Modelagem de Tópicos em Textos Curtos: uma Avaliação Experimental

Annie Amorim, Nils Murrugarra-Llerena, Vítor Silva, Daniel de Oliveira, Aline Paes

Abstract: People use social networks to express opinions or interact with other people. However, information retrieval is significantly challenging in the face of the broad scope of posted topics and the informal language in posts. Thus, automatically discovering topics from the noisy and short texts posted on social networks is paramount. Given this scenario, this paper contributes with a comparative analysis of topic modeling methods, comparing them with classical probabilistic and recent neural approaches. Also, this paper contributes with a technique for labeling topics automatically, allowing a qualitative analysis of the discovered topics.