Topic Modeling for News Recommendations: Evaluating the Performance of LDA and BERTopic

Text analysis is an important component in the evolution of recommender systems, as it enables meaningful information to be extracted from vast amounts of textual data.  This study performs a comparative analysis of two main topic modeling techniques, Latent Dirichlet Allocation (LDA) and BERTopic in the context of news recommender systems.  Using a dataset of Moroccan news articles, we evaluate the ability of these models to generate coherent and interpretable topics.  Our results demonstrate that BERTopic outperforms LDA in terms of topic consistency and semantic richness, particularly when applied to large and diverse datasets.  Moreover, the integration of BERTopic into a content-based recommender system results in improved relevance and accuracy of recommendations.  This research highlights the potential of advanced topic modeling techniques for refining recommender systems and suggests future directions, including the exploration of hybrid models and the integration of sentiment analysis for further improvements.