Item Recommendation via Latent Topic Tag
Rende Li, Qiang Guo, Qiang Yue, Jian-Guo Liu · 2018
Overwhelmed by the wide scope and variety of contents, users are facing difficult choices of selecting items. Semantics-based recommendation technology has recently received a lot of attention, since the new knowledge can be found through the semantic relation between items, which improved the content-based recommendation on user-item bipartite graphs. In this paper, we employ a hybrid method, using Latent Dirichlet Allocation (LDA)to form the topic tag and using Higher-Order Singular Value Decomposition (HOSVD)to recommend items on user-tag-item tripartite graphs. The precision of our method outperforms the traditional collaborative filtering (CF)algorithm based on user-item bipartite graphs and Topic-Sensitive PageRank system (TSPR)algorithm based on user-tag-item tripartite graphs. In addition, the performance HOSVD achieve the best when extracting 60% core information from tensor. This work could be applied to mobile app platforms recommending top-l item for.