Beyond Sentiment Analysis: Mining Defects and Improvements from Customer Feedback

Abstract

Customer satisfaction is considered as one the key performance indicators within businesses. In the current competitive marketplace where businesses compete for customers, managing customer satisfaction is very essential. One of the important sources of customer feedback is product reviews. Sentiment analysis on customer reviews has been a very hot topic in the last decade. While early works were mainly focused on identifying the positiveness and negativeness of reviews, later research tries to extract more detailed information by estimating the sentiment score of each product aspect/feature. In this work, we go beyond sentiment analysis by extracting actionable information from customer feedback. We call a piece of information actionable (in the sense of customer satisfaction) if the business can use it to improve its product. We propose a technique to automatically extract defects (problem/issue/bug reports) and improvements (modification/upgrade/enhancement requests) from customer feedback. We also propose a method for summarizing extracted defects and improvements. Experimental results showed that without any manual annotation cost, the proposed semi-supervised technique can achieve comparable accuracy to a fully supervised model in identifying defects and improvements.