Eir-Ripp: Enriching Item Representation for Recommendation with Knowledge Graph

To improve the performance of recommendation models and enhance the interpretability of results, knowledge graph is often used to add rich semantic information to the recommended items. Existing methods either use knowledge graph as an auxiliary information to mine users' interests, or use knowledge graph to establish relationships between items via their hidden information. However, these methods usually ignore the interaction between users and items. As a result, the hidden relationship between users and items are not well explored in the item representation. To address this issue, we propose an enhancement model to learn item representation based on RippleNet (Eir-Ripp). By mining the users' historical behavior and user characteristics, users' preference and the correlation between users and items are extracted to complement the semantic information of items. We conduct extensive experiments to evaluate our proposal on three public data sets. Experimental results show that our model outperforms the baseline methods in terms of an up to 8.8% improvement in the recommendation.