Ensembling Document and Graph Vectors for Improving Topic Modeling of Microblogs

Microblogging is a popular social networking service in which users interact and post messages. Also, the users are provided with facility to frequently engage in various discussions relating to their areas of interests, fields they want to explore, and people they want to connect with. Topic modeling is a mechanism to detect the category to which a group of words belong to. Employing topic modeling in microblogs is a quite productive way of organizing the content. However, topic modeling of microblogs is relatively an unexplored area. A prime challenge is to automatically segregate the data into the abstract topics that they belong to, so that a user need not manually search for posts of his/her delight, but can easily move to the particular topic which provides the sets of related records. This paper proposes a novel topic modeling method by combining graph based document representations and Bag of Words (BoW) representation in order to classify the microblog data into the topic that it belongs to. Our proposed system is validated through experiments, using real-world data. Our method performs well and helps to efficiently designate documents into its topics.