Unveiling Sustainability in Ecommerce: GPT-Powered Software for Identifying Sustainable Product Features

: In recent years, the concept of sustainability has gained signiﬁcant attention across various industries. Consumers are increasingly concerned about the environmental impact of the products they purchase, leading to a growing demand for sustainable options. However, identifying sustainable product features can be a complex and time-consuming task. This paper presents a novel approach to address this challenge by utilizing GPT (Generative Pre-trained Transformer) powered software for automatically identifying sustainable product features from product descriptions, titles, and product speciﬁcations. The software leverages the power of natural language processing and machine learning to classify products into different sustainability categories. By analyzing the textual information provided, the software can extract key sustainability indicators, such as eco-friendly materials, energy efﬁciency, recyclability, and ethical sourcing. This automated process eliminates the need for manual assessment and streamlines the evaluation of product sustainability. The proposed software not only empowers consumers to make informed and sustainable purchasing decisions but also facilitates businesses in showcasing their environmentally friendly offerings. The experimental results demonstrate the effectiveness and accuracy of the software in identifying sustainable product features. The primary objective of this article is to assess the suitability of the GPT model for the domain of sustainability assessment. By collecting a real-life dataset and employing a speciﬁc methodology, four hypotheses are formulated, which will be substantiated through the experimental outcomes. This research contributes to the ﬁeld of sustainability assessment by combining advanced language models with product classiﬁcation, paving the way for a more sustainable and eco-conscious future.


Introduction
In today's global context, sustainability has emerged as a crucial concept with significant implications for our planet and future generations.Sustainability encompasses a holistic approach to addressing the pressing environmental, social, and economic challenges we face [1].It emphasizes the need to meet the needs of the present without compromising the ability of future generations to meet their own needs [2].The significance of sustainability lies in its potential to mitigate climate change, protect natural resources, promote social equity, and foster long-term economic prosperity [3].By recognizing the interconnectedness of these pillars, sustainability encourages responsible consumption and production patterns, innovative solutions, and collective action to create a more resilient and harmonious world [4].As we navigate complex issues such as climate change, biodiversity loss, and social inequality, adopting sustainable practices and policies becomes imperative for a sustainable and thriving future for humanity and the planet we call home.
Each year, an increasing number of initiatives related to sustainability, both promotional and operational, come to the forefront, with numerous applications supporting the overall effort [5].Technology, particularly AI, can aid in the promotion and development of sustainability by providing appropriate tools to businesses and individuals, enabling them to engage in and facilitate more sustainable purchases and sales [6].However, both consumers and developers of these specific AI tools must exercise great caution regarding the limitations, potential unintended consequences, and ethical implications associated with utilizing AI for sustainability purposes.
This article presents an innovative software solution aimed at addressing sustainability challenges through the utilization of a GPT (Generative Pre-trained Transformer)-powered application.The primary objective of this software is to automate the identification of sustainable product attributes from product descriptions, titles, and product specifications.By harnessing the capabilities of natural language processing and machine learning, the software effectively categorizes products based on their sustainability characteristics.Through the analysis of textual information, it extracts crucial indicators of sustainability, such as eco-friendly materials, energy efficiency, recyclability, and ethical sourcing.By automating this process, the software eliminates the need for manual assessment, streamlining the evaluation of product sustainability.This software implementation is facilitated through an easily installable Chrome app, which informs users about the sustainability features based on the product description and title, without requiring user intervention.Significantly, this software empowers consumers to make informed and sustainable purchasing decisions, while also assisting businesses in showcasing their sustainable offerings.
The present research extends beyond the mere development of software; rather, it delves into the evaluation of the suitability of the GPT model in identifying and assessing the sustainability of products using a real-world dataset.Four pivotal hypotheses are introduced, which will be substantiated through the outcomes of the experiments: Hypothesis 1. Research Objective: The study seeks to evaluate the accuracy of GPT model responses.Alternative Hypothesis: The GPT model produces accurate responses when appropriate prompts are employed.

Hypothesis 2. Research Objective:
The study aims to examine the format of responses generated by the GPT model.Alternative Hypothesis: The format of GPT model responses is accurate and aligns with the explicitly defined structure in the prompt.

Hypothesis 3. Research Objective:
The study aims to assess the effectiveness of the GPT model in identifying sustainable features based on product titles and descriptions.Alternative Hypothesis: The GPT model is capable of evaluating the sustainable features of a product when provided with its title and description.

Hypothesis 4. Research Objective:
The study aims to evaluate the effectiveness of the GPT model in identifying product sustainable features, particularly when product titles and descriptions are less descriptive.Alternative Hypothesis: The GPT model can successfully retrieve sustainable features of a product even when the provided title and description lack sufficient descriptive information.
Through these hypotheses, we intend to gain insights into the performance and potential applications of the GPT model in the context of sustainable product identification.By exploring its accuracy, format, and effectiveness under various conditions, we aim to contribute valuable findings to the field of natural language processing and its practical implications in sustainability assessment.
The paper is organized as follows: In Section 2, an in-depth analysis of the three pillars of sustainability and their interplay with products and their corresponding attributes is presented, drawing upon a comprehensive review of current academic literature.Moreover, it rigorously investigates the predominant sustainable product attributes that align with these pillars, offering valuable insights into the prevailing trends and considerations within this domain.In Section 2.2, the Generative Pre-trained Transformer (GPT) and Natural Language Processing (NLP) models are introduced, along with their capabilities.The concept of a prompt is also presented, emphasizing its significance as a stimulus for GPT models to generate appropriate responses.In Section 3, the study provides a comprehensive presentation of the research objectives, significance, and contribution to the domain of sustainability research.Additionally, it expounds upon the rigorous methodology encompassing data collection and preprocessing, as well as the process of model evaluation.Proceeding to Section 4, the research delves into diverse aspects, including the detailed depiction of the experimental setup and results, an analysis of user interaction, insights into software implementation and system design, and illustrative case studies and examples.Furthermore, the section also presents the results derived from the research, which effectively validate the proposed research hypotheses.The final section of this paper, Section 5, offers a comprehensive analysis of the authors' reflections concerning potential future enhancements of the software.Furthermore, it conducts an exploration of the ethical considerations associated with the application of the GPT model in the context of sustainability assessment tasks.

Literature Review
The realm of e-commerce has been continuously evolving, exerting a direct impact on consumer purchasing habits [7,8].While consumer choices have expanded, the corresponding legal frameworks governing consumer protection have not progressed at an equal pace [9].Aspects such as consumer protection of privacy, misinformation, product quality, environmental impact, and related concepts necessitate addressing to safeguard consumers in the realm of e-commerce.Apart from legislative measures, educating consumers about responsible e-commerce practices is of paramount importance.Artificial Intelligence (AI), in turn, holds the potential to empower consumers to make informed decisions, whether pertaining to product quality or considerations of sustainability and environmental impact.
Within this context, this literature review chapter delves into the domains of sustainability, sustainable products, and GPT models.Through an extensive review of the existing literature, this section aims to explore and critically analyze the interplay of these themes, examining their potential implications in the context of e-commerce and consumer well-being.

Sustainable Products and Attributes
Sustainability is not an individual effort; it requires collective action and collaboration across all sectors of society to transform our consumption patterns and create a sustainable future [1,3].Both production companies and retail businesses, as well as consumers, need to collectively change their mindset regarding sustainability [5].Before designing and manufacturing a product, production companies should always consider how it will bring benefits or at least have a neutral impact on the environment and the economic and social aspects.Simultaneously, retail businesses should have sustainability awareness in order to inform and promote sustainable products to their customers, educating them on how an individual sustainable purchasing decision can make a difference [9,10].Furthermore, consumers should develop a sense of sustainability and contribute to a more sustainable future by actively supporting and prioritizing sustainable products [11].Through their support of sustainable products, individuals can effectively convey a clear message to industries and policymakers that they prioritize long-term ecological balance over shortterm profits.
Sustainability can be defined as the practice of meeting the present generation's needs without compromising future generations' ability to meet their own needs [2,9].It is a comprehensive approach that recognizes the interconnectedness of environmental, social, and economic aspects in achieving long-term well-being [12].The environmental pillar of sustainability focuses on preserving and protecting natural resources, mitigating climate change, and promoting biodiversity conservation [4,13].The social pillar emphasizes the importance of social equity, human rights, and community well-being, aiming to ensure that all share the benefits of development [4,14].The economic pillar recognizes the need for sustainable economic growth that promotes prosperity, job creation, and innovation while considering the environmental and social impacts of economic activities [4,15].By addressing these three pillars in an integrated manner, sustainability seeks to create a balanced and harmonious society that respects the limits of our planet's resources and fosters a better quality of life for current and future generations.

