Topic modeling for user feedback dataset

Abstract

In the era of big data, user feedback from mobile applications provides valuable insights for improving performance and user experience. However, extracting meaningful topics from large textual datasets remains a challenge. This study employs the Top2Vec model, a modern topic modeling technique, to analyze a dataset containing 15,000 user feedback entries from 15 different mobile applications across various categories. Unlike traditional methods like Latent Dirichlet Allocation (LDA), Top2Vec integrates word embeddings and clustering algorithms to identify topics based on semantic relationships. The research involves text preprocessing, embedding generation using Doc2Vec, and applying the Top2Vec algorithm to extract relevant topics. Results indicate that Top2Vec automatically determines topic numbers, offering richer and more interpretable topics compared to LDA and Embedded Topic Model (ETM). Evaluation metrics such as Coherence Score and Topic Diversity demonstrate that Top2Vec performs well, capturing significant patterns and addressing user concerns, including app glitches, performance issues, and user experience. This article highlights the effectiveness of Top2Vec in analyzing user feedback, making it a promising tool for understanding user needs and improving application development.