Enhancing Generative AI Shopping Assistants through Advanced Multi-Attribute Decision Making Technique

Authors

• Praveen Kumar Kanumarlapudi Induvidual

Abstract

Abstract: By enabling advanced features such as adaptability, context sensitivity, and human-like interactions, generative AI has revolutionized the capabilities of conversational agents. This article explores how generative AI is improving chatbots in industries such as banking, retail, and consumer technology by improving dynamic responses and delivering personalized, context-aware experiences. It addresses key challenges such as technical limitations, bias, and privacy issues, as well as potential solutions. In addition, the article highlights emerging trends such as multimodal AI and predicts the future of conversational agents - with greater autonomy, emotional and social intelligence, and the ability to handle complex tasks across text, voice, and visual platforms. As a transformative force in the digital landscape, generative AI is set to redefine online communication, drive innovation, and elevate user experiences. By enabling advanced features such as adaptability, context awareness, and human-like interactions, generative AI has transformed the capabilities of conversational agents. This article explores how generative AI is improving chatbots in sectors such as banking, retail, and consumer technology, empowering them to provide dynamic responses while understanding user context and delivering personalized experiences.

Research significance:Generative AI shopping assistants have significant research value due to their potential to revolutionize the online retail landscape. These assistants use advanced machine learning models to provide personalized, real-time customer interactions beyond traditional rule-based chatbots. Research in this area focuses on improving natural language understanding, context-awareness, and response generation, which are critical for creating seamless and human-like shopping experiences.The impact on customer engagement and satisfaction is a key area of ​​focus. By tailoring product recommendations and responses to individual preferences and needs, generative AI shopping assistants can improve the user experience, drive higher conversion rates, and customer loyalty. Research efforts are exploring how these systems can effectively learn from a variety of data sources, including user behavior, feedback, and product information, and are continuously refining their performance.

Mythology:

Alternative:

Content-based Filtering,Collaborative filtering, a hybrid approach, Popularity-based filtering, Retrieval Augmented Generation, Evaluation parameter:Accuracy, Personalization, Cost, and Complexity

Conclusion: Generative AI Shopping Assistant, based on the MOORA methodology.

Result:The findings show that Popularity-based Filteringholds its first position with the highest ranking, while Hybrid Approachhas the lowest ranking.

Keywords: Generative AI, conversational agents, chatbots, adaptive chatbots, context-aware chatbots, multimodal A

Introduction

The study also aims to reveal practical examples of how AI can be used to create content with ChatGPT and graphics with DALL-E and Midjourney, highlighting the potential of this technology for e-commerce businesses. This content provides an overview of the market and outlines various use cases for Manufacturing AI across various e-commerce sectors. It also explores emerging trends that indicate the growing potential of Manufacturing AI in the industry [1].

It focuses on case studies of some of the world’s leading e-commerce companies, illustrating how they are integrating Manufacturing AI technologies to improve operational efficiency and enhance customer experience. The chapter also includes a comparative assessment of Manufacturing AI solutions implemented by these companies. It provides practical explanations of popular Manufacturing AI tools such as ChatGPT, DALL·E, and Mi journey. It assesses the capabilities of these tools and explains how e-commerce businesses that have not yet adopted Manufacturing AI can effectively use them to gain a competitive advantage [2].

Recent advances in information retrieval, natural language processing, and generative AI have led to a rapid increase in the use of voice and text-based intelligent virtual assistants and catboats on major e-commerce platforms such as Amazon, Walmart, and Shopify [3].

These shopping assistants are capable of recommending products, summarizing reviews and product details, comparing items, and addressing queries related to product features, orders, and delivery. The widespread use of these conversational tools largely depends on easy-to-access, convenient, and user-friendly interfaces for customers to meet their diverse shopping needs.Looking to the future, we expect AI to become increasingly central to creating a world of conversational shopping, as digital assistants act as personalized guides for online shoppers [4].

A key and complex challenge in e-commerce is automatically generating useful and relevant recommendation questions that customers can ask these shopping assistants, which is critical to enabling fast and seamless conversations.Unlike traditional catboats, we use generative AI and long chain to build autonomous AI agents capable of handling a variety of tasks, which typically focus on single functions:

Conversational Chat Support: Engaging users in realistic, natural conversations [5].

