Analyzing User Interests and Emotions for Product Recommendations Based on Interactive Social Media Content

Abstract

Online social networks have witnessed remarkable growth, with millions of users sharing data daily. This has significantly impacted social commerce, which combines online shopping with social networking to connect businesses and customers. Businesses leverage recommendation systems to examine purchase histories and propose products that capture customer interest. However, traditional product recommendation systems have two main drawbacks: repetitive recommendations and difficulty in predicting new items, generally known as the cold-start problem. Another problem that faces the user's topical interests and behavior mining is representing the user's content through the bag-of-words model, which may include items in the predicted interests that the user no longer finds interesting. In this paper, we introduce a new framework for building user profiles that capture both explicit and implicit topics of interest, providing insights into users' needs. Additionally, the framework identifies five emotional scores-joy, anger, sadness, fear, and disgust-that characterize each user and are used to create personality profiles. We investigate the possibility of determining the interests of users who are not active on social networks called cold-start users. We utilize real-world user data from Twitter (X) and focus on interactive tweets to enhance the accuracy of interest and emotion detection by introducing a novel scoring mechanism for tweet interactivity. This framework represents an intermediate stage in our research, serving as a foundation for future work in building a comprehensive product recommendation system. Results presented in this paper are very promising.