Outer product enhanced heterogeneous information network embedding for recommendation

With the rapid development of the internet, more and more sophisticated data can be utilized by recommendation systems to improve their performance. Such data consist of heterogeneous information networks (HINs) made up of multiple nodes and link types. A critical challenge is how to effectively extract and apply the useful HIN information. In particular, the embedding-based recommendation approach has been widely used, as it can extract affluent semantic and structural information from HINs. However, the existing HIN embedding for recommendation methods only combine user embedding and item embedding through a simple concatenation or elementwise product, which does not suffer for an efficient recommendation model. In order to extract and utilize more comprehensive and subtle information from the embedding for recommendation, we propose Outer Product Enhanced Heterogeneous Information Network Embedding for Recommendation, called HopRec. The main idea is to utilize the outer product to model the pairwise relationship between user HIN embedding and item HIN embedding. Specifically, by performing an outer product between user HIN embedding and item HIN embedding, we can obtain a two-dimensional interaction matrix. Subsequently, we can obtain a rating prediction function by integrating matrix factorization (MF), user HIN embedding, item HIN embedding and interaction matrix. The results of experiments conducted on three open benchmark datasets show that HopRec significantly outperforms the state-of-the-art methods.

Automated summary

With the advancement of the internet, recommendation systems are utilizing complex data for improved performance. A new embedding method called HopRec is proposed to enhance recommendation by capturing relationships through outer product, showing significant superiority in experiments over existing methods.