Search Results (1,758)

Adversarial attacks in Natural Language Processing (NLP) present a critical challenge, particularly in sentiment analysis, where subtle input modifications can significantly alter model predictions. In search of more robust defenses against adversarial attacks on sentimental analysis, this research work introduces two novel defense mechanisms: the Lexicon-Based Random Substitute Model (LRSM) and the Word-Variant Voting Model (WVVM). LRSM employs randomized substitutions from a dataset-specific lexicon to generate diverse input variations, disrupting adversarial strategies by introducing unpredictability. Unlike traditional defenses requiring synonym dictionaries or precomputed semantic relationships, LRSM directly substitutes words with random lexicon alternatives, reducing overhead while maintaining robustness. Notably, LRSM not only neutralizes adversarial perturbations but occasionally surpasses the original accuracy by correcting inherent model misclassifications. Building on LRSM, WVVM integrates LRSM, Frequency-Guided Word Substitution (FGWS), and Synonym Random Substitution and Voting (RS&V) in an ensemble framework that adaptively combines their outputs. Logistic Regression (LR) emerged as the optimal ensemble configuration, leveraging its regularization parameters to balance the contributions of individual defenses. WVVM consistently outperformed standalone defenses, demonstrating superior restored accuracy and F1 scores across adversarial scenarios. The proposed defenses were evaluated on two well-known sentiment analysis benchmarks: the IMDB Sentiment Dataset and the Yelp Polarity Dataset. The IMDB dataset, comprising 50,000 labeled movie reviews, and the Yelp Polarity dataset, containing labeled business reviews, provided diverse linguistic challenges for assessing adversarial robustness. Both datasets were tested using 4000 adversarial examples generated by established attacks, including Probability Weighted Word Saliency, TextFooler, and BERT-based Adversarial Examples. WVVM and LRSM demonstrated superior performance in restoring accuracy and F1 scores across both datasets, with WVVM excelling through its ensemble learning framework. LRSM improved restored accuracy from 75.66% to 83.7% when compared to the second-best individual model, RS&V, while the Support Vector Classifier WVVM variation further improved restored accuracy to 93.17%. Logistic Regression WVVM achieved an F1 score of 86.26% compared to 76.80% for RS&V. These findings establish LRSM and WVVM as robust frameworks for defending against adversarial text attacks in sentiment analysis. Full article

This paper explores emotional analysis in Hebrew texts, focusing on improving machine learning techniques for depression detection by integrating psychological feature lexicons. Hebrew’s complex morphology makes emotional analysis challenging, and this study seeks to address that by combining traditional machine learning methods with sentiment lexicons. The dataset consists of over 350,000 posts from 25,000 users on the health-focused social network “Camoni” from 2010 to 2021. Various machine learning models—SVM, Random Forest, Logistic Regression, and Multi-Layer Perceptron—were used, alongside ensemble techniques like Bagging, Boosting, and Stacking. TF-IDF was applied for feature selection, with word and character n-grams, and pre-processing steps like punctuation removal, stop word elimination, and lemmatization were performed to handle Hebrew’s linguistic complexity. The models were enriched with sentiment lexicons curated by professional psychologists. The study demonstrates that integrating sentiment lexicons significantly improves classification accuracy. Specific lexicons—such as those for negative and positive emojis, hostile words, anxiety words, and no-trust words—were particularly effective in enhancing model performance. Our best model classified depression with an accuracy of 84.1%. These findings offer insights into depression detection, suggesting that practitioners in mental health and social work can improve their machine learning models for detecting depression in online discourse by incorporating emotion-based lexicons. The societal impact of this work lies in its potential to improve the detection of depression in online Hebrew discourse, offering more accurate and efficient methods for mental health interventions in online communities. Full article

