Topic modeling with latent Dirichlet allocation for cancer disease posts


Altintas V., Albayrak M. , TOPAL K.

JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY, vol.36, no.4, pp.2183-2196, 2021 (Journal Indexed in SCI) identifier

  • Publication Type: Article / Article
  • Volume: 36 Issue: 4
  • Publication Date: 2021
  • Doi Number: 10.17341/gazimmfd.734730
  • Title of Journal : JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY
  • Page Numbers: pp.2183-2196
  • Keywords: Natural language processing, topic modelling, text mining, latent Dirichlet allocation, social media, REVIEWS, LDA

Abstract

In social media platforms, users share their experiences about the events they have experienced. People talk about a recent event, a city they have just seen, a book they read, etc. They post their experiences with other people about the same specific issues. One of the topics that users often talk about is health problems and sharing their experiences on this subject. Individuals with health problems can share their illnesses, treatments and results, and the experiences they have gained at each stage in social media platforms. These shares are important for other patients, both for informative and for morale / motivation in combating the disease. Manual analysis of the posts by human beings becomes impossible due to reasons such as the high number of posts, the variety of diseases and the amount of data. In this study, posts about cancer disease were collected on the Reddit social platform and these data were studied. The main topics discussed with the " Latent Dirichlet Allocation (LDA)" algorithm, one of the artificial intelligence-based topic modeling algorithms, were found through these posts. The relationship of the subject headings with the spoken subject was examined and content analysis was made. It is aimed to determine the most talked about contents among the posts about cancer disease. In addition, the relationship between the subjects was examined using the tSNE technique. It was observed that the words in the topics obtained as a result of modeling with the LDA algorithm were compatible in the coherence test.