A Novel Automatic Sentiment Summarization from Aspect-based Customer Reviews

Abstract

Online reviews play an important role in helping companies or governments to improve product quality and services. However, these reviews are increasing day by day. It is difficult to go through the amount of these reviews and to summarize the important information manually. We proposed a novel Automatic Sentiment Summarization (ASS) system. This system has two phases. The first phase is the aspect-based representation used to represent ranked knowledge on aspect opinion calculated by using frequencies, polarity, and opinion strength. The second phase is the review summary generation used to automatically produce review summary by ranking aspect based on information of the aspect. The generated summary is more coherent by applying natural language generation technique. Furthermore, the proposed ASS system allows users to add new reviews in the same domain in order to update the generated summary. The experiments used the sentiment aspect dataset benchmarks such as customer product/service reviews for Canon, Nikon, and Laptop. The generated summaries from the proposed ASS system are well performed compared with other systems extractive summarization and abstractive summarization.