Sustainable Products
The importance of sustainable products within the broader concept of sustainability lies in their ability to serve as catalysts for positive change [9].Sustainable products embody responsible production and consumption practices, helping to transition towards a more sustainable future [5].By adhering to eco-friendly materials, minimizing waste, and prioritizing ethical sourcing, sustainable products significantly reduce environmental impact throughout their lifecycle [15].They promote social well-being by upholding fair labor practices, supporting local communities, and ensuring the welfare of workers [12].Moreover, sustainable products contribute to the long-term viability of the economy by driving innovation, creating job opportunities in green industries, and enhancing market competitiveness [14].They demonstrate that economic growth and environmental responsibility are not mutually exclusive, but rather interconnected elements that can thrive together.
In terms of environmental conservation, sustainable products prioritize the use of renewable resources and eco-friendly materials, reducing carbon emissions, minimizing waste, and preserving natural habitats [9,15].Socially, they enhance the well-being of individuals and communities by ensuring fair wages, safe working conditions, and equitable treatment throughout the supply chain, fostering social equity and empowerment [13,14].Additionally, sustainable products often support local communities through various initiatives, such as sourcing ingredients locally or reinvesting a portion of profits into community development programs [16].From an economic perspective, they drive market innovation, attract environmentally conscious consumers, inspire competitors to adopt sustainable practices, create job opportunities in green sectors, and stimulate socially and environmentally responsible economic growth [13].Overall, sustainable products serve as powerful agents for positive change, addressing environmental, social, and economic challenges for a more sustainable and prosperous future [15].

Sustainable Products Attributes
In today's world, sustainability has become a pressing concern, and understanding the environmental attributes of products is crucial for promoting a healthier planet [1].These attributes encompass factors such as resource usage, carbon footprint, and waste management, providing valuable insights into the ecological impact of products and their lifecycle [17].Alongside environmental considerations, social attributes play a vital role in assessing the sustainability of products [18].These attributes encompass aspects such as fair labor practices, community engagement, and human rights, reflecting a product's contribution to social well-being and ethical practices throughout its production and distribution [9].Recognizing that economic factors are intricately linked with sustainability, understanding the economic attributes of products becomes imperative.These attributes encompass aspects like cost-effectiveness, value creation, and long-term economic viability, shedding light on a product's potential to support sustainable business models and contribute to a thriving economy while minimizing negative externalities [19,20].

Environmental Attributes
Environmental attributes are fundamental components of sustainability, as they focus on the ecological impact of products and practices.These attributes encompass a wide range of factors, including resource usage, energy efficiency, waste management, and emissions reduction [17].Evaluating and improving the environmental attributes of products involves adopting sustainable practices such as using renewable resources, minimizing waste generation, and implementing efficient recycling and disposal systems [21].By prioritizing environmental attributes, businesses can reduce their carbon footprint, conserve natural resources, and mitigate negative impacts on ecosystems and biodiversity [9].Additionally, consumers can make informed choices by selecting products that have favorable environmental attributes, thereby encouraging companies to adopt more sustainable production and distribution methods [3].
Furthermore, environmental attributes in sustainability extend beyond individual products to consider broader environmental impacts.This includes factors like land and water conservation, climate change mitigation, and the protection of natural habitats [21,22].Sustainable development recognizes the interdependence between human activities and the environment, aiming to minimize harm and maximize the regeneration of ecosystems [9].By prioritizing environmental attributes, we can work towards creating a future where ecological balance is restored, and the well-being of both present and future generations is safeguarded.Through sustainable practices and awareness, we can build a greener, healthier planet.

Social Attributes
Social attributes play a critical role in achieving sustainability by addressing the social dimensions and impacts of products and practices.These attributes encompass a wide range of factors, including fair labor practices, community engagement, human rights, and social justice considerations [14].Sustainable initiatives prioritize social attributes by ensuring safe and fair working conditions, promoting diversity and inclusion, and respecting the rights and well-being of workers and communities affected by production processes [23].By embracing social attributes, businesses can contribute to the creation of thriving and resilient communities, foster positive social impacts, and enhance the overall quality of life for individuals and societies.
Moreover, social attributes in sustainability extend beyond the immediate impact on workers and communities to encompass broader societal concerns.This includes issues such as poverty alleviation [24], education, healthcare access, and gender equality.Sustainable development recognizes that social progress and well-being are inseparable from environmental protection and economic prosperity [25].By addressing social attributes, businesses, and organizations can proactively contribute to the achievement of the United Nations Sustainable Development Goals (SDGs) and work towards a more inclusive, just, and equitable future for all [26].Through conscious choices and responsible practices, we can foster a sustainable society where social attributes are integrated into the fabric of everyday life and decision-making.

Economic Attributes
Economic attributes are integral to the concept of sustainability, as they emphasize the financial viability and long-term economic impact of products and practices [16].These attributes encompass a range of factors, including cost-effectiveness, resource efficiency, and the potential for creating value [13].Sustainable initiatives often aim to optimize resource usage, minimize waste, and adopt innovative technologies that reduce operating costs [2].By prioritizing economic attributes, businesses can not only enhance their profitability but also contribute to a more sustainable economy by reducing resource consumption, minimizing environmental impacts, and creating new market opportunities.
Moreover, economic attributes in sustainability extend beyond individual products and encompass broader economic systems and models.Sustainable development strives to promote economic attributes that foster the equitable distribution of resources, support local communities, and contribute to social well-being [25].This includes considerations such as fair trade, responsible investment practices, and the integration of social and environmental costs into pricing mechanisms.By integrating economic factors into sustainability, we can strive for a future where economic growth is aligned with environmental protection, social equity, and the long-term prosperity of both current and future generations.

Main Sustainable Product Features
Having gained an understanding of the classification of sustainable attributes into distinct categories, we proceed to outline a selection of prevalent and pervasive sustainable product features.This compilation serves as a valuable resource for our ongoing research, as it equips us with the essential knowledge to delve deeper into the subject matter.By familiarizing ourselves with these widely recognized sustainable product features, we aim to enhance our understanding of their significance and implications within the broader realm of sustainability.

•
Organic: The term "organic" refers to a specific attribute of a product, particularly in the context of food and agriculture.Organic products are grown, produced, and processed using methods that prioritize environmental sustainability and the avoidance of synthetic chemicals [27].Organic farming practices focus on enhancing soil health, biodiversity, and ecological balance, while minimizing the use of synthetic fertilizers, pesticides, antibiotics, and genetically modified organisms (GMOs) [28].Organic products are cultivated using natural fertilizers like compost or manure, and their pest and disease control measures rely on natural methods such as crop rotation, biological pest control, and the use of beneficial insects.By adhering to these principles, organic products aim to reduce environmental impacts, promote soil fertility, protect natural ecosystems, and offer consumers a healthier and more sustainable choice [29].Organic certification standards, such as those set by regulatory bodies or third-party organizations, ensure that products labeled as "organic" meet specific criteria and undergo regular inspections to maintain their organic integrity [30,31].

•
Fair trade: Fair trade is a product feature or certification emphasizing ethical and equitable trading practices, particularly in international trade with producers in developing countries [32].It ensures that producers receive fair compensation, promotes sustainable livelihoods, and supports community development [32].Fair trade certification guarantees that the product has been sourced from producers who meet specific social, economic, and environmental standards [33].By supporting fair trade products, consumers contribute to the empowerment of small-scale farmers, artisans, and workers, addressing social and economic inequalities while promoting sustainable development [25].Fair trade serves as a significant feature in the realm of sustainable products by fostering fairer trade relationships and promoting the well-being of marginalized producers [34].

