User graph topic model

Finding coherent topics in Twitter data is difficult task because of the sparseness and informal language. Tweets also provide rich contextual and auxiliary metadata which can be used to supervise the topic modeling to get more coherent topics. In this paper, a novel topic model is proposed which extends Author Topic Model for twitter. Standard Author Topic Model cannot be used on Twitter data as every tweet has exactly one author. The proposed User Graph Topic Model (UGTM) considers the semantic relationships among tweet users based on the contextual information like hashtags, user mentions and replies to make a user graph. Related users of author of a tweet are found and used in tweet generation process. Related user information from the user graph is used to obtain the dirichlet prior for user generation. Empirical results show that the proposed UGTM outperforms standard Author Topic Model (ATM) on experimental data.