Deep Learning Model Using Transformer Bert and Topic Modeling to Identify Parasocial Interactions in Video Comments of Food Product Review

Abstract

This study proposes a novel approach to identify parasocial interactions in video comments of food product reviews by combining classification and topic modeling techniques. Social media comments often contain subtle and context-dependent expressions, making them difficult to analyze using traditional topic modeling alone. Also, the identification of parasocial interactions using topic modeling are unable to capture the entire context based on parasocial interaction scale items. To address this, we classify comments into eight categories based on the Parasocial Interaction (PSI) scale before applying topic modeling. We compare the performance of Latent Dirichlet Allocation (LDA) and BERTopic, a deep learning-based topic modeling method. Our results show that combining classification with topic modeling yields more contextual topic representations aligned with dimensions of PSI Scale. BERTopic outperforms LDA in capturing semantic relationships at lower topic counts, whereas LDA performs better at higher topic counts. This approach improves topic analysis relevance in social media data and highlights the strengths and limitations of different topic modeling techniques.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint