neutralise multiple variants of SARS-CoV-2. Covaxin BharatBiotech developed Covaxin effective on B.1.617 and B.1.1.7, emerging variants first identified in India and UK respectively—Study by Journal Clinical Infectious Diseases PIBKochi COVIDNewsByMIB PIB India KirenRijiju BSF India CISFHQrs CRPF sector GMSRailway A study published in NEJM found that a 3rd dose of the Moderna or Pfizer vaccine significantly improved its effectiveness in organ transplant recipients who take immunosuppressant drugs. Moderna created the COVID-19 vaccine using the sequence data released on the Internet. New study shows which vaccine offers the best protection against COVID-19. As of now, Covaxin seems to be the most effective & long lasting vaccine against the Delta variant!

Cambodian Prime Minister Hun Sen emphasized that Chinas vaccines are very safe and effective, and China will become the safest and most stable supplier of COVID-19 vaccines. Sinopharms COVID-19 vaccines have just arrived in this first EU country approving the Chinese vaccine. Bharat Biotech confirms deal with Brazil to supply 20 million doses of COVAXIN vaccine CoronavirusVaccine BharatBiotech Covaxin Brazil India on Friday began using SputnikV in its battle against COVID-19 with the first dose of the vaccine from Russia administered in Hyderabad. BharatBiotech announces the quick ramp-up of additional manufacturing capacities for COVAXIN at Chiron Behring Vaccines, Ankleshwar, Gujarat, a wholly-owned subsidiary of Bharat Biotech. The Kings sister had just approved on Thursday the Sinopharm COVID-19 vaccine be imported into Thailand as alternative vaccines to help the nation cope with the pandemic. Centre recently gave BharatBiotech permission to test Covaxin on 2–18 yrs age group, marking it an important milestone in vaccine development. The first batch of COVID-19 vaccine supplied by Chinas Sinopharm to COVAX, the international vaccine campaign co-led by the World Health Organization, was officially rolled off the production line on Tuesday. From June 21 Covaxin will be the most expensive of the three vaccines which will be available in private hospitals. The Drugs Controller General of India allowed Indian pharmaceutical Cipla to import the Moderna mRNA COVID-19 vaccine on Tuesday making it the fourth vaccine that will be available to Indians. Moderna approved for emergency use, 4th vaccine okayed by India COVID-19Vaccine Global coronavirus death toll surpasses 468 million COVID-19 vaccines Moderna COVIDVictoria An expert committee recommended a booster dose of Modernas anti-COVID vaccine in the United States for certain at-risk groups, a month after making a similar decision for the Pfizer shot. He further added that 96 countries have recognized both Covishield and Covaxin vaccines.

). We compare summaries with negative sentiments and positive sentiments. Among the 16,270 triples from Section 8.3, there are 926 positive or very positive triples and 5423 negative or very negative triples, which we combined into just two classes, positive and negative.

In the (S + SbS) summary (Table 9), the positive-sentiment summary focuses on encouraging vaccine developments, such as approval and distribution progress, and highlights Sinovac and Sinopharm in countries like China and regions supportive of these vaccines. The negative-sentiment summary addresses topics about vaccine discussions, primarily focusing on Sputnik V and Covaxin. The negative-sentiment summary expresses concern about vaccine supply issues, regulatory actions, and logistical challenges. For example, the summary discusses high-profile reports and decisions around emergency authorizations, delays, and shortages, reflecting a current public concern. This difference between sentiment-based summaries reveals that negative views emphasize obstacles and critical opinions, while positive views celebrate advancements and achievements in vaccine deployment.

For S + TbS (Table 10), the summaries illustrate how different sentiments shape the narrative around vaccine distribution and public policy. The negative-sentiment summary focuses on policies and actions by governments and institutions about vaccine regulations, meanwhile highlighting delays, supply chain issues, and public dissatisfaction.

On the other hand, the positive-sentiment summary emphasizes successful implementation and the progress of vaccine campaigns. This difference captures the essence of sentiment-specific discussions by focusing on the relational dynamics between policies and outcomes.