•
Recyclable: The "recyclable" feature is a crucial attribute of sustainable products, as it promotes the efficient use of resources and minimizes waste [35].Recyclable products can be collected, processed, and transformed into new materials or products, reducing the demand for virgin resources and the environmental impact of extraction [36].By designing products with recyclability in mind, manufacturers contribute to the circular economy, where materials are kept in use for as long as possible [37].Recycling also helps reduce landfill waste and the associated environmental and health hazards, making it essential to achieving a more sustainable and resource-efficient society [35].

•
Reusable: The "reusable" feature offers an alternative to single-use items that contribute to waste and resource depletion.Reusable products are designed to be used multiple times, reducing the need for constant replacements [38].By encouraging reuse, these products minimize the environmental impact associated with manufacturing, transportation, and disposal [39].Reusable products can include items such as shopping bags, water bottles, coffee cups, and food containers, providing consumers with sustainable choices that reduce their reliance on disposable options [38].By embracing reusable products, individuals can significantly contribute to waste reduction and the conservation of resources, ultimately promoting a more sustainable and environmentally friendly lifestyle [39].

•
Renewable energy: The "renewable energy" feature involves harnessing energy from renewable sources that are naturally replenished.By utilizing renewable energy sources like solar, wind, hydro, or geothermal power, sustainable products reduce reliance on fossil fuels and contribute to mitigating climate change [40].Incorporating renewable energy into product design and production processes helps minimize greenhouse gas emissions and air pollution, promoting cleaner and more sustainable energy systems [41].Furthermore, the adoption of renewable energy in manufacturing can lead to energy cost savings, increased energy independence, and a more resilient and sustainable economy [40].By embracing renewable energy, sustainable products contribute to a greener future and support the transition to a low-carbon and sustainable energy system.

•
Energy efficient: The "energy efficient" feature is a crucial characteristic of sustainable products as it helps minimize energy consumption and reduce environmental impact [42].Energy-efficient products are designed to use less energy while maintaining their functionality and performance.By incorporating energy-efficient technologies and design principles, these products contribute to lower energy bills, reduced greenhouse gas emissions, and more sustainable use of resources [43].Additionally, energy-efficient products often meet or exceed strict energy efficiency standards set by regulatory bodies, ensuring their reliability and effectiveness.By promoting energy efficiency, sustainable products help combat climate change, conserve natural resources, and support a more sustainable and energy-conscious lifestyle [42].

•
Sustainable packaging: "Sustainable packaging" focuses on minimizing the environmental impact of product packaging.It involves using materials and design strategies that prioritize resource efficiency, waste reduction, and recyclability [44].Sustainable packaging aims to minimize the use of non-renewable resources, such as petroleum-based plastics and encourages the use of renewable and biodegradable materials.Additionally, sustainable packaging considers the entire lifecycle of the product, including its production, transportation, use, and end-of-life disposal, striving for minimal environmental impact at each stage [45].By adopting sustainable packaging practices, products can contribute to reducing plastic pollution, conserving resources, and promoting a more circular and environmentally friendly economy [44].

•
Biodegradable/compostable: The "biodegradable/compostable" feature is a significant attribute of sustainable products as it relates to their ability to break down naturally and return to the environment without leaving harmful residues [46].Biodegradable products can be broken down by microorganisms into natural elements over time, minimizing their impact on ecosystems [47].Compostable products, on the other hand, are specifically designed to break down in composting systems, resulting in nutrientrich soil that can be used for agricultural purposes [48].The use of biodegradable or compostable materials reduces waste and landfill burden, promotes sustainable waste management practices, and supports the transition towards a more circular economy [46].By choosing biodegradable or compostable products, consumers can make a positive impact on environmental conservation and contribute to a more sustainable future.

•
Low carbon footprint: The "low carbon footprint" feature is a sustainable product attribute signifying products' minimal contribution to greenhouse gas emissions and climate change [49].Products with a low carbon footprint are designed, manufactured, and distributed in a manner that minimizes energy consumption and the release of carbon dioxide and other greenhouse gases.This is achieved through various strategies such as utilizing energy-efficient production processes, sourcing materials locally to reduce transportation emissions, and using renewable energy sources [50].By reducing their carbon footprint, these products help mitigate climate change, preserve natural resources, and promote a more sustainable and environmentally responsible lifestyle [49].Choosing products with a low carbon footprint allows consumers to make environmentally conscious choices and contribute to a greener and more sustainable future.

•
Carbon neutral: The "carbon neutral" feature indicates that the emissions generated throughout products' lifecycles have been effectively offset or balanced by supporting projects that reduce or capture an equivalent amount of greenhouse gases [51].Carbon neutrality is achieved by implementing measures to minimize emissions during production, transportation, and disposal, and then offsetting the remaining emissions through activities like reforestation, investing in renewable energy projects, or supporting carbon capture and storage initiatives [52].By being carbon neutral, these products have a net-zero impact on greenhouse gas emissions, making them environmentally responsible choices [51].Supporting carbon neutrality in product development helps combat climate change, fosters a more sustainable economy, and encourages a shift towards a low-carbon future [52].Choosing carbon-neutral products allows consumers to make a positive environmental impact and contribute to global climate action [52].

•
Ethically sourced: The "ethically sourced" feature focuses on ensuring that the raw materials used in products' production are obtained in a socially and environmentally responsible manner [53].Ethically sourced products prioritize fair trade, respect for human rights, and adherence to labor standards throughout the supply chain [54].This includes fair wages, safe working conditions, and the prohibition of child labor or exploitation [54].By choosing ethically sourced products, consumers can support companies that prioritize social equity, worker well-being, and environmental stewardship [53].Ethical sourcing promotes transparency, accountability, and sustainable practices, making it an essential aspect of sustainable product selection [21].

•
Water conservation: The "water conservation" feature aims to minimize water usage and promote responsible water management throughout the product lifecycle [55].
Products designed with water conservation in mind employ strategies such as efficient water use, water-saving technologies, and responsible manufacturing processes that reduce water waste and pollution [56].By minimizing water consumption, these products help alleviate water scarcity, preserve freshwater ecosystems, and protect this vital resource for future generations.Additionally, water conservation in product design encourages consumer awareness and responsible water usage, fostering a more sustainable and environmentally conscious mindset [56].Choosing products with a water conservation feature allows individuals to contribute to water sustainability and promote a more water-efficient society.

•
Non-toxic: The "non-toxic" feature ensures that products are free from harmful substances and pose minimal risks to human health and the environment [57].Non-toxic products are designed and manufactured using materials that do not contain toxic chemicals or pollutants, including heavy metals, harmful solvents, or volatile organic compounds (VOCs) [58].By prioritizing non-toxic ingredients and materials, these products minimize the potential for adverse health effects, allergies, or pollution during production, use, and disposal [57].Choosing non-toxic products promotes a safer living environment, reduces the burden on ecosystems, and supports sustainable consumption practices [57].Furthermore, non-toxic products often undergo rigorous testing and certification processes to provide consumers with assurance regarding their safety and environmental impact [58].

•
Cruelty-free: The "cruelty-free" feature indicates that products have been developed without any involvement or harm to animals throughout their production process [59].
Cruelty-free products are not tested on animals and do not contain ingredients derived from animals or animal by-products [60].By choosing cruelty-free products, consumers can support ethical and compassionate practices in the beauty, personal care, and household industries [60].This feature promotes the well-being and rights of animals, encourages the development of alternative testing methods, and contributes to the overall shift towards more ethical and sustainable choices [59].Additionally, cruelty-free products often carry certifications or labels from reputable organizations, providing transparency and assurance to consumers seeking animalfriendly options [60].

