Technical aspect extraction from customer reviews based on seeded word clustering

Abstract

Online reviews are an important source of information that customers use to make more informed purchase decisions. Attribute-centric reviews, in which the author supports her opinion with comments on the technical attributes of the product, are particularly insightful because they present deeper discussions about how technical specifications can meet the expectations of customers. However, as the number of available reviews grows, it becomes increasingly cumbersome to manually locate attribute-centric reviews as they get lost within a flood of less informative reviews. We propose a word clustering approach that uses the technical specifications of products to identify technical discussions in online reviews. Each output cluster represents a technical aspect of the products and can be used to extract its related attribute-centric reviews. We evaluate our approach by modeling technical aspects for 21,846 reviews for cameras and show that our approach can extract and rank relevant technical comments.