In the (S + EbS) summaries (Table 11), the negative-sentiment summary focuses on discussions about the supply, import, and regulatory approval of vaccines, particularly during the early stages of vaccine availability in Asian countries like China, India, and Thailand. The summary reflects a sensitivity to logistical and regulatory challenges, which aligns with prevalent concerns at that time.

In contrast, the positive-sentiment summary brings attention to the successful development and effectiveness of vaccines, highlighting endorsements by various entities and governments. The positive perspective often includes mentions of vaccine approvals and endorsements, which underscore trust in vaccine efficiency and safety.

The divergent sentiments in the summaries show how different viewpoints can provide diverse insights into public opinions on vaccines. In a polarized environment, some users might prefer summaries that reflect positive perspectives, like those from the governing party, while others may focus on the critical opinions that are prevalent in opposition narratives.

The generated summaries provide insights into public sentiments surrounding COVID-19 vaccines, as well as illustrate both the challenges and achievements in vaccine deployment. They underscore the utility of MVSF in public health communication by offering a quick, precise, and concise synthesis of diverse public opinions.

This case study serves as a practical demonstration of the MVSF’s applicability, and it is illustrative rather than evaluative. It highlights the framework’s utility in real-world events, offering a quick, perspective-driven summarization. This allows for informed decision-making in public health.

9. Summary with Perspective of Fake and Real News

In this section, the perspective analysis utilizes fake news detection models to distinguish between fake and real news. This section presents a comparative summary derived from a dataset consisting of 4080 fake news items and 4480 real news items, collected by Patwa et al. [85].

This dataset, which discerns between fake and real news related to COVID-19, was compiled through automated and manual verification methods. For real news, the collection process involved using the Twitter API to gather tweets from verified accounts, including those of the WHO, CDC, and other governmental and medical organizations. Each item underwent a review process by human annotators to ensure its factual accuracy and relevance to COVID-19. The data items focus on updates about vaccine progress, governmental responses, and pandemic statistics. For fake news, verification involved scrutinizing social media posts, news articles, and public declarations previously identified as misleading. Fact-checking platforms such as Politifact, Snopes, and Boomlive were utilized to validate the falsehoods. This involved comparing content from various platforms, including Facebook and Twitter, against original documents or reliable sources to ascertain their falsity.

Additionally, researchers evaluated the dataset using several machine learning models, including Decision Trees, Logistic Regression, Support Vector Machines (SVMs), and Gradient Boosting Trees. The SVM model exhibited a superior performance, achieving an accuracy of 93.46%, precision of 93.48%, recall of 93.46%, and F1-score of 93.46% on the validation data. The dataset comprises two columns, text and label, illustrated in

Table 12. Examples of fake news and real news.

Text	Label
Politically Correct Woman (Almost) Uses Pandemic as Excuse Not to Reuse Plastic Bag #coronavirus #nashville	Fake
COVID Act Now found “on average each person in Illinois with COVID-19 is infecting 1.11 other people. Data shows that the infection growth rate has declined over time this factors in the stay-at-home order and other restrictions put in place”.	Real

.

We leverage our MVSF to explore how different summarization perspectives capture the nuances to differentiate deceptive content from factual reporting. This section aims to enhance the MVSF’s support for informed decision-making. We present the experimental result of summaries generated by method (EbS) and method (TbS), and we perform a comparative analysis of fake news and real news.

9.1. Entity-Based (EbS) Summarization of Fake News and Real News

In this subsection, we generate entity-based summaries of fake news and real news (

Table 13. Summary sentences of real news (left) and fake news (right) using method (EbS).

