Hesham Abdelmotaleb, Małgorzata Wojtyś, Craig McNeile

Abstract: Statistical topic modelling has become an important tool in the text processing field, because more applications are using it to handle the increasing amount of available text data, e.g. from social media platforms. The aim of topic modelling is to discover the main themes or topics from a collection of text documents. While several models have been developed, there is no consensus on evaluating the models, and how to determine the best hyper-parameters of the model. In this research, we develop a method for evaluating topic models for short text that employs word embedding and measuring within-topic variability and separation between topics. We focus on the Dirichlet Mixture Model and tuning its hyper-parameters. We also investigate using the K-means clustering algorithm. In empirical experiments, we present a novel case study on short text datasets related to the telecommunication industry. We find that the optimal values of hyper-parameters, obtained from our evaluation method, do not agree with the fixed values typically used in the literature and lead to different clustering of the text corpora. Moreover, we compare the discovered topics with those obtained from the K-means clustering.

Keywords: Topic modelling, Telecommunication industry, Gibbs Sampling Dirichlet Multinomial Mixture (GSDMM), Model evaluation, Hyper-parameters tuning.

Date Published: December 6, 2023
DOI: 10.11159/jmids.2023.007

View Article