Sumarizador de avaliações usando textrank e modelagem de tópicos

Abstract

Over the past decade, the Internet has changed the way people work, shop and socialize. Those changes resulted in the increase of User Generated Content (UGC) such as: ratings, reviews, wikis, and videos. UCG contains relevant information for decision-making, especially with regard to the acquisition of goods and services. However, the large volume and dispersion of this content makes it difficult to obtain relevant information. Text summarization appears as a way to make this content more accessible to people. A summary A can be considered better than another B when A is shorter than B while maintaining the same content relevance, or when A, despite being longer, presents more relevant content. Analyzing the literature, we observed that it is possible to produce better quality summaries than those produced by algorithms that correspond to the state of the art in text summarization. We present a multilingual automatic text summarizer that combines and extends the algorithms Latent Dirichlet Allocation (LDA) and TextRank. Our approach, when compared to the state of the art, generates better text summaries in terms of size and content relevance.