Indias Total Recoveries continue to rise cross 32.5 lakh today 5 States contribute 60% of total cases 62% of active cases and 70% of total fatality reported in India. Sometimes when a state reports a large number of deaths it is because they caught up on a reporting backlog of deaths that occurred long in the past. Few states reported race and ethnicity data at the beginning of April. State decides testing rates of COVID-19 for pvt labs. Several states are seeing outbreaks of in meat and poultry processing facilities. To protect the lives of healthcare workers every state needs stay-at-home orders NOW. Together with the States of Uttar Pradesh and Tamil Nadu these 5 states contribute nearly 60% of the total active cases. Some states also provide data about the date of death. These States are seeing a sudden surge in the number of cases and some of them are also reporting high mortality. States contribute 60% of total cases 62% of cases and 70% of total repo. Seven states saw the number of people hospitalized rise by 100 or more today. States reported relatively low numbers of tests (713 k) and cases (60 k). Six states saw a rise of over 100 (FL CA TX AZ GA TN) in their number of currently hospitalized COVID-19 patients. But looking at other metrics today the state reported record hospitalizations and its 2nd-highest number of deaths. Reporting gaps in 20+ states leave the public in dark about the true scope of the pandemic.

A photo shows people infected with coronavirus lying on the sidewalk in China. Video shows coronavirus patients and doctors. Video of a doctor fainted on the floor after getting infected with coronavirus. A long message attributed to Bill Gates, the Microsoft billionaire, encouraging people to reflect positively on their lives during the coronavirus outbreak has been shared in multiple countries. Chinese government is burning down people infected with Coronavirus. People shouting Allahu Akbar in Europe after the coronavirus outbreak. A video shows a man spitting inside food packets during the coronavirus crisis. Video shows Canadian PMs wife talking about the effects of Coronavirus. People in Ukraine will be forcibly vaccinated against the new coronavirus. Video shows people behaving abnormally in China due to coronavirus. A video shows a new hospital for coronavirus patients in China. People in Ahmedabad tested positive, 11 in Kanpur and 8 in Lucknow after being exposed to vegetable infected with coronavirus. Video shows Bodies of dead novel coronavirus patients in Russia. Video showing dead bodies of coronavirus patients in Osmania Hospitals mortuary. Video shows coronavirus infected notes scattered on Indore streets being sanitized. People who have never died before are now dying from coronavirus. A video showing a police officer briefing about cases being registered against the WhatsApp group admins is shared in the context of coronavirus lockdown. People died in Hyderabad due to the 2019 coronavirus. People infected with coronavirus die in the street while doctors travel through the infection zone.

). The real news summaries focus on systemic and governmental responses to public health situations. The prominent entities (state, case, death), supported by the consistent triple score of 311.5 (

Table 14. Summary triples of real news using method (EbS).

Triple	Sentiment	Triple Score
5 states contribute 60% of total cases	2	311.5
state reports large number of deaths	2	311.5
few states reported race data at beginning of april	2	311.5
state decides testing rates of COVID-19 for pvt labs	2	311.5
several states are seeing outbreaks of in meat	2	311.5
state protect lives of healthcare workers	2	311.5
5 states together contribute nearly 60% of total active cases	1	311.5
states also provide data about date of death	2	311.5
states are seeing sudden surge in number of cases	2	311.5
states contribute 60% of total cases	2	311.5
seven states saw number of people	2	311.5
states reported relatively low numbers of tests	2	311.5
six states saw rise of over 100 in their number of currently hospitalized COVID-19 patients	2	311.5
state reported its 2nd-highest number of deaths	2	311.5
states leave public in dark about true scope of pandemic	2	311.5

), show a high level of organization and relevance to public policy and safety. All except for one sentiment are neutral, which reflects the factual and informative nature of the content.

On the other hand, fake news summaries are characterized based on the depiction of vivid scenarios such as people infected, or dramatic events related to health crises. The prominent entities (people, video, coronavirus) include individual actions and localized events, with triple scores ranging slightly and generally lower than real news. The sentiments expressed are slightly more negative than for real news, which suggests a focus on shocking the audience and emotional content (

Table 15. Summary triples of fake news using method (EbS).