•
Upcycled: The "upcycled" feature is a significant attribute of sustainable products, indicating that they have been created by repurposing or transforming waste materials into new and valuable items [61].Upcycling is a process that diverts materials from landfill and reduces the need for new resource extraction.By upcycling, products can give new life to materials that would otherwise be discarded, contributing to waste reduction and resource conservation [61,62].Upcycled products showcase creativity and innovation in design, highlighting the potential for transforming waste into unique and functional items [62].Choosing upcycled products promotes a circular economy mindset and supports sustainable production practices that minimize waste and maximize the value of existing resources.

•
Locally produced: The "locally produced" feature emphasizes products' origin from nearby sources, typically within a specific geographic region [63].Locally produced products support local economies, reduce carbon emissions associated with longdistance transportation, and promote community resilience [64].By choosing locally produced items, consumers can contribute to the vitality of their local communities and support local artisans, farmers, and businesses.Furthermore, local production often involves shorter supply chains, allowing for better transparency, traceability, and a closer connection between consumers and producers [64,65].This feature encourages sustainable consumption by fostering a sense of place, reducing environmental impact, and strengthening local economies [63].

•
Social impact: The "social impact" feature highlights products' positive influence on society and communities.Products with a focus on social impact contribute to social welfare, equality, and empowerment [66].They may support fair trade practices, provide employment opportunities for marginalized groups, or invest a portion of their profits in social initiatives [32,34].By choosing products with a social impact, consumers can align their purchasing power with their values and actively support positive change in society [67].These products often go beyond profit-driven motives and strive to create a lasting, positive difference in the lives of individuals and communities, making them an integral part of the sustainable product landscape [67].

•
Low/zero VOC: The "low/zero VOC" indicates that products contain minimal or no volatile organic compounds (VOCs) [68].VOCs are chemicals that can be released into the air and contribute to indoor and outdoor air pollution, posing risks to human health and the environment [68].Low/zero VOC products prioritize the use of nontoxic or low-emission materials, such as paints, adhesives, and cleaning products [58].By choosing low/zero VOC products, consumers can create healthier indoor environments, reduce their exposure to harmful chemicals, and contribute to improved air quality [69].This feature aligns with sustainable practices that prioritize human well-being and environmental stewardship [21].

Generative Pre-Trained Transformer (GPT) Models and Natural Language Processing (NLP)
Natural Language Processing (NLP) holds a pivotal position within the realm of artificial intelligence (AI) by facilitating machines to comprehend and engage with human language [70].In recent years, the significance of language models, particularly the groundbreaking Generative Pre-trained Transformer (GPT) developed by OpenAI, has witnessed an exponential surge across a broad spectrum of applications [71].The GPT model has exhibited exceptional aptitude in producing coherent and contextually pertinent textual content [71].This section aims to investigate four fundamental aspects pertaining to GPT and NLP.Firstly, an exploration of distinct GPT models shall be undertaken, elucidating their distinct characteristics along with an analysis of the diverse capabilities encompassed by GPT, encompassing text generation, translation, and summarization.Subsequently, the role of prompts in harnessing GPT models for targeted tasks will be examined.Lastly, the study will delve into the advancements observed in the domain of natural language comprehension and processing, underscoring the ramifications of AI-driven solutions on communication, information retrieval, and decision-making processes.

Transformer Architecture
The introduction of transformer architecture has brought about a revolution in the fields of natural language processing (NLP) and machine translation [72].This breakthrough was enabled by the self-attention mechanism, which facilitates parallel processing and enhances the capture of long-range dependencies [71].The incorporation of selfattention layers in the transformer allows the model to weigh the significance of individual words during prediction, leading to improved context comprehension [73].By replacing traditional recurrent neural network (RNN) or convolutional neural network (CNN) processing, the transformer achieves faster training and inference times [74].As a foundational framework, the transformer has elevated language-understanding tasks, serving as the basis for state-of-the-art NLP models such as BERT, GPT, and RoBERTa [71].Its success emphasizes the critical role of attention mechanisms in deep learning and stimulates continuous research in the field of NLP.The transformer's ability to process sequences in parallel contributes to its scalability and adaptability to parallel hardware systems, enhancing its widespread utilization across diverse applications and architectures [75].Moreover, the encoder-decoder structure within the transformer significantly improves machine translation and language generation, representing a pivotal advancement in the realm of natural language processing [76].

The Pre-Training Phase
The pre-training phase of GPT (Generative Pre-trained Transformer) constitutes a pivotal step wherein the model is immersed in vast and diverse text data originating from various sources [71].By harnessing the self-attention mechanism inherent in the transformer architecture, GPT gains the ability to comprehend the contextual relationships within the text and engage in unsupervised learning to predict subsequent words in a sentence [76].GPT models lack true comprehension and knowledge of the world; they do not possess genuine understanding of syntax, semantics, or discourse in the way humans do.Instead, they rely on statistical associations and patterns they have learned from the vast amounts of text during training [77].While they can generate impressive responses, especially in context, they are fundamentally different from human intelligence.As a result of the pre-training phase, GPT acquires the adaptability to be fine-tuned for specific tasks, granting it versatility in generating contextually appropriate text.The exposure to diverse textual data further enriches GPT's understanding of language and enhances its recognition of contextual nuances [78,79].The resounding success of the pre-training phase can be attributed to GPT's remarkable capacity to assimilate knowledge from diverse sources and process vast volumes of data, laying the foundational groundwork for its exceptional language-understanding capabilities [71].

The Fine-Tuning Phase
The fine-tuning phase of GPT, a critical step in its development and is essential for optimizing performance in specific downstream tasks.During this phase, the pre-trained model undergoes further training using task-specific data that include labeled examples [71].By fine-tuning, GPT can adapt its language representation to meet the unique nuances and requirements of the target task [80].This process involves adjusting the model's weights and biases to achieve optimal performance on the specific task at hand [81].The fine-tuning process empowers GPT to hone its language generation capabilities, ensuring contextually appropriate responses for the given task [82].Moreover, it facilitates GPT's specialization in a diverse array of natural language processing tasks, encompassing text classification, summarization, and sentiment analysis [82].Through fine-tuning on diverse datasets, GPT becomes amenable to customization for various real-world applications, elevating its practicality and versatility [71].The success of the fine-tuning phase is intricately tied to the quality and size of the task-specific data used for training, as these factors significantly impact the model's performance on the targeted task [83,84].

Prompts in GPT Models
In the context of GPT models, prompts refer to specific instructions or starting points provided to guide the generation of text [85].Prompts serve as input cues that help shape the output generated by the model, allowing users to influence the direction and content of the generated text [86].These prompts can take various forms, such as sentences, questions, or even keywords, and are designed to elicit desired responses from the model [86].The use of prompts provides a means for users to convey their intentions and preferences, effectively acting as a way to interact and collaborate with the language model [85].
Prompts play a crucial role in guiding the generation of text in GPT models.When a prompt is provided as input, the model leverages its pre-trained knowledge to understand the context and generate text that is coherent and relevant to the given prompt [86].The initial prompt acts as a starting point, influencing the subsequent text-generation process.The model's understanding of grammar, vocabulary, and contextual dependencies allows it to build upon the prompt and generate coherent and contextually appropriate text.By adjusting the prompt, users can control the tone, style, or specific content of the generated text, enabling a more personalized and tailored output [71,87].This flexibility in utilizing prompts empowers users to shape the output according to their desired outcomes, making GPT models a versatile tool for various applications.
Constructing effective prompts requires careful consideration to achieve desired outcomes.One approach is to provide specific instructions or constraints within the prompt to guide the model's response.For example, specifying a format, desired length, or specific keywords can influence the generated text accordingly [87].Another technique is to incorporate context within the prompt by including relevant background information or setting the stage for a particular topic [88].This helps the model generate text that aligns with the intended context.Additionally, iterative prompting can be employed, where users can progressively refine or add to the initial prompt to shape the ongoing text generation process [88].Experimentation and fine-tuning of prompts are essential to optimize the model's performance and ensure the desired output [87].By carefully crafting prompts, users can effectively guide GPT models and harness their capabilities to generate text that meets their specific requirements or objectives [85].