This study presents the first systematic empirical comparison of neural architectures for Universal Dependencies (UD) parsing in Old English, thus addressing central questions in computational historical linguistics and low-resource language processing. We evaluate three approaches—a baseline spaCy pipeline, a pipeline with a pretrained tok2vec component, and a MobileBERT transformer-based model—across datasets ranging from 1000 to 20,000 words. Our results demonstrate that the pretrained tok2vec model consistently outperforms alternatives, because it achieves 83.24% UAS and 74.23% LAS with the largest dataset, whereas the transformer-based approach substantially underperforms despite higher computational costs. Performance analysis reveals that basic tagging tasks reach 85–90% accuracy, while dependency parsing achieves approximately 75% accuracy. We identify critical scaling thresholds, with substantial improvements occurring between 1000 and 5000 words and diminishing returns beyond 10,000 words, which provides insights into scaling laws for historical languages. Technical analysis reveals that the poor performance of the transformer stems from parameter-to-data ratio mismatches (1250:1) and the unique orthographic and morphological characteristics of Old English. These findings defy assumptions about transformer superiority in low-resource scenarios and establish evidence-based guidelines for researchers working with historical languages. The broader significance of this study extends to enabling an automated analysis of three million words of extant Old English texts and providing a framework for optimal architecture selection in data-constrained environments. Our results suggest that medium-complexity architectures with monolingual pretraining offer superior cost–benefit trade-offs compared to complex transformer models for historical language processing. Full article

Background/Objective: Dry eye syndrome affects 16 million Americans with USD 52 billion in annual healthcare costs. With large language models (LLMs) increasingly used for healthcare information, understanding their performance in delivering equitable dry eye guidance across diverse populations is critical. This study aims to evaluate and compare five major LLMs (Grok, ChatGPT, Gemini, Claude.ai, and Meta AI) regarding dry eye syndrome information delivery across different demographic groups. Methods: LLMs were queried using standardized prompts simulating a 62-year-old patient with dry eye symptoms across four demographic categories (White, Black, East Asian, and Hispanic males and females). Responses were analyzed for word count, readability, cultural sensitivity scores (0–3 scale), keyword coverage, and response times. Results: Significant variations existed across LLMs. Word counts ranged from 32 to 346 words, with Gemini being the most comprehensive (653.8 ± 96.2 words) and Claude.ai being the most concise (207.6 ± 10.8 words). Cultural sensitivity scores revealed Grok demonstrated highest awareness for minority populations (scoring 3 for Black and Hispanic demographics), while Meta AI showed minimal cultural tailoring (0.5 ± 0.5). All models recommended specialist consultation, but medical term coverage varied significantly. Response times ranged from 7.41 s (Meta AI) to 25.32 s (Gemini). Conclusions: While all LLMs provided appropriate referral recommendations, substantial disparities exist in cultural sensitivity, content depth, and information delivery across demographic groups. No LLM consistently addressed the full spectrum of dry eye causes across all demographics. These findings underscore the importance for physician oversight and standardization in AI-generated healthcare information to ensure equitable access and prevent care delays. Full article

We investigate how the factors influencing aircraft utilization have changed during and post-Pandemic depending on the business model before. We classify the Pandemic into three periods (pre-, during and post- Pandemic) and the business models into three types (Total, FSC and LCC). For each group, we analyze the importance of factors using the SHAP and Random Forest models. Through group-difference tests on factor importance, we examine whether there are significant differences across the three periods and business models. According to the findings of the ANOVA (Analysis of Variance) and the Kruskal–Wallis assay, the importance of factors influencing aircraft utilization has changed across all business models over the three periods. Pre-Pandemic, a coincident index and a consumer price index were the principal factors. However, the exchange rate (KRW/EUR) gained significant importance during the Pandemic. This suggests that the Pandemic’s impact on the aviation industry was not limited to reduced demand but was also associated with changes in the importance of exchange rates and key business indicators for airline operations. Pre-Pandemic, there were significant differences among the business model groups. However, no meaningful differences were observed during and post-Pandemic. In other words, it seems that the leading indexes were closely interconnected pre-Pandemic, whereas lagging indexes and exchange rate became closely interconnected afterward. A group-difference test confirmed that no differences were observed among the business models, but differences were evident when considering the groups in their entirety. We presented the implications for changes in airline decision-making to understand changes in the aviation industry caused by the Pandemic, by identifying how the factors influencing aircraft utilization were altered. Full article

