Analyzing User Interaction Patterns to Improve Chatbot Recommendations

Abstract E-commerce chatbots play a vital role in assisting users with product searches, recommendations, and order-related queries, enhancing the overall online shopping experience. This study investigates user interaction patterns in chatbot sessions and evaluates the impact of AI-based recommendation improvements using a small anonymized dataset of 10 sessions. Key parameters such as session length, query types, user feedback, and item click behavior were analyzed to identify patterns affecting recommendation acceptance. A simple AI simulation was applied to generate improved recommendations, which were then compared with original user feedback. Results indicate that product recommendation queries demonstrate higher acceptance rates compared to general product searches or order-status queries. Longer session lengths were associated with multi-intent interactions and more complex recommendation scenarios, while shorter sessions often corresponded to single-intent queries. The simulated AI-based recommendations improved prediction accuracy, particularly for sessions involving multiple recommendations. The study highlights the importance of incorporating session-aware analysis, user feedback, and query type categorization in chatbot design to optimize recommendation effectiveness. Limitations of the study include the small dataset size and exclusion of multimodal inputs such as voice or images. Future work can focus on expanding the dataset, integrating real-time AI ranking models, and exploring adaptive recommendation strategies in live e-commerce environments. Overall, this research underscores the potential of combining user interaction analysis with AI-driven recommendations to enhance chatbot performance, user engagement, and customer satisfaction in e-commerce settings. Keywords- Chatbot, E-commerce, AI-based recommendations, User interaction patterns, Session-aware modeling, Feedback analysis