Natural Language Processing (NLP)
Natural language processing (NLP) holds immense significance in the field of AI, as it enables machines to comprehend, interpret, and manipulate human language.NLP plays a crucial role in bridging the gap between human communication and machine understanding [80].It encompasses a wide range of tasks, including speech recognition, text classification, sentiment analysis, machine translation, information extraction, and question answering [80].NLP techniques allow for the extraction of meaningful insights from vast amounts of unstructured textual data, facilitating efficient information retrieval, analysis, and decision-making processes [89].NLP has numerous practical applications, such as chatbots, virtual assistants, language translation services, content summarization, and sentiment analysis in social media.By unlocking the power of language, NLP enables the development of intelligent systems that can understand and communicate with humans in a more natural and intuitive manner.
GPT models make significant contributions to advancing natural language processing capabilities [90].These models, based on Generative Pre-trained Transformer architecture, have revolutionized the field of NLP [71].GPT models have the ability to capture and understand complex linguistic structures, context, and semantics.Through extensive pretraining on vast amounts of unlabeled text data, GPT models acquire a deep understanding of language patterns and relationships [80].This pre-training enables GPT models to generate coherent and contextually relevant text [71].Furthermore, GPT models can be fine-tuned for specific NLP tasks, enhancing their performance and adaptability [90].GPT models excel in various NLP applications, including text generation, language translation, summarization, and content completion [71].By leveraging GPT models, NLP systems can achieve more accurate language understanding, improve the quality of the generated text, and provide users with more contextually relevant and personalized results [90].The capabilities of GPT models contribute to pushing the boundaries of natural language processing, enabling more sophisticated and effective interactions between humans and intelligent systems [89].

Materials and Methods
In Section 2.1, the concept and necessity of sustainability were analyzed, as well as the characteristics that products must possess in order to be considered sustainable.It was also found that sustainability is not solely about the material composition of a product and its environmental footprint, but social and economic factors can also influence it.In Section 2.2, the Generative Pre-trained Transformer (GPT) Models and Natural Language Processing (NLP) were discussed as emerging AI technologies.The concept of pre-training and fine-tuning in GPT models was explained, along with how prompts are used to guide the generation of text, and how GPT models contribute to advancing natural language processing capabilities.

Research Objectives
The primary objective of this study is to propose a novel approach for automating the identification of sustainable product features using GPT-powered software.The proposed software will leverage natural language processing techniques to classify products into distinct sustainability categories based on their descriptions, titles, and product specifications.

Expected Outcomes and Benefits
By achieving this research objective, we anticipate several significant outcomes and benefits:

•
Enhanced Consumer Empowerment: The GPT-powered software will empower consumers to make more informed and sustainable purchasing decisions effortlessly.By automatically extracting sustainability indicators from product descriptions, titles, and product specifications, consumers will have access to crucial environmental and ethical information at their fingertips, enabling them to align their purchasing choices with their sustainability values.

•
Time and Resource Savings: The automated sustainability assessment through the GPT-powered software will streamline the evaluation process for consumers and businesses alike.The software will eliminate the need for time-consuming and laborintensive manual assessments, saving valuable resources and allowing businesses to allocate their efforts to other sustainability initiatives.

•
Environmentally Conscious Business Practices: The proposed approach will encourage businesses to showcase and prioritize their environmentally friendly offerings.By providing automated sustainability classifications for products, the software will motivate businesses to adopt sustainable practices, ultimately contributing to a more eco-conscious market.

Significance and Contribution to Sustainability Research:
This research is highly significant in addressing the existing challenges in sustainability assessment.The increasing global focus on sustainability necessitates efficient and accurate methods for evaluating product eco-friendliness.The proposed GPT-powered software represents an innovative fusion of advanced language models with product classification, bridging the gap between natural language processing and sustainability evaluation.
The integration of GPT technology in sustainability assessment not only ensures more comprehensive and objective analysis but also extends the applicability of natural language processing to environmental and social domains.By combining language-understanding capabilities with machine learning techniques, the software can capture subtle nuances and context-specific sustainable features that might be challenging to identify using traditional methods.
The contribution of this research lies in its potential to revolutionize the way sustainable product features are identified and communicated.By automating sustainability assessment, the proposed approach paves the way for a more sustainable and eco-conscious future.It facilitates a shift towards greener consumption patterns, which can significantly reduce the environmental impact of products and promote circular economy practices.
Overall, the proposed approach holds promising implications for both consumers and businesses, fostering a more sustainable marketplace and supporting efforts towards achieving global sustainability goals.It represents a valuable addition to the field of sustainability research, offering an innovative and efficient solution to address the pressing challenges of sustainability assessment across various industries.

Methodology
To validate the efficacy of the software and the suitability of the GPT model for assessing and identifying sustainable features in products, a specific methodology was employed.A representative sample of products from three well-established marketplaces was gathered, and the GPT model was utilized to evaluate and identify the sustainable features of each product.The subsequent sections detail both the data collection and preprocessing process, as well as the model evaluation based on the responses obtained.

Data Collection and Preprocessing
Ensuring data quality and relevance is crucial to maintain the integrity and reliability of the research findings.In this study, the following data-preprocessing steps were undertaken to achieve this:

•
Data Collection: For the purpose of this research, we utilized the most well-known marketplaces, namely Amazon, eBay, and AliExpress, as reference points.From each of these marketplaces, we collected a total of 30 randomly selected products, spanning across 6 distinct product categories.Care was be taken to cover various industries and product categories to ensure the software's and GPT model's applicability across different domains.

•
Data Cleaning: Raw text data often contain noise, irrelevant information, or typographical errors.Data cleaning involved removing special characters, punctuation, and irrelevant tags or metadata.Additionally, any duplicated or redundant entries were eliminated to enhance data quality.

•
Text Normalization: Text normalization is crucial to standardize the data, ensuring consistency in word representations.This step involves converting all text to lowercase, handling contractions, and applying stemming or lemmatization techniques to reduce words to their base form.

•
Tokenization: The process of tokenization breaks down the textual data into individual tokens (words or subwords).Tokenization is a critical step for language models like GPT, as it allows the software to process and understand text at a granular level.

•
Feature Extraction: Relevant features related to sustainability, such as "eco-friendly materials," "energy efficiency," "recyclability," and "ethical sourcing," will be identified and extracted from the preprocessed text.This step is vital to building a robust classification model that can differentiate products based on their sustainability characteristics.

•
Software assessments: With the GPT-powered Chrome app activated, we accessed the webpage of each product, and the application was tasked with extracting the product title, description, and available product specifications, as applicable.Subsequently, the collected data were transmitted to a remote server, which established communication with the appropriate GPT model, soliciting responses concerning the sustainability features of the product based on its title and description.The resulting outcomes were stored in a CSV file to facilitate further analysis.A comprehensive description of this process can be found in Section 4.2, titled "Software Implementation and System Design."The sample products and software's assessments are readily available in CSV format on GitHub repository [91].
By rigorously implementing these data preprocessing steps, we aim to guarantee the quality and relevance of the dataset used to test the GPT-powered software.These steps ensure that the software can effectively and accurately identify sustainable product features across various domains, providing reliable and actionable insights for consumers and businesses alike.
Upon data collection and processing, the CSV file was imported into a Python script responsible for populating a data frame (df) to facilitate the extraction of meaningful plots.For this purpose, two Python libraries, namely pandas and matplotlib, were utilized.Pandas is a robust Python library extensively employed in data manipulation and analysis tasks, offering essential data structures such as data frames that facilitate the effective management of substantial datasets [92].On the other hand, matplotlib, a widely adopted Python library, specializes in data visualization endeavors, providing a versatile selection of plotting techniques to generate diverse graph types and visual depictions derived from data stored in arrays or data frames [93].
In Figure 1, the percentage of utilization of sustainable features among the products in our dataset is presented, while in Figure 2, the frequency of sustainable feature occurrences categorized by marketplace is illustrated.

