Explainable User Clustering in Short Text Streams

摘要

User clustering has been studied from different angles: behaviorbased,rnto identify similar browsing or search patterns, and contentbased,rnto identify shared interests. Once user clusters have beenrnfound, they can be used for recommendation and personalization.rnSo far, content-based user clustering has mostly focused on staticrnsets of relatively long documents. Given the dynamic nature ofrnsocial media, there is a need to dynamically cluster users in therncontext of short text streams. User clustering in this setting isrnmore challenging than in the case of long documents as it is difficultrnto capture the users’ dynamic topic distributions in sparserndata settings. To address this problem, we propose a dynamic userrnclustering topic model (or UCT for short). UCT adaptively tracksrnchanges of each user’s time-varying topic distribution based bothrnon the short texts the user posts during a given time period and onrnthe previously estimated distribution. To infer changes, we proposerna Gibbs sampling algorithm where a set of word-pairs fromrneach user is constructed for sampling. The clustering results arernexplainable and human-understandable, in contrast to many otherrnclustering algorithms. For evaluation purposes, we work with arndataset consisting of users and tweets from each user. Experimentalrnresults demonstrate the effectiveness of our proposed clusteringrnmodel compared to state-of-the-art baselines.