A proposal of a recommendation method of reviews describing chracteristic minority opinions

Abstract

Recently, there have been many studies that analyzed the comments in online product reviews to find customers' opinions. However, the existing analysis methods based on text mining techniques usually focus on the highly frequent words in reviews. This resulted in ignoring the minor but important opinions. Therefore, in this study, we propose a method to recommend reviews describing minor but characteristic opinions. In order to take account of such opinions, we utilized ternary relationships between products, reviews and dependency relations. We assigned the values representing the importance to dependency relations, reviews and products respectively. These values were mutually recursively calculated based on the extended idea of HITS algorithm