Unlike traditional catboats, we use generative AI and long chain to build autonomous AI agents capable of handling a variety of tasks, which typically focus on single functions: Conversational Chat Support: Engaging users in realistic, natural conversations.Examples of tasks include:Data Scraping: Extracting product information from various e-commerce websites.Product Search & Recommendations: Providing personalized recommendations based on individual user preferences [6].

Examples of tasks include:Data Scraping: Extracting product information from various e-commerce websites.Product Search & Recommendations: Providing personalized recommendations based on individual user preferences. Generative Artificial Intelligence (AI) represents a major breakthrough by producing synthetic data that closely resembles real-world data. This article provides an in-depth overview of generative AI, covering its definition and basic concepts such as generative adversarial networks (GANs), variational autoencoders (VAEs), and autoregressive models [7].

It explores its application in a variety of fields, including content creation, data augmentation, healthcare, gaming, fashion, finance, retail, cybersecurity, transportation, energy, education, entertainment, law, agriculture, real estate, and manufacturing. The article also highlights future opportunities for generative AI, emphasizing advances in creativity, improved collaboration between humans and AI, and more personalized user experiences.

Generative Artificial Intelligence (AI) is a subset of AI dedicated to generating data that resembles existing datasets. Unlike conventional AI models that typically classify data or predict outcomes from patterns, generative AI models generate new, synthetic data by learning those patterns. This article explores the definition, key concepts, and various applications of generative AI, explores its future potential, and outlines the pros and cons for businesses integrating this technology [8].

Generative AI involves algorithms capable of generating new content, such as text, images, audio, or other data types, that closely resemble a given dataset. These algorithms learn the underlying distribution of the data and create new models that conform to that distribution. Important generative AI models include generative adversarial networks (GANs), variational autoencoders (VAEs), and autoregressive models such as GPT (Generative Pre-trained Transformer) [9].

ChatGPT is just one accessible example of generative AI, a technology made up of algorithms designed to generate new content, such as audio, code, images, text, simulations, and videos. Instead of simply recognizing and classifying data, generative AI generates original information using underlying models - deep learning systems that can perform multiple complex tasks at once.Although this emerging technology has only recently become widely available and still faces numerous challenges and shortcomings, all signs point to rapid progress that could revolutionize many business sectors. According to McKinsey, over the next three to five years, generative AI has the potential to increase operating profits in the apparel, fashion, and luxury industries by a conservative estimate of $150 billion, and possibly as much as $275 billion. By assisting in tasks such as co-design and accelerating content creation, generative AI opens up new opportunities for creativity [12].

It can process a wide variety of unstructured data, such as raw text, images, and videos, and can generate new media formats, including fully written scripts, 3D designs, and lifelike virtual models for video marketing campaigns.This is where generative AI makes a difference. Unlike traditional AI models that rely on fixed rules, generative AI uses advanced algorithms to instantly generate unique, human-like responses. This ability to generate dynamic content that takes into account context and user preferences is revolutionizing Chabot functionality. With generative AI, we are entering an era of adaptive and context-aware interactions that enable continuous learning, improvement, and highly personalized user experiences, opening up new possibilities for Chabot development [14].

Generative AI models excel at generating entirely new responses rather than choosing from pre-set responses. AI-based chatbots offer significant flexibility, adapting to a wide range of interactions. They can learn from user feedback, personalize their responses, and handle unexpected questions more naturally. Examples like Siri and Google Assistant use machine learning to continuously improve personalization and accuracy. Thanks to generative AI, conversational agents are evolving rapidly, becoming smarter, more context-aware, and capable of generating unique responses [16].

Unlike previous AI models that relied on fixed knowledge bases and predefined formats or limited pre-recorded responses, generative AI can generate entirely new content by using information drawn from extensive data sources.Generative AI is a type of artificial intelligence designed to generate new content, whether written, visual, or audio, based on the input it receives. It is particularly valuable in conversational agents because it creates spontaneous, natural-sounding dialogue that is not limited to predefined word templates [17].