Social network platforms have a big impact on the development of companies by influencing clients’ behaviors and sentiments, which directly affect corporate reputations. Analyzing this feedback has become an essential component of business intelligence, supporting the improvement of long-term marketing strategies on a larger scale. The implementation of powerful sentiment analysis models requires a comprehensive and in-depth examination of each stage of the process. In this study, we present a new comparative approach for several feature extraction techniques, including TF-IDF, Word2Vec, FastText, and BERT embeddings. These methods are applied to three multilingual datasets collected from hotel review platforms in the tourism sector in English, French, and Arabic languages. Those datasets were preprocessed through cleaning, normalization, labeling, and balancing before being trained on various machine learning and deep learning algorithms. The effectiveness of each feature extraction method was evaluated using metrics such as accuracy, F1-score, precision, recall, ROC AUC curve, and a new metric that measures the execution time for generating word representations. Our extensive experiments demonstrate significant and excellent results, achieving accuracy rates of approximately 99% for the English dataset, 94% for the Arabic dataset, and 89% for the French dataset. These findings confirm the important impact of vectorization techniques on the performance of sentiment analysis models. They also highlight the important relationship between balanced datasets, effective feature extraction methods, and the choice of classification algorithms. So, this study aims to simplify the selection of feature extraction methods and appropriate classifiers for each language, thereby contributing to advancements in sentiment analysis. Full article

Falkland Islands English (FIE) began its development in the first half of the 19th century. In part, as a consequence of its youth, FIE is an understudied variety. It shares some morphosyntactic features with other anglophone countries in the Southern Hemisphere, but it also shares lexical features with regional varieties of Spanish, including Rioplatense Spanish. Che is one of many South American words that have entered FIE through Spanish, with its spelling ranging from “chay” and “chey” to “ché”. The word has received some marginal attention in terms of its meaning. It is said to be used in a similar way to the British dear or love and the Australian mate, and it has been compared to chum or pal, and is taken as an equivalent of the River Plate, hey!, hi!, or I say!. In this work, we explore the hypothesis that che entered FIE through historical contact with Rioplatense Spanish, drawing on both linguistic and sociohistorical evidence, and presenting survey, corpus, and ethnographic data that illustrate its current vitality, usage, and social meanings among FIE speakers. In situ observations, fieldwork, and an online survey were used to look into the vitality of che. Concomitantly, by crawling social media and the local press, enough data was gathered to build a small corpus to further study its vitality. A thorough literature review was conducted to hypothesise about the borrowing process involving its entry into FIE. The findings confirm that the word is primarily a vocative, it is commonly used, and it is indicative of a sense of belonging to the Falklands community. Although there is no consensus on the origin of che in the River Plate region, it seems to be the case that it entered FIE during the intense Spanish–English contact that took place during the second half of the 19th century. Full article

In some languages, prosodic cues on word stems can be used to predict upcoming suffixes. Previous studies have shown that second language (L2) users can process such cues predictively in their L2 from approximately intermediate proficiency. This ability may depend on the mapping of the L2 prosody onto first language (L1) perceptual and functional prosodic categories. Taking as an example the Danish stød, a complex prosodic cue, we investigate an acquisition context of a predictive cue where L2 users are unfamiliar with both its perceptual correlates and its functionality. This differs from previous studies on predictive prosodic cues in Swedish and Spanish, where L2 users were only unfamiliar with either the perceptual make-up or functionality of the cue. In a speeded number judgement task, L2 users of Danish with German as their L1 (N = 39) and L1 users of Danish (N = 40) listened to noun stems with a prosodic feature (stød or non-stød) that either matched or mismatched the inflectional suffix (singular vs. plural). While L1 users efficiently utilised stød predictively for rapid and accurate grammatical processing, L2 users showed no such behaviour. These findings underscore the importance of mapping between L1 and L2 prosodic categories in second language acquisition. Full article