Triple	Sentiment	Triple Score
people infected with coronavirus	2	175.3
video shows coronavirus patients	2	177.4
video fainted getting with coronavirus	2	177.4
people reflect positively on their lives during coronavirus outbreak	2	175.3
people infected with coronavirus	2	175.3
people shouting allahu akbar after coronavirus outbreak	1	175.3
video shows man spitting inside food packets during coronavirus crisis	1	177.4
video shows canadian pms wife talking about effects of coronavirus	2	177.4
people will forcibly vaccinated against new coronavirus	2	175.3
people behaving abnormally due to coronavirus	2	175.3
video shows new hospital for coronavirus patients in china	2	177.4
people tested exposed to vegetable infected with coronavirus	2	175.3
video shows bodies of dead novel coronavirus patients in russia	1	177.4
video showing dead bodies of coronavirus patients in osmania hospitals mortuary	2	177.4
video shows coronavirus infected notes scattered on indore streets	1	177.4
people are now dying from coronavirus	2	175.3
video is shared in context of coronavirus lockdown	2	177.4
people died due to 2019 coronavirus	2	175.3
people infected with coronavirus	2	175.3

). These summaries appear to capture attention through specific and impactful imagery and scenarios.

The emphasized entities are apparently different between real news and fake news. Real news summaries focus on systemic and governmental responses, such as state actions, case updates, and death tolls. This shows a consistent approach in how these entities are scored, which is supported by the consistent triple score, 311.5. This consistency indicates a standardized method of evaluating the importance of each entity, reinforcing the objective and factual nature of real news reporting.

On the other hand, fake news summaries frequently spotlight individual, localized, or even imaginary events, including specific people, dramatic videos, or aspects related to the coronavirus. The triple scores in fake news vary between 175.3 and 177.4, reflecting a broader and less consistent approach to how information is valued. This variability suggests that the evaluation of entities in fake news does not adhere to a consistent standard. This inconsistency is likely due to the more sensational and diverse topics covered, aiming to engage the audience on an emotional level.

Understanding these differences in entity focus and scoring consistency is important. It helps to discern the reliability and intent of news content. This also helps emphasize the need for critical evaluation when consuming news, especially in distinguishing between credible and misleading information. Our analytical results underscore the importance of recognizing the veracity in news summaries.

9.2. Triple-Based (TbS) Summarization of Fake News and Real News

Table 16. Summary sentences of real news (left) and fake news (right) using method (TbS).

Principal Secretary to Prime Minister directed all concerned for an evidence based preparedness of all aspects of with active participation of Districts and States for effectiveness. Testing a drug will determine if an emergency use authorization comes by late fall. Our estimates suggest that once an effective vaccine has been distributed and international travel and trade is fully restored the economic gains will far outweigh the $38 billion investment required for the ACT Accelerator- Dr Tedros. India records more than 82,000 Recoveries for two days in a row Total Recoveries cross 40 lakhs Recovered Cases exceed Active Cases by more than 30 lakhs. CDCMMWR finds steps that help slow the spread of may also reduce if widely practiced. We continue to call on all countries to use every tool at their disposal to suppress transmission & save lives until & after we have a vaccine- Dr Tedros. The recovered cases (4,674,987) exceed active cases (966,382) by more than 37 lakh. In Cross River our Rapid Response Team (R) is supporting the state to enhance sample collection and ramp-up testing for The also worked with the Cross River State Response Team to assess a 100-bed isolation facility in Adiabo Tinapa. County health depts (Maricopa AZ and St. Louis, MO) provide both states most reliable source of LTC data. High level teams will assist State/UT in strengthening public health measures for surveillance drharshvardhan Prakash Javdekar PIB India. Across 50 states and DC we’ve tracked 16,502 total 1953 positive 13,419 negative and 1130 pending.