By analyzing the context of messages, sequences of words, or characters, it creates coherent sentences or paragraphs that resemble human language. Simply put, generative AI predicts the next word or sentence in a conversation. Instead of choosing from a list of fixed answers, it generates answers tailored to the language and intent within a given context. This ability makes interactions with generative AI chatbots more realistic, smooth, and natural compared to traditional chatbots.Generative AI shopping assistants are transforming how consumers interact with online retail platforms. Unlike traditional chatbots that rely on static responses, these assistants use advanced generative AI technologies to understand context, preferences, and nuances in real time. This allows them to engage customers in natural, human-like conversations, provide personalized product recommendations, answer detailed questions, and assist with order management [18].

By leveraging massive amounts of data, generative AI shopping assistants can analyze customer behavior, preferences, and feedback and continuously improve their responses. This not only improves the shopping experience by making it more intuitive and efficient, but also helps retailers increase customer satisfaction and loyalty. These AI-powered assistants can streamline the buying process by quickly comparing products, summarizing reviews, and providing recommendations tailored to individual needs. As e-commerce continues to grow rapidly, generative AI shopping assistants are becoming essential tools for businesses looking to differentiate themselves in a competitive marketplace. Their ability to create dynamic, context-aware interactions represents a significant step towards more personalized, convenient, and engaging online shopping experiences that will shape the future of retail [20].

Materials And Method

In developing a generative AI shopping assistant, the MOORA (Multi-Objective Optimization Based on Ratio Analysis) methodology was used to evaluate and rank various recommendation system approaches. The goal was to identify the most suitable technique for integration with a retail-focused AI assistant that balances performance, cost, and implementation feasibility.The data items included five widely used recommendation methods: content-based filtering, collaborative filtering, hybrid approach, popularity-based filtering, and retrieval augmented generation (RAG).

Each method was evaluated based on four important decision criteria: accuracy, personalization, cost, and complexity. These criteria were selected because they represent essential trade-offs faced when using AI systems in practical e-commerce environments.

In the methodology, the process began by creating a decision matrix in which each alternative (recommendation method) was evaluated against each criterion. Each criterion was assigned an equal weight of 0.25 to ensure a balanced assessment. Next, the raw data was normalized to remove unit discrepancies and standardize the values.

Then, the weighted normalized decision matrix was calculated by multiplying the normalized scores by the assigned weights.Following this, MOORA’s reference point approach was used to derive a rating value for each alternative. This value reflects the performance of each method compared to the others. Positive values ​​indicate a favorable position, while negative values ​​indicate less performance. Based on these values, final rankings were assigned to each method. A bar chart was created to visually compare the rankings, making it easier to interpret the relative performance of each approach. This structured method ensured an objective, data-driven selection process, allowing us to identify the most practical and effective recommendation system for use within a generative AI shopping assistant.

Alternatives:

Content-based filtering recommends products by analyzing user preferences and item features. It excels at personalization, but can struggle with novelty and diversity.

Collaborative filtering relies on user behavior and similarities between users or items. It is useful for detecting trends, but suffers from cold-start issues.

A hybrid approach combines multiple techniques, such as content-based and collaborative filtering, to improve accuracy and personalization. However, it is complex and resource-intensive.

Popularity-based filtering recommends items based on overall popularity. It is simple and cost-effective, but lacks personalization.

Retrieval Augmented Generation (RAG) uses AI to generate recommendations with contextual understanding. It is powerful but computationally expensive.

Evaluation parameter:

Accuracy refers to how well a recommender system predicts what a user will actually want or select. High accuracy ensures that recommended items are relevant and match user preferences, which increases satisfaction and trust in the system.

Personalization measures the system’s ability to tailor recommendations to individual user behaviors, interests, or history. Better personalization improves user engagement by providing more meaningful and personalized recommendations.

Cost estimates the financial and computational resources required to implement and maintain the recommender system. Lower costs make a system more accessible and scalable, especially for businesses with limited budgets.

Complexity estimates the level of difficulty in building, deploying, and managing the system. High complexity may require more technical expertise and longer development time, which may limit practical use in small or fast-paced environments.

