Analysis of restaurant reviews in terms of semantic frames in order to provide an opinion profile that reflects customer satisfaction along several dimensions.
Context and challenges
This project works towards a joint perspective on how users of social media convey meanings to their peers. An algorithm for sentiment analysis that is enriched by qualitative semantic analyses is proposed. Our primary source of data is the Yelp’s Academic Dataset, a collection of restaurant reviews in the online city guide Yelp.
Objectives
Our goal is to analyze restaurant reviews in terms of semantic frames in order to provide an opinion profile that reflects customer satisfaction along several dimensions. We thus conduct fine-grained sentiment analysis that is sensitive to different product features that are expressed by customers.
Partners and funding
Funded by Cogito Foundation
Results
We developed a web prototype that allows users to search the N closest restaurants from a location and then color them from red to green according to their preferences. An evaluation performed on the Yelp’s Academic dataset highlights a recall of 52% and a precision of 67% over a manually labeled excerpt.
Valorisation
William Droz, Hatem Ghorbel, Martin Hilpert, Magdalena Punceva, Mehdy Davary. Opinion profile construction from social media. Poster in SwissText Conference, Zürich, 2016.