During Lock down period such a fantastic natural scenery on sea beach near Chandrabhaga, Puri to Konark marine drive road. BBC replaces Nichola Sturgeons pandemic briefings with Jamie Oliver making curried haggis. Spain corrected their number of deaths by COVID-19 from more than 26,000 to 2000. Nigel Farage to teach kids climate change denial for balance in BBC Lockdown Learning Scheme. Canadas COVID alert app warns that the virus is calling from inside the house. Well, I think climate change still counts as the worst federal response to a national emergency in our nations history. Are you struggling to work out the difference between real and fake news during the crisis? We fact checked Night 3 of Trumps actions to prevent COVID-19 Biden on school choice and defunding the police. Military COVID infected 118,984. A majority of COVID-19 deaths in the United States happen in a medical facility but people die at home too. Mr. Mandetta singles out the presidents COVID-19 denialism as the biggest problem in tackling the pandemic. Takes us back to our childhood when the first line of defence against a common cold sore throat was an iodine tincture, or Betadine gargle. Trump ensures Americas stimulus checks will bounce by writing his name on them. There are sterilization agents in the COVID vax which can cause sterility not only in the patient but also in the sexual partners of people who have taken the shot. Rock Legend Beats Bug. Hindu gods quarantined due to COVID-19.

shows triple-based summaries of fake news and real news. The real news summary illustrates a focus on responses to current public health challenges. The summary discusses strategies, initiatives, and evaluations surrounding public health crises. The results reflect a systematic approach to managing and communicating about public health events. By choosing triples with shortest distances from the centroid, we ensure that the topics discussed are central to the ongoing discourse and aligned with the core issues. Therefore, the experimental result suggests a focused and relevant dissemination of information.

On the other hand, the fake news summary is characterized by the distortion of facts, including the misleading presentation and incorrect information under the guise of news. These narratives often include sensational or controversial topics to attract attention or provoke emotional responses. The content involves outrageous claims or conspiracy theories that diverge from genuine news reporting.

The difference in method (TbS) between real and fake news lies in the alignment and integrity of the presented information. Real news aligns with current public discourse, reflecting ongoing societal or governmental responses. The real news summary aims to provide data and updates to promote knowledge and safety. In contrast, fake news deviates from such discussions, instead introducing irrelevant topics to distort perceptions or manipulate readers’ emotions. Fake news lacks integrity and focuses on generating engagement through controversial or sensational content at the expense of relevance and truthfulness.

10. Prototype Application

In this section, we present our web prototype (Figure 4). This app allows users to easily summarize social media discussions and other input text. Users can input a segment of text such as a social media thread or article and customize summarization preferences based on key elements of the MVSF including the method (entity-based or triple-based), sentiment perspective (negative or positive), or political perspective (left or right).

We also offer an ongoing dataset that captures COVID-19 government health policies [63], organized on a policy and monthly basis. Additionally, the website contains a built-in sample set of 100 social media posts of fake news [71] about the COVID-19 pandemic. Users can utilize these datasets to create summaries according to their preferred summarization techniques and viewpoints. With an input token limit of 4096 for each month’s policy tweet compilation, the application employs a random selection mechanism to ensure it stays within this token boundary.

Figure 5 shows the summary of tweets about the Business Closing policy in Dec. 2021 (

Table 17. Summary of Business Closing policy tweets in 2021-12.

I managed a restaurant in BH during COVID, not only did we never close, but business was booming due to delivery services being the main option of that soft shutdown. It never felt safe the entire time Senate passes stopgap funding bill, avoiding shutdown, despite TED CRUZ VOTING TO SHUT DOWN GOVERNMENT. @SenTedCruz Its crazy how in the beginning of the pandemic we were closing down and disinfecting everything because of the chance of COVID now when everyone basically has it its business as usual. we need to shutdown the nation and pay people and business owners to stay home and let COVID-19 run it’s course. It’s absolutely clear you can’t make Americans take the jab. Capitalism only benefit the few. #COVID-19 This is pretty important information. We are closing in on looking at COVID in a very different way, and the messaging should support that. Get vaccinated. Get boostered. Time to trust science and live your lives. So we learned nothing from last year or the social justice work that was done? Business as usual @BroadwayLeague not considering industry shutdown amid COVID cancellations| The restaurant i work at just got shutdown bc we dont get enough business due to COVID and ion really know how im gonna pay for the hotel i live in this week, or even this month. really not tryna live n my car again but ill do what i got to. soon as im almost up they push me down

), using the entity-based method, combined with the perspectives of negative sentiment and left political bias, while Figure 6 shows the summary of the same tweet set using the triple-based method, combined with the perspectives of positive sentiment and right political bias.