Multi-Attribute Decision Making Technique

The MOORA (Multi-Objective Optimization Based on Ratio Analysis) method is a decision-making framework used to evaluate multiple alternatives across multiple conflicting criteria. It provides a structured and objective approach to identify the most appropriate option when faced with complex decisions, such as selecting the best recommendation system for a generative AI shopping assistant. The content used in the MOORA method includes five main recommendation techniques: content-based filtering, collaborative filtering, hybrid approach, popularity-based filtering, and retrieval-augmented generation.

These alternatives are evaluated using four essential evaluation parameters: accuracy, personalization, cost, and complexity. Each alternative is initially analyzed based on raw performance data for the given criteria. This data is then normalized to remove unit discrepancies and bring all values ​​to a common scale. After normalization, the values ​​are weighted equally (in this case 0.25 for each criterion), creating a weighted normalized decision matrix.

This matrix indicates the relative performance of each alternative considering the importance of each criterion. The evaluation values ​​are calculated using the MOORA method’s reference point approach. This involves comparing beneficial criteria (such as accuracy and personalization) with non-beneficial ones (such as cost and complexity). The resulting values ​​determine how favorableeach recommendation system is. Alternatives are ranked based on their evaluation values ​​– positive values ​​indicate greater relevance, while negative values ​​indicate less desirability.

This ranking helps decision maker’s select the most effective and practical recommendation system to integrate into a generative AI system. The MOORA method provides a balanced, transparent, and data-driven way to compare different recommendation systems.It ensures that both technical performance and practical considerations are taken into account when selecting the best alternative for real-world applications. It provides an overall score for each country or investment option. The alternatives are ranked based on these scores, with the highest indicating the most favorable option according to the selected criteria. This ranking system aids strategic decision-making by highlighting which option offers the most well-rounded benefits across a range of financial objectives.

The systematic and transparent nature of the method helps to fully evaluate international financial options. When applied to global financial analysis, the MOORA method improves decision-making processes in areas such as investment evaluation, portfolio diversification, and assessing country-specific risks, ultimately supporting more informed and efficient financial strategies globally.

1. Analyze And Discussion

Table 1.Generative Ai Shopping Assistant

	Accuracy	Personalization	Cost	Complexity
Content-based Filtering	7.5	8	4	5
Collaborative Filtering	8	7.5	4.5	6
Hybrid Approach	9	8.5	6.5	7.5
Popularity-based Filtering	6	5.5	2.5	3
Retrieval Augmented Generation	8.5	8	5.5	6.5

The comparison table presents five recommendation techniques - content-based filtering, collaborative filtering, hybrid approach, popularity-based filtering and retrieval-augmented generation (RAG) - evaluated on four criteria: accuracy, personalization, cost and complexity, each rated on a 10-point scale. Content-based filtering scores moderately well in accuracy (7.5) and personalization (8), indicating good performance in tailoring recommendations to users' preferences. However, its cost (4) and complexity (5) are relatively low, making it a simple and affordable solution, although it has limitations in versatility and scalability.Collaborative filtering offers slightly better accuracy (8) and a stronger personalization score (7.5). Its cost (4.5) and complexity (6) are slightly higher than content-based methods, reflecting the need for more extensive data and user interaction. It may perform better with larger user bases, but may suffer from cold start issues. The hybrid approach leads in both accuracy (9) and personalization (8.5), combining the strengths of the first two methods. However, it comes at the cost of high cost (6.5) and complexity (7.5), making it more suitable for applications where performance outweighs implementation challenges. Popularity-based filtering is simple, with very low scores for accuracy (6), personalization (5.5), cost (2.5), and complexity (3). It is suitable for quick, general recommendations, but lacks user-specific nuance. Retrieval Augmented Generation balances high accuracy (8.5) and personalization (8), but requires moderate cost (5.5) and complexity (6.5), reflecting its reliance on advanced models and external knowledge retrieval.

