A new method for recommendation based on embedding spectral clustering in heterogeneous networks (RESCHet)

The advancement in internet technology has enabled the use of increasingly sophisticated data by recommendation systems to enhance their effectiveness. This data is comprised of Heterogeneous Information Networks (HINs) which are composed of multiple nodes and link types. A significant challenge is effectively extracting and incorporating valuable information from HINs. Clustering has been proposed as one of the main methods in recommender systems, but in Heterogeneous Information Networks for recommender systems has received less attention. In this paper, we intend to present a new method for Recommendation Based on Embedding Spectral Clustering in Heterogeneous Networks (RESCHet), which uses the embedding spectral clustering method, whose similarity matrix is generated by a heterogeneous embedding approach. Subsequently, we employed the concepts of submeta-paths and atomic meta-paths to uncover the relationships between users and items that are pertinent to each cluster. Finally, we generated recommendations for users by computing the Hadamard product between the relevant vectors. Experiments carried out on three open benchmark datasets have demonstrated that RESCHet outperforms current leading methods in a significant manner.