Identificação e desambiguação de menções a produtos em conteúdo gerado por usuários : um estudo de caso no domínio de jogos

Abstract

A very important issue for the analysis of comments posted by users in social networks is the identification of the entities that are the target of these comments. However, correctly identifying the entities mentioned in texts produced by users is a challenging task, since the same entity can be mentioned in several different ways, depending on the user and on how the mention is being made. In addition, these comments are characterized by text with low-quality writing, misspellings, grammatical errors, etc. In this work, we present a case study on the problem of identification and disambiguation of mentions to entities in user-generated content, focused on the domain of games. The choice of this domain is due to the economic and cultural importance of this type of content and also because most of the work in recent literature related to this problems focuses on the context of electronics (televisions, smartphones, etc.). As a strategy for carrying out this case study, we have developed a tool called GameSpotter, which uses methods of named entity recognition - NER and named entity disambiguation - NED to identify and disambiguate mentions to games in comments posted on a real Web forum. Therefore, we have developed two alternative NER methods and one NED method focused on the domain of games. Our experimental results showed that our NER and NED methods are effective, achieving an average precision of 0.93 and 0.83 in the recognition and disambiguation mentions of games, respectively.