Model Evaluation
As stated in Section 2.2, the GPT model utilized in this research is a pre-trained model that has undergone extensive training on diverse textual datasets from various sources.This pre-training process exposed the GPT model to vast amounts of text, enabling it to learn language patterns, syntactic structures, and contextual relationships.The pre-training phase aimed to establish a robust language understanding foundation, which is crucial for subsequent fine-tuning on specific tasks.
For the purposes of this research, the GPT pre-trained model was utilized without undergoing fine-tuning.This decision was based primarily on the strong language and contextual capabilities of the pre-trained model, as well as its training on diverse sources of textual data.Another reason was to assess the quality of responses generated by the pre-trained model for this specific task.
To evaluate the performance of the GPT-powered software in accurately identifying sustainable attributes, a manual inspection of the 90 sampled products was conducted.Additionally, the evaluation process incorporated a comprehensive qualitative analysis of the results.This qualitative assessment sought to identify potential patterns, limitations, or challenges encountered by the GPT-powered software in sustainable feature identification.The objective of this analysis was to gain valuable insights into the model's decision-making process and to pinpoint areas for potential improvements in the proposed approach.

User Interaction
The initial idea was to create a desktop or mobile app that would allow consumers to receive an answer about whether a product has sustainable features or not, either by pasting a link or a product description.However, adding another mobile app to the thousands already existing on our mobile devices, which are likely to never be used, seemed counterproductive.The rationale from the beginning was to facilitate the user making a more sustainable purchasing decision without burdening them in any way.For this reason, we chose to develop a Google Chrome app that, once installed, would function autonomously without any user intervention.
The developed software facilitates user interaction with an e-commerce or marketplace platform.Upon accessing a product page, the accompanying Chrome application assumes the responsibility of extracting the product's title, description, and features.Through communication with a remote server, employing a GPT model, the gathered information undergoes analysis to ascertain the presence or absence of sustainable attributes within the product.Consequently, the system provides the user with a response indicating whether the currently viewed product exhibits sustainable characteristics.In summary, the user's act of opening a product page prompts the GPT-powered Chrome application to deliver information essential for making informed sustainable purchasing decisions.

Software Implementation and System Design
The implementation of the software involved the utilization of various technologies and programming languages.JavaScript, JQuery, and Ajax with Manifest version 3.0 were employed for creating the Chrome APP/Extension [94].On the server side, Python was utilized along with the Flask framework and relevant libraries.For the implementation of the GPT component, Python was responsible for invoking the OpenAI API [95], retrieving the response, and returning it to the Chrome APP, where JavaScript would handle the printing of the results to the user.
A detailed description is provided below, outlining the steps undertaken by the Chrome app to identify and make a decision regarding the sustainable attributes of a product.
1.The user proceeds with the installation of the Chrome app on the Google Chrome browser, utilizing the extensions area in developer mode; 2. The user navigates through one of the prominent marketplaces/e-commerces, such as Amazon, and arrives at a product page; 3. The background-running Chrome app detects the user's presence on a product page and utilizes JavaScript and JQuery to extract the title, description, and characteristics of the product; 4. Through Ajax, the Chrome app initiates an asynchronous POST request to a remote server; 5.The script is hosted on PythonAnywhere [96], a project associated with Anaconda [97], renowned among both data scientists and other users; 6.The server employs Flask [98], a micro Python-based framework, to handle the incoming POST request from the Chrome app; 7. Python is responsible for loading the OpenAI library and generating an appropriate prompt containing the title and description of the product observed by the user; 8. Upon receiving the response from the OpenAI API, Python undertakes the processing of the response data and formats them in a manner suitable for the Chrome app to interpret.These formatted data are then returned to the Chrome APP; 9. Depending on the response, the Chrome app asynchronously displays a badge on the page being viewed by the user.On mouse hover, a popup message appears, presenting text that elucidates the sustainable attributes identified by the GPT for the product.In the scenario where no sustainable attributes are detected, a distinct badge is printed on the page, which, upon hover, notifies the user that no sustainable attributes were found for the specific product.
Figure 3 depicts the prompt employed to elicit the desired response from the GPT model and OpenAI API.The placeholder {product_title} is automatically replaced with the product title, {product_description},which represents the product description, and {sustainable_attributes_to_string} corresponds to a stringified array containing all the sustainable attributes analyzed in Section 2.2.4.
Sustainability 2023, 15, x FOR PEER REVIEW 17 of 27 5.The script is hosted on PythonAnywhere [96], a project associated with Anaconda [97], renowned among both data scientists and other users; 6.The server employs Flask [98], a micro Python-based framework, to handle the incoming POST request from the Chrome app; 7. Python is responsible for loading the OpenAI library and generating an appropriate prompt containing the title and description of the product observed by the user; 8. Upon receiving the response from the OpenAI API, Python undertakes the processing of the response data and formats them in a manner suitable for the Chrome app to interpret.These forma ed data are then returned to the Chrome APP; 9. Depending on the response, the Chrome app asynchronously displays a badge on the page being viewed by the user.On mouse hover, a popup message appears, presenting text that elucidates the sustainable a ributes identified by the GPT for the product.In the scenario where no sustainable a ributes are detected, a distinct badge is printed on the page, which, upon hover, notifies the user that no sustainable a ributes were found for the specific product.
Figure 3 depicts the prompt employed to elicit the desired response from the GPT model and OpenAI API.The placeholder {product_title} is automatically replaced with the product title, {product_description},which represents the product description, and {sustainable_attributes_to_string} corresponds to a stringified array containing all the sustainable a ributes analyzed in Section 2.2.4.For the purposes of this article, the text-DaVinci-003 model, which belongs to the GPT-3 model family, was employed [99].
The text-DaVinci-003 model incorporates several enhancements, including the ability to generate higher-quality writing, thereby facilitating the delivery of clearer, more engaging, and more compelling content through applications.Moreover, it demonstrates improved proficiency in handling complex instructions, enabling users to explore more creative ways to leverage its capabilities.Additionally, the model exhibits enhanced competence in generating longer-form content, expanding its utility to encompass tasks that were previously deemed challenging [100].
Undoubtedly, the software can be readily upgraded in the future to utilize the GPT-4 model, which is presently undergoing limited beta testing and is exclusively accessible to a select group of authorized individuals [101].
The source code of the Chrome APP, both for the client-side and server-side components, is made available as open-source on GitHub [91].

Case Studies and Examples
This section presents two case studies pertaining to the functioning of the software when a user navigates a product page with the GPT-powered Chrome app enabled.Subsequently, an associated badge is displayed to the user following the evaluation of the sustainable features of the product.Upon hovering the mouse over the badge, the user receives a textual representation of all the sustainable features that the GPT model identified within the specific product.For the purposes of this article, the text-DaVinci-003 model, which belongs to the GPT-3 model family, was employed [99].
The text-DaVinci-003 model incorporates several enhancements, including the ability to generate higher-quality writing, thereby facilitating the delivery of clearer, more engaging, and more compelling content through applications.Moreover, it demonstrates improved proficiency in handling complex instructions, enabling users to explore more creative ways to leverage its capabilities.Additionally, the model exhibits enhanced competence in generating longer-form content, expanding its utility to encompass tasks that were previously deemed challenging [100].
Undoubtedly, the software can be readily upgraded in the future to utilize the GPT-4 model, which is presently undergoing limited beta testing and is exclusively accessible to a select group of authorized individuals [101].
The source code of the Chrome APP, both for the client-side and server-side components, is made available as open-source on GitHub [91].

Case Studies and Examples
This section presents two case studies pertaining to the functioning of the software when a user navigates a product page with the GPT-powered Chrome app enabled.Subsequently, an associated badge is displayed to the user following the evaluation of the sustainable features of the product.Upon hovering the mouse over the badge, the user receives a textual representation of all the sustainable features that the GPT model identified within the specific product.

