A hybrid Latent Dirichlet Allocation - BERT approach for topic discovery of market places

Abstract The tourism industry is now intimately linked to the web, as It was one of the first industries to fit on digital transformation. The tourist is more autonomous in organizing and preparing his trip, and the possibilities of leaving at competitive prices are numerous. New technologies have an impact on tourism and it is now necessary to think about communication and marketing accordingly. Market places in Morocco are reached by millions of tourists each year. Tourists share their experiences in social medias to encourage or discourage other tourists who are interested in visiting such places. The main goal of this work is to propose a hybrid architecture based on Latent Dirichlet Allocation (LDA) and Bidirectional Encoder Representations from Transformers (BERT) with a latent space representation technique to perform topic modeling and have as a result an idea of each tourist's review and particularly what he thinks about a specific market place in Morocco.