This study investigates how topic-specific expression by women delivery riders on digital platforms predicts their community engagement, emphasizing the mediating role of self-disclosure and the moderating influence of cognitive and emotional language features. Using unsupervised topic modeling (Top2Vec, Topical Vectors via Embeddings and Clustering) and psycholinguistic analysis (LIWC, Linguistic Inquiry and Word Count), the paper extracted eleven thematic clusters and quantified self-disclosure intensity, cognitive complexity, and emotional polarity. A moderated mediation model was constructed to estimate the indirect and conditional effects of topic probability on engagement behaviors (likes, comments, and views) via self-disclosure. The results reveal that self-disclosure significantly mediates the influence of topical content on engagement, with emotional negativity amplifying and cognitive complexity selectively enhancing this pathway. Indirect effects differ across topics, highlighting the heterogeneous behavioral salience of expressive themes. The findings support a statistically grounded, semantically interpretable framework for predicting user behavior in high-dimensional text environments. This approach offers practical implications for optimizing algorithmic content ranking and fostering equitable visibility for marginalized digital labor groups. Full article

This research study focuses on the social preference parameters and financial decisions among welfare populations receiving social benefits in Miami, Florida. Understanding the attitudes and primary motivations that shape financial decision-making is of great interest to economists, marketers, and other social scientists. The implications of developing a solid understanding of these attitudes and motivations are vast in terms of erecting tangible and sensitive workforce development policies to assist the specific population studied. This study is designed to determine whether significant differences exist in the strength of preference parameters between welfare participants and other populations. The preference parameters assessed in this paper were self-interest, altruism, trust, and reciprocity, both positive and negative. The control group in this study is college students. The results from the experiments show that welfare recipients exhibit similar behavioral patterns and make financial decisions in a manner similar to the general population. In other words, the control group and the experimental group did not differ significantly in their financial decision processes. This finding has several implications for how economists and policymakers assess and approach policymaking; nevertheless, the question remains whether or not there are other preference parameters that differ between the two groups. Full article

Background: Sexuality is a central aspect of being human, even if people experience it in different ways in various stages of life. Sexuality in older people may be expressed, as well as affection, companionship, touch, and physical contact. However, older peoples’ sexual needs are not properly considered by themselves, caregivers, or healthcare professionals. Reviews on barriers related to identification and satisfaction of sexual needs of people living in nursing home are scarce. In this scenario we intended to summarize the state of evidence regarding sexual need identification and satisfaction among older people living in nursing homes and possible barriers that could limit sexual need identification and satisfaction. Methods: We carried out a narrative review. The included studies responded to the research question, using the following key words: nursing homes, sexuality or sexual need, or sexual behavior, older people. Searched databases included PubMed, Embase, CINAHL, PsycInfo, and Scopus. Results: After searching and screening we included 22 studies, finding three main topics: 1. identification of sexual needs by residents and healthcare personnel attitude and practice; 2. barriers and reasons hindering the identification of sexual needs; and 3. manifestation and satisfaction of sexual needs. Conclusions: The findings showed that nursing homes’ residents have different sexual needs, but there are many organizational, educational, and cultural barriers and negative attitudes of healthcare personnel. Supporting nursing home residents to express their sexual needs is a challenge for the healthcare professionals and managers of nursing homes. Full article

Vocabulary acquisition is a fundamental challenge in second language (L2) learning. Recent research highlights the benefits of using gestures to enhance vocabulary retention. This comprehensive review explores the theoretical, empirical, and neuroscientific foundations of gesture-enhanced learning. Findings show that the human body, specifically sensorimotor engagement, can be harnessed as an effective cognitive tool to support long-term word learning. This paper examines the limitations of traditional vocabulary learning methods, introduces embodied cognition as a theoretical framework, presents behavioral and neuroscientific evidence supporting gesture-based learning, and offers practical applications for educational settings. This integration of multidisciplinary research provides a robust foundation for reconceptualizing the role of physical engagement in second language acquisition. Full article

