Mining E-Commerce Feedback Comments for Dimension Rating Profiles

摘要

Opinion mining on regular documents like movie reviews and product reviews has been intensively studied. In this paper we focus on opinion mining on short e-commerce feedback comments. We aim to compute a comprehensive rating profile for sellers comprising of dimension ratings and weights. We propose an algorithm to mine feedback comments for dimension ratings, combining opinion mining and dependency relation analysis, a recent development in natural language processing. We formulate the problem of computing dimension weights from ratings as a factor analytic problem and propose an effective solution based on matrix factorisation. Extensive experiments on eBay and Amazon data demonstrate that our proposed algorithms can achieve accuracies of 93.1% and 89.64% respectively for identifying dimensions and ratings in feedback comments, and the weights computed can accurately reflect the amount of feedback for dimensions.