Service Annotation and Profiling by Review Analysis
Masafumi Yamamoto, Toshihiko Yamasaki, Kiyoharu Aizawa · 2016
With the increase in the number of user reviews on user review sites, useful tools for extracting good and bad points of services so that users can easily and intuitively understand the quality of the services are required. If the annotations are selected from the pre-defined list, there can always be missing keywords. Supervised annotation approaches would suffer from the same problem. In this paper, we present an unsupervised method for extracting unique aspects of services and user opinions on these aspects from plain user reviews and apply it to service annotation using Yelp's Academic Dataset. Our method is simple and easy to extend to other languages and domains. By using only the term frequency (TF), general aspects such as whether the food or service is good can be extracted. Further, what is praised particularly to the specific service can be extracted by using the term frequency and inverse document frequency (TF-IDF).