In our experiments we were not able to coax ChatGPT into using extractive summarization. Thus, our algorithms differ from the web application built with ChatGPT.

The two summaries differ in emphasis and tone based on the requested perspectives. The first (Figure 5), leaning left, emphasizes entities and negative sentiments: it accentuates Ted Cruz’s controversial vote, underscores pandemic-induced hardships, and critiques some pandemic responses. The second summary (Figure 6), in contrast, is leaning right, and highlights positive events: it applauds the Senate’s actions and emphasizes business resilience during the pandemic. Through selective extraction and strategic phrasing, the same input is transformed to resonate with different biases and foci.

Leveraging the power of ChatGPT [28], the application can rapidly produce a customized summary highlighting the most relevant information tailored to the user’s preferences. This provides an accessible and user-friendly tool for text summarization that can supplement the computationally demanding algorithms within our framework. The integration of ChatGPT enables high-quality summarization capabilities within the overall MVSF architecture, while opening up these techniques to a broader general audience.

11. Discussion and Conclusions

We presented the Multiple-View Summarization Framework (MVSF), which enables users to generate summaries of diverse perspectives with diverse methods from the same dataset. Entity-based (EbS) summarization focuses on frequent entities, triple-based (TbS) summarization emphasizes the events by ranking prominent triples, while social signal-based (SbS) summarization captures social prominence of tweets. Additionally, combining methods and views enables users to narrow down summaries to specific topics or sentiments. For instance, a user may focus on a summary of entities that are related to negative sentiments or possibly derived with a fake news analysis [71]. A notable challenge for our evaluation was the lack of a dataset of social media posts with gold-standard summaries. Despite this, the MVSF’s competitive performance against state-of-the-art models on a news dataset showcases its efficacy.

Our MVSF offers several key theoretical implications for the field of summarization. By combining multiple extractive summarization perspectives within a unified framework, the MVSF enables targeted information extraction from social media content. This provides a more nuanced understanding of the underlying data. Additionally, the MVSF’s ability to cater to specific user interests and viewpoints introduces a novel approach to personalized summarization. It contributes to the advancement of research in extractive summarization by demonstrating the effectiveness of a multi-perspective approach.

The practical implications of the MVSF are to offer content curation and decision-making support in the era of vast social media datasets. By offering personalized and fine-grained summaries, the MVSF empowers users to efficiently extract vital information aligned with their specific interests and requirements. This capability might be particularly valuable for professionals, journalists, politicians, and social media analysts, who deal with large volumes of daily information.

The MVSF sets itself apart from existing work in several key ways. While traditional transformer-based models [86] such as BertSum, SBert, T5, and Bart-Large-CNN are limited in handling long input sequences due to the performance of their attention mechanism [87], the MVSF is more flexible when accepting input text of greater length. The comparative evaluations of state-of-the-art models demonstrate the MVSF’s better performance, with an average gain of 14% in Rouge scores.

However, it is important to recognize the limitations of Rouge, which measures lexical overlap and may not fully capture semantic accuracy. This suggests a potential undervaluation if summaries are paraphrased or semantically similar but lexically different from source texts. To enhance evaluation robustness, integrating metrics like BERTScore [88] or MoverScore [89] could provide deeper semantic insights. We also acknowledge the limitation of not including a comparison against human-generated summaries. Human evaluation is considered a gold standard as it can capture nuances in coherence, informativeness, and readability that automated metrics might miss. Human evaluation through crowdsourcing could validate coherence and readability from a human perspective. Such approaches would help overcome the limitations of Rouge and offer a more comprehensive assessment. Future research could incorporate a human evaluation, where human evaluators rate the summaries based on various criteria such as relevance, accuracy, and overall quality. This can potentially provide a more comprehensive understanding of approach performance.

