Phrase-Based Topic Discovery from Spanish Social Media Texts

Recently, there has been a large amount of valuable information on various topics in social networks. For example, many worldwide appreciate clinical information about suffering, signs, symptoms, procedures, drugs or treatments related to diseases. However, this information needs to be more structured since it is in natural language, which makes it difficult for readers to find respected data manually. In addition, it is a tedious, expensive and time-consuming process. Clinical information of interest, such as disease names, is usually expressed in phrases, not just words. It requires computational methods based on Natural Language Processing for a phrasal textual analysis to find topics of interest from texts. Therefore, this paper presents an approach for automatic topic discovery from social media texts in Spanish about three prevalent diseases in Mexico: COVID, Diabetes, and Cancer. This approach introduces an analysis based on phrase level as entries for LDA and BTM topic discovery algorithms. An evaluation process has been performed using the coherence topics metric and relevant phrases obtained for discovered topics. The results show encouraging 0.7463 coherence for COVID texts using noun phrases and the BTM algorithm.