TopC-CAMF: A Top Context-Based Matrix Factorization Recommender System

Abstract

Online activities have been more and more vital as the digital business has expanded. Users can conduct most activities online such as online shops, hotel bookings, or online educations and courses. A large number of social users are drawn to the abundance of goods available on the Internet. The huge amount of information makes it impossible for social users to navigate it properly and efficiently.  Many companies have offered a personalization to tackle this issue. It is proven that the personalized recommendation systems are able to suggest items to users based on their interests and needs that best suit them, which can be captured from user’s contextual information. However, most of the studies capture this contextual information from the predefined contexts such as location and time. In this study, the personalized user context from the user’s text review that they posted as they gave rating to an item was obtained. To this end, a new approach based on the matrix factorization recommendation model, TopC-CAMF, was proposed. TopC-CAMF investigates and finds the most important contexts or needs for each user by leveraging the deep learning model. First, all important contexts from user’s text reviews were extracted. The next step was representing user preferences with the variations of most important contexts, namely top 5, top 10, top 15, top 20, and top 25 contexts. Then, the best top context variation was evaluated and the optimal one was used as the input for the matrix factorization method in providing better recommendations.  Extensive experiments using three real datasets were conducted to prove the effectiveness of the TopC-CAMF in terms of root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), normalized discounted cumulative gain (NDCG), and Recall.