Avaliando o dicionário em português do método de análise de sentimentos SentiStrength

Abstract

The sentiment analysis is one of the techniques used to identify this customer satisfaction in social networks. Among the most used social networks is Twitter, being used by about 319 million users. It is considered a micro-blog and allows users, in real time, to send and receive texts of up to 140 characters, known as tweets, to from their contacts. These tweets can contain various types of information such as news, complaints, opinions, and many express feelings. There are several methods of sentiment analyzing. Among these, SentiStrength is a method that uses a lexical dictionary annotated by humans and improved with the use of Machine Learning. The method has shown in some surveys a good performance in sentiment analysis with tweets in English. The method provides some other dictionaries, including a dictionary in Portuguese that has been little used in searches. In this context, this work aims to analyze the accuracy of the SentiStrength method with the Portuguese dictionary for tweets also in Portuguese. To perform the analysis of the accuracy of the method, two 1000 tweets were collected each on different themes and the results were compared with a manual analysis. After a suggestion of improvement in the dictionary, it was possible to verify an improvement in the results, arriving to the conclusion that the effectiveness of the method is directly related to the generalization of the dictionary used.