Mining Users' Intentions from Thai Tweets Using BERT Models

Abstract

In this paper, we explore the mining of users' intentions in text. We view that being able to identify the intentions of users expressed in textual data allows us to specifically know their aims and what they want to do. In the experiment, we collected tweets, constructed a Thai intention corpus, and performed a binary classification task on the corpus. We investigated the intent classification results derived through the application of three different Bidirectional Encoder Representations from Transformers (BERT) models, a word-embedding model, and a bag-of-words model. The results revealed that the BERT Base EN-TH Cased model outperforms the other models in both classification and processing-time aspects: it achieves an F1 score of 0.81 and performs the classification task up to 15% faster than the other BERT models.

Keywords: intent mining, intention mining, intent classification, intent detection, text mining, natural language processing

References

Download paper (PDF) Download BibTeX

@article{2023_Sanchan,
  title   = {Mining Users' Intentions from Thai Tweets Using BERT Models},
  author  = {Sanchan, Nattapong},
  journal = {Journal of Information Science and Technology},
  volume  = {13},
  number  = {1},
  pages   = {17--25},
  year    = {2023}
}

Sanchan, N. (2023). Mining Users' Intentions from Thai Tweets Using BERT Models. Journal of Information Science and Technology, 13(1), 17–25.

More information

This paper is a continuing work in the Thai Intention Mining project:

1. Mining Users' Intentions from Thai Tweets Using BERT Models
2. Intent Mining of Thai Phone Call Text Using a Stacking Ensemble Classifier with GPT-3 Embeddings