Personalized Item Ranking from Implicit User Feedback: A Heterogeneous Information Network Approach

In today’s era of the digital world with information overload, generating personalized recommendations for the e-commerce users is a challenging and interesting problem. Recommendation of top-N items of interest to a user of e-commerce is highly challenging using binary implicit feedback. The training data is usually very sparse and have binary values capturing a user’s action or inaction. Due to the sparseness of data and lack of explicit user preferences, the recommendations generated by model-based and neighborhood-based approaches are not effective. Of late, network-based item recommendation methods, which utilize item related metainformation, are beginning to attract increasing attention for binary implicit feedback data. In this work, we propose a heterogeneous information network based recommendation model for personalized top-N recommendations using binary implicit feedback data. To utilize the potential of meta-information related to items, we utilize the concept of meta-path. To improve the effectiveness of the recommendations, the popularity of items and interest of users are leveraged simultaneously. Personalized weight learning of various meta-paths in the network is performed to determine the intrinsic interests of users from the binary implicit feedback data. To show the effectiveness, the proposed model is experimentally evaluated using the real-world dataset.