SentiWordNet: A High-Coverage Lexical Resource for Opinion Mining

摘要: Opinion mining (OM) is a recent subdiscipline at the crossroads of information retrieval and computational linguistics which concerned not with topic document about, but opinions it expresses. OM has rich set applications, ranging from tracking users’ about products or political candidates as expressed in online forums, to customer relationship management. In order aid extraction text, research tried automatically determine “PN-polarity” subjective terms, i.e. identify whether term that indicates presence an opinion positive negative connotation. Research on determining “SO-polarity” indeed (a term) (an objective, neutral been instead much scarcer. this paper we describe SentiWordNet, lexical resource produced by asking automated classifier Φ associate each synset s WordNet (version 2.0) triplet scores Φ(s, p) (for p ∈ P ={Positive, Negative, Objective}) describing how strongly terms contained enjoy three properties. The method used develop SentiWordNet based quantitative analysis glosses associated synsets, use resulting vectorial representations for semi-supervised classification. score derived combining results committee eight ternary classifiers, all characterized similar accuracy levels extremely different classification behaviour. We present evaluating assigned triplets publicly available benchmark. freely purposes, endowed Web-based graphical user interface.