Review tree: An unsupervised method to autogenerate visual summary of online reviews

摘要

in an online review and rating portal, summarization of the text review is important as it provides a detailed glimpse of what is working and what is not. Usual techniques of text summarization do not work well in this domain due to unstructured nature of such reviews. In the case when the reviews are about services that do not have well-defined entity and aspects, doing aspect based opinion analysis is also a challenge. This paper is about summarization of online review in such domains. Additionally, extractive summarization sometimes fails to represent all the important topics due to its constraint of using sentences as information units. Such summaries also do not depict linkages between topics and building that is left to the reader. In our work, we have used sentiment analysis to extract the subjective part of the reviews and then adopted an unsupervised graph theoretical approach to extract the key phrases. Finally, semantic similarities between these key phrases are calculated to discover the interlinkages and a tree like review summary is generated using standard graph theory techniques.