Film recommendation using scenario-based approach to solve the cold-start problem for new users

Recommendation systems are a crucial component of modern internet activities. Typically, these systems rely on ratings given by users as the primary feature; however, this approach makes the system susceptible to the cold-start problem. To address this issue, the present study proposes using sentiment polarity as a feature in recommendation systems, especially in scenarios where user history information is unavailable. Subsequently, film ratings are used to evaluate the weighted score of sentiment polarity. For systems with access to user history data, a similarity-based algorithm is employed. In this study, the Fine-Tuned BERT + CNN model is utilised as a sentiment classifier. Additionally, Cosine Similarity is applied to calculate the semantic similar ity between two vectors—specifically, the query and the document—by leveraging BERT to capture the contextual meaning of each, thereby enhancing the accuracy of the outcomes.