Discovering Topic Representative Terms for Short Text Clustering

Abstract

Clustering short texts are one of the most important text analysis methods to help extract knowledge from online social media platforms, such as Twitter, Facebook, and Weibo. However, the instant features (such as abbreviation and informal expression) and the limited length of short texts challenge the clustering task. Fortunately, short texts about the same topic often share some common terms (or term stems), which can effectively represent a topic (i.e., supported by a cluster of short texts), and we also call them topic representative terms. Taking advantage of topic representative terms, it is much easier to cluster short texts by grouping short texts into the most similar topic representative term groups. This paper provides a novel topic representative term discovery (TRTD) method for short text clustering. In our TRTD method, we discover groups of closely bound up topic representative terms by exploiting the closeness and significance of terms. The closeness of the topic representative terms is measured by their interdependent co-occurrence, and the significance is measured by their global term occurrences throughout the whole short text corpus. The experimental results on real-world datasets demonstrate that TRTD achieves better accuracy and efficiency in short text clustering than the state-of-the-art methods. INDEX TERMS Short text, clustering, topic representative terms.

ETM: Enrichment by topic modeling for automated clinical sentence classification to detect patients’ disease history

“…Several techniques have been proposed to tackle the challenges posed by sentence classification, including dimension reduction (Zelikovitz and Hirsh 2000;Sriram et al 2010;Khoo et al 2006;Bollegala et al 2018), topic modeling (Cheng et al 2014;Chen et al 2011;Yang et al 2015), clustering (Kozlowski and Rybinski 2019;Yin et al 2017;Bollegala et al 2018;Dai et al 2013;Kozlowski and Rybinski 2017;Yang et al 2019), and word embedding (Kozlowski and Rybinski 2019;Kim 2014;Lee and Dernoncourt 2016;Hill et al 2016). Kim (Kim 2014) proposed a single layer of CNN applied for sentence classification.…”

A Novel Sentence Embedding Based Topic Detection Method for Microblogs

“…Jia et al proposed WordCom [13], which creates concept vectors by identifying semantic word communities in a weighted word co-occurrence network. TRTD [12] Weibo also has a forwarding function, and this process brings about the rapid dissemination and diffusion of information.…”

“…Content may change prior to final publication. in (12), in which a = |S * |, where FMI is an external evaluation method, which is used to determine the similarity between two clusterings. FMI can be calculated as shown in (13).…”

“…First, there is no prior knowledge on the exact number of topics. For many previous topic models [11] [12] [13], the number of topics is an essential parameter. However, in practical application, the number of topics and what they are about are both unknown.…”

“…Feature extraction from short text is another key problem of topic detection models, which is especially useful for blogs in Chinese because they are typically short, dispersed, sparse and noisy [15] [17]. To resolve these problems, researchers usually adopt an embedding representation-based method [18] [17] [19] or a graph-based method [15] [12] [13] to extract topics. Furthermore, these two methods can be used together, similar to using GNN [20] for short text representation generation [21].…”