Soft-Constraint Based Online LDA for Community Recommendation

With the number of social communities grows, social community recommendation has gradually become a critical technique for users to efficiently find their favorite communities. Currently a variety of recommendation techniques have been developed, such as content-based method, collaborative filtering, etc. There methods either easily overfit the data due to the limitation of observations or suffer the heavy computational cost. Besides, they don't consider the relationships between users and communities, and cannot handle incoming users. In this paper, we propose a soft-constraint based online LDA (SO-LDA) method. We use the number of user's posts within each community as soft-constraint to estimate the latent topics across the communities by an online LDA algorithm, in which an incremental method is adopted to facilitate model updating when incomes a new user. Experiment on the well-known MySpace community data shows that the proposed method takes much less time and outperforms the state-of-the-art community recommendation methods.