Analysing news portal comments

Abstract

With news media popularity switching from newspapers to online news portal formats with comment sections, this has created a repository of articles and comments talking about the article content which was previously not possible. We extracted data ranging 15 months starting from January 2019 to March 2020 containing 14,188 articles and 228,249 comments from 5385 unique users to provide analytical data and metrics regarding the users, articles and their interactions. We used the LDA topic modelling technique to classify according to the topics found in the article. Using comments we also discovered which topics created a more vocal response and how the users reacted by calculating the relevance and sentiment of the users’ comments to the article. Comment relevance was determined by using the LDA model generated by the articles on the comments and comparing the topic assigned with the article, while sentiment was calculated using the rule based sentiment analysis model VADER. Evaluation for how well these approaches classified this information was performed through a usability study involving 10 participants. From the results it emerged that topics were labelled correctly 63.93% of the time, topic relevance was detected correctly 58.14% of the time and sentiment was classified correctly 56.97% of the time. This information was then used to develop a prototype aimed at presenting user and article details in an easy to navigate manner using intuitive design and clear visualisations.