Figure1.Generative Ai Shopping Assistant

The bar chart, titled “Generative AI Shopping Assistant,” compares five recommendation methods—content-based filtering, collaborative filtering, a hybrid approach, popularity-based filtering, and retrieval-augmented generation—on four criteria: accuracy, personalization, cost, and complexity. The hybrid approach stands out with high scores in both accuracy (9) and personalization (8.5). However, it also has high cost (6.5) and complexity (7.5), making it a powerful but resource-intensive option that’s suitable for applications that prioritize performance. Retrieval-augmented generation also performs strongly in accuracy (8.5) and personalization (8), offering a good balance of intelligence and user adaptation. With moderate cost (5.5) and complexity (6.5), it’s well-suited for sophisticated recommendation systems that leverage external knowledge. Aggregate filtering maintains solid accuracy (8) and personalization (7.5), but its cost (4.5) and complexity (6) imply a moderate implementation effort. It is a well-rounded option where user interaction data is abundant. Content-based filtering offers moderate accuracy (7.5) and personalization (8), but excels in cost (4) and complexity (5), making it suitable for budget-conscious applications.

Table 2.Normalized Data

Normalized Data
	Accuracy	Personalization	Cost	Complexity
Content-based Filtering	0.4263	0.4724	0.3730	0.3852
Collaborative Filtering	0.4547	0.4429	0.4196	0.4622
Hybrid Approach	0.5116	0.5020	0.6061	0.5778
Popularity-based Filtering	0.3411	0.3248	0.2331	0.2311
Retrieval Augmented Generation	0.4832	0.4724	0.5129	0.5007

The normalized data provides a standardized comparison of five recommendation methods: content-based filtering, collaborative filtering, hybrid approach, popularity-based filtering, and retrieval-augmented generation, across four dimensions: accuracy, personalization, cost, and complexity. Values ​​between 0 and 1 allow for easy cross-benchmarking. The hybrid approach clearly outperforms the others, achieving higher normalized scores for accuracy (0.5116), personalization (0.5020), cost (0.6061), and complexity (0.5778). While this confirms its strength in overall performance, the higher cost and complexity indicate that it may be more suitable for advanced applications where resource investment can be justified. Retrieval Augmented Generation scores highly across all dimensions, especially cost (0.5129) and complexity (0.5007), reflecting a balanced trade-off between performance and operational requirements. Its strong accuracy (0.4832) and personalization (0.4724) make it a strong choice for intelligent and adaptive systems. Collaborative filtering demonstrates consistent, moderate performance across all metrics, with no extreme highs or lows, making it a practical and balanced choice. Content-based filtering performs slightly less well, but offers decent personalization (0.4724) with low cost and complexity, making it suitable for simple systems. Popularity-based filtering ranks lowest across all metrics, indicating minimal sophistication but likely to perform well on basic recommendation tasks.

Table 3. Weight

Weight
	Accuracy	Personalization	Cost	Complexity
Content-based Filtering	0.25	0.25	0.25	0.25
Collaborative Filtering	0.25	0.25	0.25	0.25
Hybrid Approach	0.25	0.25	0.25	0.25
Popularity-based Filtering	0.25	0.25	0.25	0.25
Retrieval Augmented Generation	0.25	0.25	0.25	0.25

The table provides a comparison of five recommender system approaches: content-based filtering, collaborative filtering, hybrid approach, popularity-based filtering, and retrieval-enhanced generation. They are evaluated on four key criteria: accuracy, personalization, cost, and complexity. Each approach is assigned equal weight (0.25) across all criteria, indicating a balanced evaluation framework in which no single factor is prioritized over the others. Content-based filtering relies on individual user preferences and item features, providing greater personalization but often limited in handling novel user situations. Collaborative filtering uses user behavior patterns and is generally good at detecting social trends but can suffer from cold-start issues. The hybrid approach combines multiple techniques, aiming to balance strengths and minimize individual weaknesses, although it can be more complex to implement. Popularity-based filtering is straightforward and cost-effective, recommends widely preferred items, but lacks personalization and may reduce recommendation diversity. Finally, emerging AI approaches such as retrieval-enhanced generation (RAG) generate dynamic recommendations based on retrieved information. While this promises contextual relevance and flexibility, it often incurs high complexity and cost due to its reliance on large language models.

Table 4. Weighted Normalized Dm

Weighted normalized decision matrix
	Accuracy	Personalization	Cost	Complexity
Content-based Filtering	0.1066	0.1181	0.0933	0.0963
Collaborative Filtering	0.1137	0.1107	0.1049	0.1156
Hybrid Approach	0.1279	0.1255	0.1515	0.1444
Popularity-based Filtering	0.0853	0.0812	0.0583	0.0578
Retrieval Augmented Generation	0.1208	0.1181	0.1282	0.1252

The weighted normalized decision matrix provides a refined assessment of the five recommendation system techniques, including their performance – accuracy, personalization, cost, and complexity – and the importance of each criterion. The values ​​reflect normalized and weighted scores, making the comparison more accurate than simple equal-weighted assessments.

