A semantic approach for topic-based polarity detection: a case study in the Spanish language

In recent years, surprising amounts of news, messages, and reviews of products and services are generated in the online social media. Several efforts are being dedicated to detecting topics, as well as mining opinions in these unstructured texts. There are several approaches that compute opinion polarity, and some of them consider topics for their textual analysis. Nevertheless, discovering topics in opinions continues being challenging; due to opinions are generally short and informally written. Besides, opinions do not have a defined structure in several paragraphs; they are presented most of the time as a composition in a paragraph. In this paper, we propose a method to detect polarity in opinions by topics. Our proposal contributes to the fuzzy polarity calculation of detected topics in Spanish opinions. This method is comprised of three main steps: (1) preprocessing, (2) topic detection and (3) fuzzy polarity detection. It is important to notice the added values of this paper are: (1) the topic detection proposal based on the semantic processing when applying the clustering algorithm on opinion sentences, and (2) the evaluation of different aggregation operators for determining the opinion polarity from a fuzzy logic perspective. Aiming at assessing the quality of the resultant polarity detection by topics, we have conducted two main experiments over the Spanish corpus of opinions about Andalusian Hotels from the TripAdvisor site. The results have shown that our method is able to detect the topics correctly, as well as calculating their opinion polarities.