Análise de sentimento baseada em aspectos usando aprendizado profundo: uma proposta aplicada à língua portuguesa

Abstract

he increase in Internet access and the popularization of User Generated Content systems make a large number of opinionated texts became publicly available. Lots of valuable information can be found in this kind of text. Individuals and entities (private or governmental) can take advantage of that information. However, it is a lot of data to be manually processed, extracting relevant information from these sources is challenging. Sentiment Analysis is the Computer Science area focused on extracting people‘s opinions towards entities, like companies or products. However, despite the growth of this area year by year, there is a recognized lack of works developed for Portuguese. This work aims to help fill this gap. We propose an approach using Character Level Convolutional Neural Networks to solve Aspect Based Sentiment Analysis (ABSA) tasks in Portuguese texts. ABSA is a type of approach that analyzes sentiment expressed at a very specific level, seeking to explain which parts or properties (aspects) of the entities are arousing the sentiment presented in the text. In this work, we created a corpus of hotel reviews with aspect-level annotations. Based on this corpus, we developed a new approach for ABSA. Obtained results demonstrate that the proposed method has superior performance than the state-of-the-art method for Portuguese. We also compared our results with an attention-based LSTM method, with state-of-the-art performance for English.