Leveraging Transformer-Based Topic Modeling using BERTopic for Publication Venue Recommendation

Topic modeling has become essential in a variety of text mining applications, such as document clustering and recommendation systems. This study investigates the potential of BERTopic, a transformer-based method that leverages BERT embeddings to recommend publication venues. Traditional topic modeling methods are made better by BERTopic, which uses the deep contextual embeddings that BERT (Bidirectional Encoder Representations from Transformers) offers. This investigation assesses the efficacy of BERTopic in clustering academic papers according to their content in order to suggest appropriate publication venues. We also compare the performance between different kinds of pre-trained sentence embedding models and input types (title, abstract, and title+abstract). The experimental results indicate that BERTopic performs well with the GARUDA dataset, which comprises 5272 records from three different research subjects. The results emphasize the transformative influence of the transformer-based model. The best performance of all cases was 99.17% mean average precision in the Top 3 position, using Title+Abstract as input data and all-mpnet-base-v2 as the sentence embedding model.