The hybrid approach scores the highest in most categories, especially cost (0.1515) and complexity (0.1444). This indicates that despite its inherent complexity, its overall performance makes it a strong contender. It balances personalization and accuracy well, suggesting that it is well suited for applications that require robust, adaptive recommendations. Retrieval Augmented Generation highlights its enhanced capabilities and adaptability, especially in cost (0.1282) and complexity (0.1252). It may be suitable for dynamic and content-rich environments where deep contextual understanding is valuable.

Collaborative filtering is competitive, especially in terms of complexity (0.1156) and accuracy (0.1137), indicating that it is effective but resource-intensive. Meanwhile, content-based filtering shows moderate performance across all metrics, with a slight advantage over personalization. Popularity-based filtering ranks very low across all dimensions, especially in terms of cost (0.0583) and complexity (0.0578), indicating that it is simpler and less customizable, better suited for situations where personalization is not important.

Table 5. Assessment Value

Assessment value
Content-based Filtering	0.0351
Collaborative Filtering	0.0039
Hybrid Approach	-0.0426
Popularity-based Filtering	0.0504
Retrieval Augmented Generation	-0.0145

The rating values, after combining all weighted and normalized criteria, reflect the overall performance of each recommendation system technique. These values ​​indicate the relative desirability of each approach, with positive values ​​indicating above-average performance and negative values ​​indicating underperformance based on the rating model.

Popularity-based filtering has the highest rating value (0.0504), indicating that it performed well overall in this particular evaluation. Despite its simplicity, it likely benefits from lower cost and complexity, making it a practical choice for situations where performance and broad appeal are more important than personalization. Content-based filtering also has a positive rating value (0.0351), indicating a more balanced performance. It may offer a good compromise between personalization and resource demands, making it a solid choice for moderately tailored recommendations.

In contrast, collaborative filtering shows the lowest positive score (0.0039), indicating that it is almost neutral in performance. Reliance on user data may limit its performance in heterogeneous or sparse data environments. The hybrid approach has the lowest score (-0.0426), which may be due to its high complexity and resource requirements outweighing its performance benefits in this environment. Similarly, Retrieval Augmented Generation shows a negative value (-0.0145), likely due to its computational cost despite its robust capabilities.

Table 6. Rank

Rank
Content-based Filtering	2
Collaborative Filtering	3
Hybrid Approach	5
Popularity-based Filtering	1
Retrieval Augmented Generation	4

The ranking values ​​provide a final ranked evaluation of the five recommender system approaches, based on their overall rating of accuracy, personalization, cost, and complexity. A lower ranking number indicates better overall performance within the decision-making framework used.Popularity-based filtering comes in first place, indicating that its simplicity, low cost, and ease of implementation outweigh its limitations in terms of personalization and accuracy. This approach is particularly well-suited for applications where broad recommendations are sufficient, such as popular product recommendations or general content discovery.Content-based filtering comes in second place, representing a strong balance of personalization and manageable complexity. This method is best suited for systems that require moderately tailored recommendations without excessive resource requirements, such as e-commerce sites with extensive product metadata.Collaborative filtering comes in third, performing adequately but not excelling in any particular area. It can be useful in environments with rich user interaction data, but can struggle with cold-start issues. Retrieval Augmented Generation is in fourth place, with its improved customization and contextual capabilities offsetting its higher costs and complexity. It is well suited for dynamic, data-rich environments despite its lower ranking. The hybrid approach is in fifth place, with its higher complexity and cost reducing its overall desirability despite strong individual performance metrics. It may be better suited for high-stakes or highly personalized recommendation systems.

Figure 2.Rank

The bar chart labeled “RANK” displays the comparative ranking of five recommendation techniques used in a generative AI shopping assistant, as determined by the MOORA method. The rankings are based on several criteria, including accuracy, personalization, cost, and complexity. From the chart, popularity-based filtering (green bar) takes the first rank, indicating that it is the most favorable option among the evaluated methods. This is likely due to its simplicity, low implementation cost, and broad applicability in common recommendation scenarios. Content-based filtering (gray bar) follows closely in second place, showing a solid balance between personalization and system management. It works well in environments rich in product information even when user data is scarce. Collaborative filtering (yellow bar) is in third place, indicating moderate performance – strong in personalization but can suffer from high data requirements or cold-start issues. Although it provides advanced contextual recommendations, retrieval augmented generation (dark blue bar) ranks fourth due to its higher complexity and operational costs. The hybrid approach (light blue bar) ranks fifth, indicating that while it may offer greater accuracy and customization, its complexity and resource requirements may make it less practical for some retail applications.

Conclusion

A generative AI shopping assistant using the MOORA method provides a structured, multi-criteria decision-making approach to selecting the most appropriate recommendation technique to improve customer experience in digital retail environments. MOORA (Multi-Objective Optimization Based on Ratio Analysis) helps balance the trade-offs between essential factors such as accuracy, personalization, cost, and complexity. Using this method, five popular recommendation strategies – content-based filtering, collaborative filtering, hybrid approach, popularity-based filtering, and retrieval augmented generation – were systematically evaluated. The analysis revealed that popularity-based filtering achieved the highest overall ranking due to its simplicity, low implementation cost, and minimal complexity. While it lacks deep personalization, it is very effective at quickly and efficiently recommending products that have broad appeal, making it ideal for general-purpose shopping assistants or platforms that serve a broad customer base. Content-based filtering also performed well, providing a good level of personalization with relatively low complexity. Augmented generation of combined filtering and retrieval, while promising in some contexts, suffered from data requirements and high costs. Hybrid approaches, despite their potential for accuracy and customization, ranked very low due to their significant complexity and resource intensity.The MOORA method proves to be a valuable tool for objectively evaluating and selecting recommendation techniques for a generative AI shopping assistant. It emphasizes the importance of aligning technical capabilities with practical constraints such as cost and scalability. While advanced models may provide better personalization, simpler models such as popularity-based filtering or content-based filtering may be more suitable for real-world applications, especially in resource-constrained or high-traffic environments. MOORA’s integration ensures data-driven, transparent decision-making, allowing businesses to deploy AI assistants that are not only intelligent but also efficient and strategically aligned with business goals.

References:

1. Szumniak-Samolej, Justyna. "The Role of Generative AI in E-commerce."
2. Vedula, Nikhita, Oleg Rokhlenko, and ShervinMalmasi. "Question suggestion for conversational shopping assistants using product metadata." In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 2960-2964. 2024.
3. Ramya, B., M. Kiran, P. Nagendra, K. Srinivas, and P. Harsha. "Conversational AI Assistant for Online Shopping." International Journal of Innovative Science and Research Technology 10, no. 4 (2025): 1027-1033.
4. Oluwagbenro, MogbojuriBabatunde. "Generative AI: Definition, Concepts, Applications, and Future Prospects." Authorea Preprints (2024).
5. Harreis, Holger, Theodora Koullias, Roger Roberts, and Kimberly Te. "Generative AI: Unlocking the future of fashion." McKinsey & Company (2023).
6. Bhoyar, Manoj. "Enhancing Conversational Agents with Generative AI: A Framework for Creating More Adaptive and Context-aware chatbots." (2023).
7. Sterne, Jim, and Thomas H. Davenport. "Improving voice of the customer analysis with generative AI." Applied Marketing Analytics 10, no. 1 (2024): 32-41.
8. Costa, Charles, and Souvick Ghosh. "Empowering customer service with generative AI: enhancing agent performance while navigating challenges." Information Research 30, no. iConf (2025) (2025): 150.
9. Grewal, Dhruv, Cinthia B. Satornino, Thomas Davenport, and AbhijitGuha. "How generative AI Is shaping the future of marketing." Journal of the Academy of Marketing Science (2024): 1-21.
10. Mogaji, Emmanuel, and Varsha Jain. "How generative AI is (will) change consumer behaviour: Postulating the potential impact and implications for research, practice, and policy." Journal of consumer behaviour 23, no. 5 (2024): 2379-2389.
11. Shevchyk, Yurii. "Generative AI in e-commerce: a comparative analysis of consumer-brand engagement with AI-generated and human-generated content." PhD diss (2024).
12. Lee, Gun Ho, Kyoung Jun Lee, BaekJeong, and Taekyung Kim. "Developing personalized marketing service using generative AI." IEEE Access 12 (2024): 22394-22402.
13. Lee, Gun Ho, Kyoung Jun Lee, BaekJeong, and Taekyung Kim. "Developing personalized marketing service using generative AI." IEEE Access 12 (2024): 22394-22402.
14. Naik, Dishita, IshitaNaik, and NitinNaik. "Applications of AI chatbots based on generative AI, large language models and large multimodal models." In The International Conference on Computing, Communication, Cybersecurity & AI, pp. 668-690. Cham: Springer Nature Switzerland, 2024.
15. Kshetri, Nir, Yogesh K. Dwivedi, Thomas H. Davenport, and NikiPanteli. "Generative artificial intelligence in marketing: Applications, opportunities, challenges, and research agenda." International Journal of Information Management 75 (2024): 102716.
16. Sai, Siva, RevantSai, and VinayChamola. "Generative AI for Industry 5.0: Analyzing the impact of ChatGPT, DALLE, and other models." IEEE Open Journal of the Communications Society (2024).
17. Zhang, Shuo, BociPeng, Xinping Zhao, Boren Hu, Yun Zhu, Yanjia Zeng, and Xuming Hu. "Llasa: Large language and e-commerce shopping assistant." arXiv preprint arXiv:2408.02006 (2024).
18. Recchia, Francesco. "Next-Generation Cloud-Based Chatbots: Enhancing Customers Interactions Using Generative AI on AWS." PhD diss., Politecnico di Torino, 2024.
19. Ding, Min, Songting Dong, and Rajdeep Grewal. "Generative AI and usage in marketing classroom." Customer Needs and Solutions 11, no. 1 (2024): 5.
20. Chui, Michael, Eric Hazan, Roger Roberts, Alex Singla, and Kate Smaje. "The economic potential of generative AI." (2023).
21. Genc, ElifGuneren, and OzlemDenizBasar. "Comparison of country ratings of credit rating agencies with moora method." Business and Economics Research Journal 10, no. 2 (2019): 391-404.
22. Karande, Prasad, and Shankar Chakraborty. "Application of multi-objective optimization on the basis of ratio analysis (MOORA) method for materials selection." Materials & Design 37 (2012): 317-324.
23. Brauers, Willem Karel M., RomualdasGinevičius, and ValentinasPodvezko. "Regional development in Lithuania considering multiple objectives by the MOORA method." Technological and economic development of economy 16, no. 4 (2010): 613-640.
24. Dincer, Hasan. "Profit-based stock selection approach in banking sector using Fuzzy AHP and MOORA method." Global Business and Economics Research Journal 4, no. 2 (2015): 1-26.
25. Brauers, Willem KM. "Location theory and multi-criteria decision making: an application of the MOORA method." Contemporary Economics 12, no. 3 (2018): 241-252.
26. Brauers, Willem K., and Edmundas K. Zavadskas. "Robustness of the multi‐objective MOORA method with a test for the facilities sector." Technological and economic development of economy 15, no. 2 (2009): 352-375.
27. Brauers, Willem Karel M. "Multi-objective contractor’s ranking by applying the MOORA method." Journal of Business Economics and management 4 (2008): 245-255.
28. Ersöz, Filiz, CelalHakanKıncı, and TanerErsöz. "A model proposal for course selection with the fuzzy MOORA approach." AvrupaBilimveTeknolojiDergisi 14 (2018): 369-377.
29. Murat, Dilek, and SimlaGüzel. "Determining the Inclusive Development Performances of the BRICS Countries and Turkey with MULTIMOORA." EskişehirOsmangaziÜniversitesiİktisadiveİdariBilimlerDergisi 17, no. 2 (2022): 535-560.
30. Başar, Ö. D., and M. Turanlı. "Ranking of high human development countries with respect to variables relating with female living standards by using MOORA method." Internatonal Journal of Gender and Women's Stud es 6, no. 1 (2018): 111-120.