Case Study 1
Figure 4 presents a visual representation of a product page in its initial state before receiving any response from the Chrome app.In Figure 5, the same product page is depicted after the response generated by the Chrome app has been incorporated.Notably, a green badge has appeared above the title, which, upon hovering, displays the GPT model's responses concerning the sustainable features of the product.Figure 4 presents a visual representation of a product page in its initial state before receiving any response from the Chrome app.In Figure 5, the same product page is de picted after the response generated by the Chrome app has been incorporated.Notably, a green badge has appeared above the title, which, upon hovering, displays the GPT model's responses concerning the sustainable features of the product.Figure 4 presents a visual representation of a product page in its initial state before receiving any response from the Chrome app.In Figure 5, the same product page is depicted after the response generated by the Chrome app has been incorporated.Notably, a green badge has appeared above the title, which, upon hovering, displays the GPT model's responses concerning the sustainable features of the product.

Results
The assessment of a domain-specific model typically necessitates the use of domainspecific evaluation metrics.In the domain of sustainability, coupled with the GPT model, there are currently no available evaluation metrics.
To confirm the inadequacy of general evaluation metrics in appraising a sustainable evaluation software, as presented in this specific paper, the BLEU (Bilingual Evaluation Understudy) [102] metric was experimentally employed.Loading the CSV file into the Jupyter Notebook and comparing the product titles and descriptions with the responses generated by the GPT model yielded discouraging results, with a BLEU score of 0.07259360665469289 on average.This score suggests that the model-generated text bears some resemblance to the reference data, but the degree of similarity is relatively low.A higher BLEU score, closer to 1, would indicate a more favorable alignment between the model's outputs and the reference data.
Following the methodology outlined in Section 3.2, human evaluation was conducted to scrutinize each line of the CSV in order to ascertain whether the responses from the GPT-powered software adequately align with the hypotheses presented in the Introduction.

Hypothesis 1
Research Objective: The study seeks to evaluate the accuracy of GPT model responses.


Null Hypothesis (H0): There is no significant accuracy in GPT model responses when appropriate prompts are utilized.


Alternative Hypothesis (H1): The GPT model produces accurate responses when appropriate prompts are employed.
The investigation involved the use of multiple prompts in an iterative manner until the desired and appropriate response was achieved from the GPT model.Throughout this process, the generated responses consistently maintained relevance to the topic and demonstrated precision.This observation indicates that tailoring prompts to elicit specific information from the model can effectively enhance the accuracy and pertinence of its outputs.As a result, the null hypothesis H0 can be rejected, leading to the conclusion that

Results
The assessment of a domain-specific model typically necessitates the use of domainspecific evaluation metrics.In the domain of sustainability, coupled with the GPT model, there are currently no available evaluation metrics.
To confirm the inadequacy of general evaluation metrics in appraising a sustainable evaluation software, as presented in this specific paper, the BLEU (Bilingual Evaluation Understudy) [102] metric was experimentally employed.Loading the CSV file into the Jupyter Notebook and comparing the product titles and descriptions with the responses generated by the GPT model yielded discouraging results, with a BLEU score of 0.07259360665469289 on average.This score suggests that the model-generated text bears some resemblance to the reference data, but the degree of similarity is relatively low.A higher BLEU score, closer to 1, would indicate a more favorable alignment between the model's outputs and the reference data.
Following the methodology outlined in Section 3.2, human evaluation was conducted to scrutinize each line of the CSV in order to ascertain whether the responses from the GPT-powered software adequately align with the hypotheses presented in the Introduction.

Hypothesis 1
Research Objective: The study seeks to evaluate the accuracy of GPT model responses.

•
Null Hypothesis (H0): There is no significant accuracy in GPT model responses when appropriate prompts are utilized.

•
Alternative Hypothesis (H1): The GPT model produces accurate responses when appropriate prompts are employed.
The investigation involved the use of multiple prompts in an iterative manner until the desired and appropriate response was achieved from the GPT model.Throughout this process, the generated responses consistently maintained relevance to the topic and demonstrated precision.This observation indicates that tailoring prompts to elicit specific information from the model can effectively enhance the accuracy and pertinence of its outputs.As a result, the null hypothesis H0 can be rejected, leading to the conclusion that with meticulous prompt selection and refinement, the GPT model can serve as a valuable tool for providing accurate and pertinent information across various contexts.

Hypothesis 2
Research Objective: The study aims to examine the format of responses generated by the GPT model.

•
Null Hypothesis (H0): The responses produced by the GPT model do not conform to the appropriate format.

•
Alternative Hypothesis (H1): The format of GPT model responses is accurate and aligns with the explicitly defined structure in the prompt.
In the course of the research, multiple prompts were utilized to prompt the GPT model to generate responses in a dictionary format suitable for retrieval in the Chrome app via an Ajax request.However, the returned dictionaries presented multiple choices, leading to instances of truncated responses due to the imposed constraint on the maximum tokens per request by OpenAI.Addressing this issue programmatically involved merging the choices and eliminating duplicates, but this did not provide sufficient evidence to confirm the alternative hypothesis (H1).
As a result, the null hypothesis cannot be rejected, and the encountered limitation in response format indicates the need for further investigation and refinement to achieve the desired response structure.Future research efforts should focus on developing more effective methods to ensure that the GPT model consistently delivers complete and unambiguous dictionary-style responses.By achieving this objective, the usability and applicability of the GPT model within the context of the Chrome app can be significantly enhanced.

Hypothesis 3
Research Objective: The study aims to assess the effectiveness of the GPT model in identifying sustainable features based on product titles and descriptions.

•
Null Hypothesis (H0): The GPT model is unable to identify sustainable features of a product when given its title and description.

•
Alternative Hypothesis (H1): The GPT model is capable of evaluating the sustainable features of a product when provided with its title and description.
Following a comprehensive manual evaluation of 90 products within the sample, the responses generated by the GPT model regarding the sustainable features of these products demonstrated a high degree of accuracy.The model exhibited proficiency in recognizing sustainable features through both the identification of relevant keywords and the comprehension of contextual cues present in the product titles and descriptions.As a result, the null hypothesis H0 is rejected, indicating that the GPT model possesses the capability to effectively assess and identify sustainability-related attributes of products.This finding underscores the potential of the GPT model as a valuable tool for sustainable product analysis and evaluation.

Hypothesis 4
Research Objective: The study aims to evaluate the effectiveness of the GPT model in identifying product sustainable features, particularly when product titles and descriptions are less descriptive.

•
Null Hypothesis (H0): The GPT model is incapable of identifying product sustainable features when confronted with limited or less descriptive product titles and descriptions.

•
Alternative Hypothesis (H1): The GPT model can successfully retrieve sustainable features of a product even when the provided title and description lack sufficient descriptive information.
Upon conducting a meticulous manual evaluation, notable observations were made, revealing the GPT model's ability to delve into its own training data and retrieve pertinent information about a product, even in scenarios where the product titles and descriptions were inadequately descriptive.Despite encountering limitations in the available details, the model skillfully employed its pre-trained knowledge to generate accurate responses pertaining to the sustainable features of the products.Consequently, the null hypothesis H0 can be confidently rejected.The GPT model's remarkable adaptability and capacity to complement incomplete information with acquired knowledge from its training data underscore its robustness and versatility in identifying sustainable attributes, regardless of the quality of input descriptions.As a result, the GPT model emerges as a promising tool for effectively assessing sustainability in products with varying levels of descriptive information.

Discussion
In previous sections, an in-depth examination was conducted to elucidate the methodology and tools employed during the software implementation and evaluation process, as well as to address the limitations and challenges encountered.In this section, the authors offer their contemplations regarding potential future enhancements of the software.Additionally, they delve into the ethical considerations surrounding the application of the GPT model in sustainability assessment tasks.

