Explainable User Clustering in Short Text Streams

Abstract

User clustering has been studied from different angles: behaviorbased, to identify similar browsing or search patterns, and contentbased, to identify shared interests. Once user clusters have been found, they can be used for recommendation and personalization. So far, content-based user clustering has mostly focused on static sets of relatively long documents. Given the dynamic nature of social media, there is a need to dynamically cluster users in the context of short text streams. User clustering in this setting is more challenging than in the case of long documents as it is difficult to capture the users' dynamic topic distributions in sparse data settings. To address this problem, we propose a dynamic user clustering topic model (or UCT for short). UCT adaptively tracks changes of each user's time-varying topic distribution based both on the short texts the user posts during a given time period and on the previously estimated distribution. To infer changes, we propose a Gibbs sampling algorithm where a set of word-pairs from each user is constructed for sampling. The clustering results are explainable and human-understandable, in contrast to many other clustering algorithms. For evaluation purposes, we work with a dataset consisting of users and tweets from each user. Experimental results demonstrate the effectiveness of our proposed clustering model compared to state-of-the-art baselines.

Collaboratively Tracking Interests for User Clustering in Streams of Short Texts

“…Since the well-known topic models, probabilistic latent semantic indexing [12] and LDA (Latent Dirichlet Allocation) [7], were proposed, topic models with dynamics have been widely studied. These include the Dynamic Topic Model (DTM) [13], Dynamic Mixture Model (DMM) [14], Topic over Time (ToT) [15], Topic Tracking Model (TTM) [16], infinite topic-cluster model [17], and more recently, generalized dynamic topic model [18], dynamic User Clustering Topic model (UCT) [6,10], dynamic topic model for search diversification [19], Dynamic Clustering Topic model (DCT) [20] and scaling-up dynamic model [21]. All of these models except DCT aim at inferring documents' dynamic topic distributions rather than user clustering.…”

“…But, so far, these have not been used for user clustering yet. Zhao et al [6] and Liang et al [10] propose user clustering algorithms in the context of streams of short texts. But they do not take both users' followees and their long-term interest distributions into account during tracking users' interests for clustering, and thus there is still some room to improve the performance.…”

“…We use Twitter as our default setting of streams of short texts and provide an overview of our proposed UCIT/UCIT-L models in Algorithm 1. Following [6,33,34], we represent each user's interests by topics. Thus, the interests of each user u ∈ u t at time period t are represented as a multinomial distribution θ t,u = {θ t,u,z } Z z=1 over topics.…”

“…Most previous work [4,5] on user clustering uses collections of static, long documents, and hence makes the assumption that users' interests do not change over time. Recent work [6] clusters users in the context of streams of short documents, however it ignores any collaborative information, such as friends' messages. Our hypothesis is that accounting for this information is critical, especially for those users with limited activity, infrequent posts, and thus sparse information.…”

Search Result Diversification in Short Text Streams

“…The Dynamic Clustering Topic (DCT) model [29] aims at clustering short documents rather than diversifying search results by a dynamic topic model, where topic distributions of the documents are assumed to change over time. The dynamic User Clustering Topic (UCT) model [51] and User Collaborative Interest Tracking (UCIT) model [28] propose to tackle the problems of user clustering in the context of streaming short texts by topic models. Twitter-LDA [50] is a topical keyphrase extraction LDA-based topic model and assumes that the content of documents generated from Twitter is rich enough for the inference of topic distributions.…”