Grouping Product Aspects from Short Texts Using Multiple Classifiers

摘要

In this paper we present and evaluate a classification model to group product aspects from short user comments, found as pros and cons in consumer review websites. Because of the distinct vocabulary used by consumers to describe the same aspects of a product, it is necessary to group pros and cons to support consumers’ decision making. For this purpose we propose a supervised classification model, consisting of an ensemble classifier that combines a main text classifier (e.g. Naive Bayes) and several string-based classifiers. Furthermore we make use of WordNet as a domain independent ontology to detect semantically related words. Experimental results using pros and cons from five heterogeneous product groups show, that the proposed method outperforms existing approaches to group pros and cons from short texts. We also found that the reusable short comments from our sample follow a power law distribution, that is usually present in social tagging systems.