Abstract

Background. Topic modelling is a method of automated probabilistic detection of topics in a text collection. Use of topic modelling for short texts, e.g. tweets or search engine queries, is complicated due to their short length and grammatical flaws, including broken word order, abbreviations, and contamination of different languages. At the same time, as our research shows, human coding cannot be perceived as a baseline for topic quality assessment. Objectives. We use biterm topic model (BTM) to test the relations between two topic quality metrics independent from topic coherence with the human topic interpretability. Topic modelling is applied to three cases of conflictual Twitter discussions in three different languages, namely the Charlie Hebdo shooting (France), the Ferguson unrest (the USA), and the anti-immigrant bashings in Biryulevo (Russia), which represent, respectively, a global multilingual, a large monolingual, and a mid-range monolingual type of discussions. Method. First, we evaluate the human baseline coding by providing evidence for the Russian case on the coding by two pairs of coders who have varying levels of knowledge of the case. We then measure the quality of modelling on the level of topics by looking at topic interpretability (by experienced coders), topic robustness, and topic saliency. Results. The results of the experiment show that: 1) the idea of human coding as baseline needs to be rejected; 2) topic interpretability, robustness, and saliency can be inter-related; 3) the multilingual discussion performs better than the monolingual ones in terms of interdependence of the metrics. Conclusion. We formulate the idea of an ‘ideal topic’ that rethinks the goal of topic modelling towards finding a smaller number of good topics rather instead of maximization of the number of interpretable topics.

Original languageEnglish
Title of host publicationSocial Computing and Social Media. Design, Ethics, User Behavior, and Social Network Analysis - 12th International Conference, SCSM 2020, Held as Part of the 22nd HCI International Conference, HCII 2020, Proceedings
EditorsGabriele Meiselwitz
Place of PublicationCham
PublisherSpringer Nature
Pages19-26
Number of pages8
ISBN (Print)9783030495695
DOIs
Publication statusPublished - 2020
Event12th International Conference on Social Computing and Social Media, SCSM 2020, held as part of the 22nd International Conference on Human-Computer Interaction, HCII 2020 - Copenhagen
Duration: 19 Jul 202024 Jul 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12194 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference12th International Conference on Social Computing and Social Media, SCSM 2020, held as part of the 22nd International Conference on Human-Computer Interaction, HCII 2020
CountryDenmark
CityCopenhagen
Period19/07/2024/07/20

Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'The ideal topic: interdependence of topic interpretability and other quality features in topic modelling for short texts'. Together they form a unique fingerprint.

  • Cite this

    Blekanov, I. S., Bodrunova, S. S., Zhuravleva, N., Smoliarova, A., & Tarasov, N. (2020). The ideal topic: interdependence of topic interpretability and other quality features in topic modelling for short texts. In G. Meiselwitz (Ed.), Social Computing and Social Media. Design, Ethics, User Behavior, and Social Network Analysis - 12th International Conference, SCSM 2020, Held as Part of the 22nd HCI International Conference, HCII 2020, Proceedings (pp. 19-26). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12194 LNCS). Springer Nature. https://doi.org/10.1007/978-3-030-49570-1_2