Collaborative tensor-topic factorization model for personalized activity recommendation

摘要

Activity recommendation is a new aspect of location-based social networks (LBSNs) that is being increasingly researched in academia and industry. Previous studies focus mainly on the identification of behavioral regularity by users and use sporadic check-in data, so they suffer severely from data sparsity problems and provide inaccurate recommendations. Furthermore, tips that imply a user's interests and the semantic data available for locations have not been extensively investigated. In this paper, we describe a collaborative tensor-topic factorization (CTTF) model that incorporates user interest topics and activity topics into a tensor factorization framework to create improved activity recommendations for users. We represent user activity feedback with a third-order tensor and penalize false preferences inferred from check-ins using term frequency-inverse document frequency. A biterm topic model was used to learn user interest topics and activity topics from location content information. We learned the latent relations between users, activities, and times by incorporating user interest topics and activity topics into a tensor factorization framework. Experimental results on real world datasets show that the CTTF model outperforms current state-of-the-art approaches.