Start Date

10-12-2017 12:00 AM

Description

As an essential medium for online knowledge sharing and discovery, online community of interests has experienced explosive growth in recent years. Due to the increasingly large number of communities, it is a challenging task for users to find suitable communities for the knowledge sharing and discovery. Recommendation methods, e.g., latent factor-based methods and graph-based methods, are proposed to help users find the interested communities. But they ignore indirect but valuable connections between users and communities and fail to capture the difference of various connections in terms of inferring the relevance between users and communities. This paper presents a novel community recommendation method, where users, communities, items, and their interaction relations are represented as a heterogeneous information network. The proposed method is evaluated based on a public dataset. The evaluation results show that the proposed method significantly outperforms the baseline methods.

Share

COinS
 
Dec 10th, 12:00 AM

Leveraging Heterogeneous Information Network for Community Recommendation

As an essential medium for online knowledge sharing and discovery, online community of interests has experienced explosive growth in recent years. Due to the increasingly large number of communities, it is a challenging task for users to find suitable communities for the knowledge sharing and discovery. Recommendation methods, e.g., latent factor-based methods and graph-based methods, are proposed to help users find the interested communities. But they ignore indirect but valuable connections between users and communities and fail to capture the difference of various connections in terms of inferring the relevance between users and communities. This paper presents a novel community recommendation method, where users, communities, items, and their interaction relations are represented as a heterogeneous information network. The proposed method is evaluated based on a public dataset. The evaluation results show that the proposed method significantly outperforms the baseline methods.