Making recommendations on microblogs through topic modeling

Abstract

The large-data era has made microblogs important platforms for propagating and searching for information. Analyzing microblog content with topic models facilitates the search for users and microblogs of interest from a vast amount of information. However, traditional topic model doesn’t work well on microblog because these blogs are short and have irregular writing patterns. Considering that microblog have obviously concentration on the field and time they were posted, we propose a microblog recommender approach called time-field latent Dirichlet allocation (TF-LDA), which effectively makes topics more discriminative and thus improves recommender performance. An experiment shows that user and microblog recommendations based on TF-LDA increases accuracy compared with those based on traditional topic models.