Application of Findings and Future Research Directions
Commencing with the software, it aspires to serve as a solution with the aim of raising awareness among the broader consumer base and encouraging potentially more sustainable purchasing decisions that may positively impact the environment and socioeconomic aspects of life.Concurrently, it can assist product producers in crafting more comprehensive descriptions that accurately delineate the sustainable attributes of their products.The issue of a lack of a sustainable mindset should not solely rest upon consumers.The issue originates from the lack of appropriate information available to consumers.At the consumer level, if individuals were aware that a product is more sustainable than another, without exerting any effort, they would undoubtedly choose the former over a non-sustainable alternative.Furthermore, e-commerce platforms and marketplaces should also assume a corresponding role, either by developing their own tools to inform consumers about the sustainability features of products or by incorporating the automated open-source solution provided by this article simply by integrating the Chrome app's JavaScript into their underlying source code.
From the perspective of GPT and NLP, their integration into emerging technologies is inevitable, and they undeniably have the potential to serve as the cornerstone of an AI technological revolution.However, it would be prudent for the OpenAI API, in particular, to facilitate development by providing documentation not only on how to communicate with the API but also on how to handle the numerous choices that often differ or are truncated due to token limitations.For instance, the "create" method within the completion class could potentially be enhanced to accept parameters defining the expected format of the output, such as JSON encoded, text, or integer representations.Additionally, since the creation of a suitable prompt is highly significant, the API itself should assist developers in generating appropriate prompts through a corresponding tool.This would aid the GPT model in understanding the intention of the prompt and providing the precise responses that developers expect.The ChatGPT, currently based on GPT 3.5/GPT 4, could be leveraged for generating prompts.A corresponding graphical user interface (GUI) could be integrated into the user panel, enabling developers to input the expected outcomes and their respective formats.The GPT model would then respond with the appropriate prompt, assisting the developers in obtaining the desired responses.
As explicated in the section concerning model evaluation, the GPT pre-trained model was employed in this research without undergoing fine-tuning, aiming to assess its existing capabilities in detecting sustainable product features.Subsequent investigations may consider fine-tuning the GPT model by training it with additional knowledge specifically related to sustainability.However, such fine-tuning would necessitate a substantial and well-balanced dataset to accurately train the GPT model to perform sustainability tasks more effectively.

Software Limitations and Challenges
Undoubtedly, no software is immune to problems and challenges, which can be addressed and are presented below:

•
Prompt-related issues: To achieve the desired outcome described in Section 4.2, numerous attempts were made to find the appropriate prompt that would guide the GPT model in understanding the required actions.For instance, despite prompting the model to evaluate the product based on specific attributes, it would still generate its own assessments for attributes not provided as options.This problem was resolved by refining the prompt.

•
Efforts were also devoted to handling the responses returned by the OpenAI API.
After each response, the API provided multiple choices, with each choice containing fragmented answers due to token limitations.By refining the prompt and providing the exact expected format of the result, the answers became more accurate.However, the presence of multiple choices continued to complicate the extraction of results.

•
At present, the software is fully functional with the Amazon marketplace, which has well-structured source code using CSS classes and IDs.In contrast, in similar attempts on eBay, it proved almost impossible to extract product descriptions using JavaScript, as the eBay rich description operates on iframe technology.This issue can be addressed by accessing the product information through the official APIs provided by the marketplaces instead of relying on JavaScript-based extraction.

Ethical Considerations
Ethical considerations surrounding the utilization of the GPT model for identifying sustainable features of a product based solely on its title and description are of paramount importance.While the model's ability to analyze textual data and provide insights can be valuable in promoting sustainable practices, certain ethical concerns must be acknowledged.One major concern is the potential for biased or inaccurate assessments due to the limitations of context and comprehensiveness in the provided information.Relying solely on the title and description may overlook critical details, leading to incomplete or misleading conclusions.Moreover, there is a risk of perpetuating greenwashing practices, where products are falsely labeled as sustainable, leading to consumer deception.To address these ethical challenges, researchers and developers must ensure transparency in the methodology, actively acknowledge and disclose the model's limitations, and consider incorporating Supplementary Materials sources to enhance the accuracy and reliability of the sustainability assessments.Additionally, implementing robust validation mechanisms and engaging with domain experts can help mitigate potential ethical pitfalls and foster responsible and meaningful use of the GPT model for sustainability evaluations.

Chrome Extensions and APPs in Similar Contexts
In the market, various chrome extensions and mobile applications are available, mirroring the proposal presented in this paper.These applications, like our GPT-powered extension, contribute to the promotion of sustainability and incentivize consumers to make more informed and sustainable decisions.For instance, the Ethical Shopper chrome extension informs consumers about the sustainability practices adopted by different brands.The Ecowiser chrome extension aids consumers in selecting more sustainable products from marketplaces like Amazon, by identifying corresponding keywords in the recommended products.Additionally, the Good On You Android app serves as a valuable resource for sustainability ratings in the realm of fashion.
The achievement of sustainability is further attainable with the availability of appropriate tools and applications.By leveraging the capabilities of GPT-powered technologies, along with other innovative approaches, consumers are better equipped to align their choices with sustainability principles, fostering a more eco-conscious future.

Conclusions
In conclusion, this article highlights the growing importance of sustainability in today's industries and the increasing consumer demand for sustainable products.It introduces a novel approach, utilizing GPT-powered software to automatically identify sustainable product features from product descriptions, titles, and product specifications.By leveraging natural language processing and machine learning, the software successfully classifies products into different sustainability categories and extracts key indicators of sustainability.This automated process eliminates the need for manual assessment and enhances the evaluation of product sustainability.The software empowers consumers to make more sustainable purchasing decisions and enables businesses to showcase their environmentally friendly offerings.The experimental results demonstrate the effectiveness and accuracy of the GPT model in the field of sustainability assessment, rendering it a valuable tool for promoting sustainability and paving the way for a more eco-conscious future.Prospective investigations hold the promise of further narrowing the gap between AI models and sustainability, where the latter will be meticulously fine-tuned for specific sustainability objectives.The realization of such advancements bears the potential to augment and optimize human existence, thereby fostering heightened eco-consciousness in decisionmaking processes.

27 Figure 1 .
Figure 1.Percentage of Use of Sustainable Features Across Products.

Figure 2 .
Figure 2. Number of Occurrences of Sustainable Features by Marketplace.3.2.2.Model Evaluation As stated in Section 2.2, the GPT model utilized in this research is a pre-trained model that has undergone extensive training on diverse textual datasets from various sources.This pre-training process exposed the GPT model to vast amounts of text, enabling it to

Figure 1 . 27 Figure 1 .
Figure 1.Percentage of Use of Sustainable Features Across Products.

Figure 2 .Figure 2 .
Figure 2. Number of Occurrences of Sustainable Features by Marketplace.3.2.2.Model EvaluationAs stated in Section 2.2, the GPT model utilized in this research is a pre-trained model that has undergone extensive training on diverse textual datasets from various sources.This pre-training process exposed the GPT model to vast amounts of text, enabling it to learn language pa erns, syntactic structures, and contextual relationships.The pre-train-

Figure 3 .
Figure 3. Prompt to obtain the appropriate response from the GPT model.

Figure 3 .
Figure 3. Prompt to obtain the appropriate response from the GPT model.

Figure 4 .
Figure 4. Screenshot of a product page before receiving the response from the Chrome app and GPT model.

Figure 5 .
Figure 5.The same product page displaying modifications after the response generated by the Chrome app and GPT model.

4. 3 . 2 . Case Study 2 Figure 6
Figure6displays a non-sustainable product along with the corresponding badge tha appears to the user following the evaluation performed by the GPT model.

Figure 4 .
Figure 4. Screenshot of a product page before receiving the response from the Chrome app and GPT model.

Figure 4 .
Figure 4. Screenshot of a product page before receiving the response from the Chrome app and GPT model.

Figure 5 .
Figure 5.The same product page displaying modifications after the response generated by the Chrome app and GPT model.

4. 3 . 2 . Case Study 2 Figure 6
Figure6displays a non-sustainable product along with the corresponding badge that appears to the user following the evaluation performed by the GPT model.

Figure 5 .
Figure 5.The same product page displaying modifications after the response generated by the Chrome app and GPT model.

4. 3 . 2 . Case Study 2 Figure 6
Figure6displays a non-sustainable product along with the corresponding badge that appears to the user following the evaluation performed by the GPT model.

Figure 6 .
Figure 6.The non-sustainable product page displaying modifications after the response generated by the Chrome app and GPT model.

Figure 6 .
Figure 6.The non-sustainable product page displaying modifications after the response generated by the Chrome app and GPT model.