Extracting implicit customer viewpoints from product reviews

Abstract

In online merchant sites such as Amazon.com, online customers share their opinions on productsin a form of product reviews. Such opinions from customers are important since they greatly affectsthe others` purchase decision. Sentiment analysis is a field of study that aims to extract and analyzeopinions contained in text. Although the level of analysis has been developed up to aspect level (i.e.aspect-level sentiment analysis ), current techniques have limitation in that they focused on recognizingthe polarity (i.e., positivity, negativity) of opinions. In this paper, we try to extract viewpoints thatpeople have in mind when they express opinions toward targets (i.e., entities or aspects). If we couldextract such viewpoints, we could gain much finer-grained insight on how people think of the targets. Wetry to extract viewpoints implied in adjectives, which is the most frequently used part-of-speech whenexpressing opinions. We pay attention to the sole function of an adjective, which is to assign a value (e.g.heavy) to an attribute (e.g. weight) of the noun that it modifies. We hypothesize that such attributeof an adjective coincide with the viewpoint of the person who used the adjective to evaluate the target.In order to extract an attribute of an adjective, we use WordNet, which maintains such relation (i.e.,attributes relation ) between adjective synsets (adjective) and noun synsets (attribute). We asked humanannotators to see if the attribute extracted from an adjective actually coincide with the viewpoint onecan infer from opinion expression. The result shows that the task is promising.