Graph-based Key Opinion Analyzer: Multifaceted Evaluation of Critical Opinions and Omission Detection in Summarization
We introduce the Graph-based Key Opinion Analyzer (GKOA), an innovative opinion mining framework that extracts key opinions from online discourse, such as product reviews or social media posts.By constructing opinion graphs using LLMs, GKOA enables the quantitative and multifaceted evaluation of opinion importance through network analysis techniques, such as centrality measures.Additionally, GKOA facilitates the detection and incorporation of overlooked opinions into refined summaries, promoting inclusivity and balance in high-stakes applications (e.g., broadcasting and policymaking).We evaluated the quality of the opinion graph in the Key Point Analysis setting, demonstrating superiority over baseline models.GKOA is available on GitHub 1 with an introduction video 2 .
Authors
- Tomoki Fukuma (ORCID: https://orcid.org/0000-0002-1489-9868)
- Kyosuke Kambe (ORCID: https://orcid.org/0009-0007-9451-721X)
- Koki Noda (ORCID: https://orcid.org/0009-0009-7937-805X)
- Fujio Toriumi (ORCID: https://orcid.org/0000-0003-3866-4956)
- Yoshiharu Ichikawa (ORCID: https://orcid.org/0009-0000-8035-9141)
- Kousuke Hosoi
- Yu Masubuchi (ORCID: https://orcid.org/0009-0002-2872-4847)
- Toshihide Ubukata (ORCID: https://orcid.org/0009-0004-1033-3383)
Institutions
- Japan Broadcasting Corporation (Japan) (JP)
- The University of Tokyo (JP)
Publication Details
- Published
- 2025-05-08
- DOI
- https://doi.org/10.1145/3701716.3715179
- Primary Topic
- Advanced Text Analysis Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00