Empowering neural collaborative filtering with contextual features for multimedia recommendation

A rapid growth in multimedia on various application platforms has made essential the provision of additional assistive technologies to handle information overload issues. Consequently, various multimedia recommendation systems have been developed by the research community. Among these Neural Collaborative Filtering (NCF) is one of the most commonly adopted recommendation frameworks. In this research, we argue that weighing contextual features can help the underlined learning model to develop a better understanding of a user’s behavior. We propose a Weighted Context-based Neural Collaborative Filtering (WNCF) model to supplement weighted contextual information into NCF for learning the user–item interaction function with respect to the different contextual conditions. We introduced an interactive mechanism for addressing the user ratings on items in various contextual situations. Learned contextual weights describe the importance of each item in specific contextual conditions. The proposed model can also assign different weights to the contextual conditions depending on their significance. We performed extensive experiments on three real-world datasets and the outcomes demonstrate the significance of our proposal in comparison with the state-of-the-art models. Empirical results highlight that integrating weighted contextual information with NCF has enhanced recommendation performance. Also, the in-depth analysis leads us toward a completely new research direction on context-aware recommender systems.