Another challenge we encountered is the ongoing instability of Twitter (now X), our primary source of social media posts. With policies changing frequently and new restrictions being imposed, such as limitations on the number of tweets users can access, Twitter’s dynamics have become unpredictable. Additionally, Twitter is now charging fees. Despite these challenges, we have tailored our methods to effectively handle Twitter’s unique characteristics, including its short, disconnected, and often ungrammatical posts. Moreover, we believe that the MVSF holds potential for applications beyond Twitter, including other social networks such as Reddit. As digital landscapes evolve, we anticipate adapting our framework to suit varying contexts. Looking ahead, the MVSF’s adaptability and user-centric approach render it a valuable tool for comprehending the diverse perspectives of online discussions. For example, our MVSF study has incorporated a comparative analysis of fake and real news. This offers a unique method for discerning and contrasting the reliability of information on social media. It facilitates the development of a deeper understanding of misleading information dynamics and enhances media literacy among users. By addressing user preferences and offering comprehensive insights into social media content, our framework paves the way for a more informed and insightful exploration of social media data by lay audiences and decision-makers alike.

In conclusion, the MVSF represents an advance in the field of extractive summarization, with its combination of multiple perspectives and extractive methods within a unified framework. The practical implications of the MVSF are that it offers insights for decision-making support and analysis, while its theoretical implications contribute to the understanding of multiple information extraction perspectives and methods within a unified framework.

We are currently integrating the disparate elements of our summarization methodology into a unified real-time summarization web application. The design of this web application will enable users who are not affiliated with this research group to choose a set of keywords, a time range, a geographical range, and a collection period for tweets from X. Common-sense limitations will be imposed, and password registration will be required. Users can then return after the end of the collection period and select the exact combinations of methods and views that can help them comprehend “what Twitter/X users say about their topic of interest”. This web application will be based on our own algorithms, as opposed to ChatGPT.

Our MVSF primarily employs extractive summarization techniques, which select significant pieces from the source text, maintaining the original phrasing and factual accuracy. However, this approach potentially misses subtler nuances that abstractive methods can capture. Abstractive summarization, by contrast, involves generating entirely new sentences, sometimes leading to a more coherent and concise representation. This method can offer fresh perspectives and synthesize broader themes that extractive techniques may overlook. Future work could explore hybrid models that integrate the reliability of extractive methods with the fluidity of abstractive techniques to improve both the accuracy and readability of generated summaries.

To address the limitations that social signals may not always correlate with the content quality, as stated in Section 6.3, future research could explore integrating additional metrics that provide a better content quality. Metrics such as stance detection [90] could offer more nuanced evaluations of tweet importance. Stance detection assesses the sentiment and position expressed in the content regarding a specific topic. These enhancements would contribute to a more comprehensive understanding of content value, reducing the reliance on simple engagement metrics.

The introduction of ChatGPT has allowed us to build a rapid web frontend within our MVSF that is not inhibited by the hardware limitations of the system on which we are running our own summarization algorithms. It remains to be determined, in a comparative study between ChatGPT and our implementation, whether our purpose-built algorithms perform better than ChatGPT. While tailored methods often have an edge over generalized implementations, the question is still open as to whether this is the case for MVSF text summarization. Furthermore, we plan to experiment with having ChatGPT generate triples from social media posts and then use our methods to summarize them. This exploration will further allow us to compare the effectiveness of our customized algorithms with ChatGPT’s handling of similar tasks.

Lastly, we are carefully monitoring the rapid advancement of ChatGPT, Google Bard, and their kin. These generative AI tools have enabled many new approaches to all AI tasks, while reducing long-established AI methods to technological anachronisms. We intend to make full use of the power of these new tools in our research.