Description
The proliferation of social media data presents an immense opportunity to extract useful business insights. The continuous flow of user-generated reviews, comments, recommendations, ratings, and feedbacks represents a co-mingled dataset of facts and opinions. Detecting and separating facts from opinions in social media will significantly improve subsequent opinion mining tasks. We present an algorithm that automatically separates facts from opinions in a social media corpus. We test our algorithm using Twitter data. The algorithm analyzes not only the actual text of the posts, but also the contextual metadata and supporting reference datasets. Our approach yielded an accuracy of 73.47% in classifying facts and opinion, compared to 52.55% accuracy of the baseline models. To further demonstrate its usefulness, we applied our algorithm in an external opinion mining application that leverages social media to track customer complaints. Results show that by integrating our algorithm, the application could achieve a 10% improvement in performance.
Recommended Citation
Chatterjee, Swayambhu; Deng, Shuyuan; and Liu, Jun, "Improving Opinion Mining by Classifying Facts and Opinions in Twitter" (2017). AMCIS 2017 Proceedings. 17.
https://aisel.aisnet.org/amcis2017/SocialComputing/Presentations/17
Improving Opinion Mining by Classifying Facts and Opinions in Twitter
The proliferation of social media data presents an immense opportunity to extract useful business insights. The continuous flow of user-generated reviews, comments, recommendations, ratings, and feedbacks represents a co-mingled dataset of facts and opinions. Detecting and separating facts from opinions in social media will significantly improve subsequent opinion mining tasks. We present an algorithm that automatically separates facts from opinions in a social media corpus. We test our algorithm using Twitter data. The algorithm analyzes not only the actual text of the posts, but also the contextual metadata and supporting reference datasets. Our approach yielded an accuracy of 73.47% in classifying facts and opinion, compared to 52.55% accuracy of the baseline models. To further demonstrate its usefulness, we applied our algorithm in an external opinion mining application that leverages social media to track customer complaints. Results show that by integrating our algorithm, the application could achieve a 10% improvement in performance.