A Recommendation Method For Fusion of Multiple HIN Meta-Paths

With the rapid development of information technology and the Internet, people have stridden into the era of "information overload". The personalized recommendation is an effective way to solve the problem of "information overload" and plays a pivotal role in people's daily life. Feature extraction based on heterogeneous information networks can improve the performance of recommendation systems. This paper considers different search paths between nodes and proposes a recommendation method that integrates multiple heterogeneous information network meta-paths. Combining different meta-paths in a personalized nonlinear way, the network representation learning vector is generated through the meta-path and deepwalk algorithm. The embeddings of users and items are spliced and MLP is used for training and scoring prediction to improve the performance of the recommender system. The experimental results show that on the Yelp dataset and Douban Book dataset, the proposed model is better than other algorithms. In terms of improving the recommendation effectiveness, the MAE and RMSE are reduced by 5.4%~15%, which has good effectiveness and interpretability.