Graph-based Semi-Supervised Classification for Online Customer Reviews Using Consensus Clustering
Kenjiro Torizuka, Fumiaki Saıtoh, Syohei Ishizu · 2019
The purpose of this research is to present a graph-based semi-supervised learning (GBSSL) method with high classification accuracy for handling a large number of customer reviews. Following recent developments in information and communication technology, it has become essential for companies to employ efficient methods for analyzing customer reviews for improving their products and services. In analyzing these reviews, it is necessary that they are classified based on the review content. However there are only a few labelled customer reviews that can be used for the purpose of classification. Semi-supervised learning is effective in such case. In this research, we introduce GBSSL and show how it improves classification accuracy with the use of unsupervised clustering. The proposed learning method defines the result of consensus clustering based on the similarity between nodes. This research classifies customer reviews for a beauty salon using the proposed method and compares its accuracy with that of other machine learning methods to demonstrate the former's effectiveness.