The ideal topic: interdependence of topic interpretability and other quality features in topic modelling for short texts

Результат исследований: Публикации в книгах, отчётах, сборниках, трудах конференцийстатья в сборнике материалов конференциинаучнаярецензирование

1 Цитирования (Scopus)

Аннотация

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.

Язык оригиналаанглийский
Название основной публикацииSocial 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
РедакторыGabriele Meiselwitz
Место публикацииCham
ИздательSpringer Nature
Страницы19-26
Число страниц8
ISBN (печатное издание)9783030495695
DOI
СостояниеОпубликовано - 2020
Событие12th 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, Дания
Продолжительность: 19 июл 202024 июл 2020

Серия публикаций

НазваниеLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Том12194 LNCS
ISSN (печатное издание)0302-9743
ISSN (электронное издание)1611-3349

конференция

конференция12th International Conference on Social Computing and Social Media, SCSM 2020, held as part of the 22nd International Conference on Human-Computer Interaction, HCII 2020
СтранаДания
ГородCopenhagen
Период19/07/2024/07/20

Предметные области Scopus

  • Теоретические компьютерные науки
  • Компьютерные науки (все)

Ключевые слова

  • Human coding
  • Ideal topic
  • Inter-ethnic discussions
  • Topic modelling
  • Topic quality
  • Twitter

Fingerprint Подробные сведения о темах исследования «The ideal topic: interdependence of topic interpretability and other quality features in topic modelling for short texts». Вместе они формируют уникальный семантический отпечаток (fingerprint).

Цитировать