Fuzzy Sentiment Analysis Using Spanish Tweets

Opinion mining and sentiment analysis are two topics with growing interest in artificial intelligence. Last years' research on these areas has evolved in increasing complexity and sophistication with tasks such as opinion retrieval, sentiment extraction, classification and summarization being carried out using text mining and natural language processing techniques. Possible strategies go from concept-based approaches (using web ontologies or semantic networks) and data mining (passing a labeled training corpus of texts to a machine learning algorithm), to the most crude keyword spotting methods. These last approaches use sentiment lexicons defining the prior polarity of particular terms, and aggregate these scores to determine the text's polarity. Recent literature has taken down the text analysis to sentence-level, as different opinions and sentiments towards the same topic can be present across the span of a document. Twitter is a popular social media platform where the number of characters in a message is limited to 140. These messages, called tweets, can convey opinions about a range of topics. On the other hand, fuzzy sets theory is a valuable tool in situations dealing with imprecision and ambiguous or incomplete information. Its use can be of interest to tackle the assignment of polarity to small pieces of text such as the tweets. This paper describes a sentiment analysis performed to Spanish tweets following a fuzzy sets approach and using an existing Spanish lexicon. A profiling method to evaluate the quality of the resulting indicator is proposed later.