Abstract
We present a new method based on episodic Knowledge Graphs (eKGs) for evaluating (multimodal) conversational agents in open domains. This graph is generated by interpreting raw signals during conversation and is able to capture the accumulation of knowledge over time. We apply structural and semantic analysis of the resulting graphs and translate the properties into qualitative measures. We compare these measures with existing automatic and manual evaluation metrics commonly used for conversational agents. Our results show that our Knowledge-Graph-based evaluation provides more qualitative insights into interaction and the agent’s behavior.- Anthology ID:
- 2022.ccgpk-1.3
- Volume:
- Proceedings of the 1st Workshop on Customized Chat Grounding Persona and Knowledge
- Month:
- October
- Year:
- 2022
- Address:
- Gyeongju, Republic of Korea
- Editors:
- Heuiseok Lim, Seungryong Kim, Yeonsoo Lee, Steve Lin, Paul Hongsuck Seo, Yumin Suh, Yoonna Jang, Jungwoo Lim, Yuna Hur, Suhyune Son
- Venue:
- CCGPK
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 15–28
- Language:
- URL:
- https://aclanthology.org/2022.ccgpk-1.3
- DOI:
- Cite (ACL):
- Selene Baez Santamaria, Piek Vossen, and Thomas Baier. 2022. Evaluating Agent Interactions Through Episodic Knowledge Graphs. In Proceedings of the 1st Workshop on Customized Chat Grounding Persona and Knowledge, pages 15–28, Gyeongju, Republic of Korea. Association for Computational Linguistics.
- Cite (Informal):
- Evaluating Agent Interactions Through Episodic Knowledge Graphs (Baez Santamaria et al., CCGPK 2022)
- PDF:
- https://preview.aclanthology.org/naacl-24-ws-corrections/2022.ccgpk-1.3.pdf
- Code
- selbaez/evaluating-conversations-as-ekg
- Data
- Topical-Chat