Semi-supervised topic representation through sentiment analysis and semantic networks

This paper proposes a novel approach to topic detection aimed at improving the semi-supervised clustering of customer reviews in the context of customers' services. The proposed methodology, named SeMi-supervised clustering for Assessment of Reviews using Topic and Sentiment (SMARTS) for Topic-Community Representation with Semantic Networks, combines semantic and sentiment analysis of words to derive topics related to positive and negative reviews of specific services. To achieve this, a semantic network of words is constructed based on word embedding semantic similarity to identify relationships between words used in the reviews. The resulting network is then used to derive the topics present in users' reviews, which are grouped by positive and negative sentiment based on words related to specific services. Clusters of words, obtained from the network's communities, are used to extract topics related to particular services and to improve the interpretation of users' assessments of those services. The proposed methodology is applied to tourism review data from Booking.com, and the results demonstrate the efficacy of the approach in enhancing the interpretability of the topics obtained by semi-supervised clustering. The methodology has the potential to provide valuable insights into the sentiment of customers toward tourism services, which could be utilized by service providers and decision-makers to enhance the quality of their services.