Handwritten Keyword Spotting (KWS) remains a challenging task, particularly in segmentation-free scenarios where word images must be retrieved and ranked based on their similarity to a query without relying on prior page-level segmentation. Traditional KWS methods primarily focus on visual similarity, often overlooking the underlying semantic relationships between words. In this work, we propose a novel NLP-driven re-ranking approach that refines the initial ranked lists produced by state-of-the-art KWS models. By leveraging semantic embeddings from pre-trained BERT-like Large Language Models (LLMs, e.g., RoBERTa, MPNet, and MiniLM), we introduce a relevance feedback mechanism that improves both verbatim and semantic keyword spotting. Our framework operates in two stages: (1) projecting retrieved word image transcriptions into a semantic space via LLMs and (2) re-ranking the retrieval list using a weighted combination of semantic and exact relevance scores based on pairwise similarities with the query. We evaluate our approach on the widely used George Washington (GW) and IAM collections using two cutting-edge segmentation-free KWS models, which are further integrated into our proposed pipeline. Our results show consistent gains in Mean Average Precision (mAP), with improvements of up to $2.3\%$ (from $94.3\%$ to $96.6\%$) on GW and $3\%$ (from $79.15\%$ to $82.12\%$) on IAM. Even when mAP gains are smaller, qualitative improvements emerge: semantically relevant but inexact matches are retrieved more frequently without compromising exact match recall. We further examine the effect of fine-tuning transformer-based OCR (TrOCR) models on historical GW data to align textual and visual features more effectively. Overall, our findings suggest that semantic feedback can enhance retrieval effectiveness in KWS pipelines, paving the way for lightweight hybrid vision-language approaches in handwritten document analysis. Full article

This study focuses on a profound paradox in eco-hotel evaluations: why do consumer ratings for location, a static asset, exhibit dynamic fluctuations? To solve this puzzle, we construct a two-stage signal-processing theoretical framework that integrates Signaling Theory and the Elaboration Likelihood Model (ELM). This framework posits that the dynamic trajectory of a hotel’s green attributes (encompassing eco-facilities, sustainable practices, and ecological experiences) constitutes a high-fidelity market signal about its underlying quality. We utilized natural language processing techniques (Word2Vec and LSA) to conduct a longitudinal analysis of over 60,000 real consumer reviews from Booking.com between 2020 and 2023. This study confirms that continuous improvements in green attributes result in significant positive spillovers to location scores, while any degradation triggers strong negative spillovers. More critically, consumer environmental involvement (CEI) acts as an amplifier in this process, with high-involvement consumers reacting more intensely to both types of signals. The research further uncovers complex non-linear threshold characteristics in the spillover effect, subverting traditional linear management thinking. These findings not only provide a dynamic and psychologically deep theoretical explanation for sustainable consumption behavior but also offer forward-thinking practical implications for hoteliers on how to strategically manage dynamic signals to maximize brand value. Full article

The tip-of-the-tongue (TOT) phenomenon is characterized by a temporary inability to retrieve a word despite a strong sense of familiarity. While extensive research has linked phonological processing to TOT, the exact nature of this relationship remains debated. The “blocking hypothesis” suggests that the retrieval of target words is interfered with by phonological neighbors, whereas the “transmission deficit hypothesis” posits that TOT arises from insufficient phonological activation of the target words. This study revisited this issue by examining the relationship between the microstructural integrity of the phonological processing brain network and TOT, utilizing graph-theoretical analyses of neuroimaging data from the Cambridge Centre for Ageing and Neuroscience (Cam-CAN), which included diffusion tensor imaging (DTI) data from 576 participants aged 18–87. The results revealed that global efficiency and mean degree centrality of the phonological processing network positively predicted TOT rates. At the nodal level, the nodal efficiency of the bilateral posterior superior temporal gyrus and the clustering coefficient of the left premotor cortex positively predicted TOT rates, while the degree centrality of the left dorsal superior temporal gyrus (dSTG) and the clustering coefficient of the left posterior supramarginal gyrus (pSMG) negatively predicted TOT rates. Overall, these findings suggest that individuals with a more enriched network of phonological representations tend to experience more TOTs, supporting the blocking hypothesis. Additionally, this study highlights the roles of the left dSTG and pSMG in facilitating word retrieval, potentially reducing the occurrence of TOTs. Full article