Latent Factor Model with User and Fused Item Embeddings for Recommendation
Dahui Zhu, Jianxiong Tang, LI Qia · 2020
Collaborative filtering (CF) is commonly used in recommender systems, and latent factor model (LFM) is one of the most popular methods for CF. Though LFMs achieve great success in CF, most of them are not satisfactory enough for data with negative feedback. Recently, embedding techniques are applied in LFMs and enhance the performance. However, it is hard to keep the convergence speed and evaluation metrics in balance since the joint learning embeddings bring a large number of trainable parameters. By using items and users embeddings, in this paper, we propose an improved LFM. Different from the existing embedding based LFMs, the proposed model learns item latent representations from the combination of co-liked item co-occurrence matrix (LICM) and co-disliked item co-occurrence matrix (DICM). This property reduces the trainable parameters and brings a quick convergence. Experimental results on MovieLens datasets show that the proposed model performs well.