Abstract: Aiming at the defects that traditional user interest models have such as non-dynamicity, noise, high computational complexity, and single dimension of interest evolution, and based on the sliding-window technique, the genetic factor of interest topics is introduced to maintain the continuity of topics, and the common topics of interest for capturing general semantics and noise interference words are defined. An SGC-LDA (Sliding-window, Genetic factor and Common topic-Latent Dirichlet Allocation) user interest topic model was proposed, and the subject evolution analysis was carried out by the proposed model, from the three dimensions of interest topic intensity, interest topic state, and interest topic path to analyze user interest preferences and evolution rules. Sina microblog corpus text is used for empirical analysis. The experimental results show that SGC-LDA user interest topic model is better than the traditional LDA topic model, and can characterize the evolution of user interests accurately. The model has lower non-response rate, false positives rate and normalized costs than the Baseline method without topic association filtering, and the validity of the model is proved.

Key words:  user interests, topic evolution, Latent Dirichlet Allocation (LDA) model, evolution relationship, microblog