Leveraging heterogeneous information based on heterogeneous network and homophily theory for community recommendations

Online communities benefit information sharing on various social platforms, but their quantity is getting large and hinders online users to join appropriate communities. Personalized recommendation methods have been designed to suggest appropriate communities; however, previous recommendation methods consider communities as virtual users and ignore heterogeneous relations that connect users with communities via various items. We aim to integrate the heterogeneous information into community recommendations and propose a new community recommendation approach based on heterogeneous network and homophily theory. The proposed approach models users, communities, items, and their relations as a heterogeneous network and extracts information based on homophily theory from the network to measure user-community proximities. Next, it incorporates the proximities into a collaborative filtering method to generate recommendations. Experiments on real-world data show that the proposed approach provides better recommendations than multiple baseline methods and the improvements of the proposed approach are significant. This is the first research that employs homophily theory for community recommendations with heterogeneous network. This research not only extends the application scope of homophily theory but also provides a new solution to social platforms with numerous communities.