EMNLP2024! CoXQL - A Dataset for Parsing Explanation Requests in Conversational XAI Systems
Exciting News from hashtag#EMNLP2024! 🚀 I’m delighted to share that Qianli Wang from the XplaiNLP research group Technische Universität Berlin and Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI) presented our latest work at EMNLP 2024! 🎤
His paper, titled “CoXQL: A Dataset for Parsing Explanation Requests in Conversational XAI Systems” (https://lnkd.in/eFMpJpjT), dives into the challenges of building conversational explainable AI (ConvXAI) systems that accurately interpret user requests for model explanations. This research is a significant step forward in enhancing user trust and comprehension by improving the reliability of intent recognition in ConvXAI applications.
🔍 Key Insights from Our Paper:
- We introduce CoXQL, the first dataset tailored for user intent recognition in ConvXAI, covering 31 distinct explanation intents, including complex operations requiring multi-slot parsing.
- Our novel parsing approach, Multi-prompt Parsing with template checks (MP+), outperforms existing methods, doubling accuracy in several evaluation scenarios.
- Despite advancements, recognizing intents with multiple slots remains a significant challenge for large language models (LLMs), signaling important directions for future research.
It’s been a great experience discussing this work with fellow researchers and practitioners at EMNLP, and Qianli Wang is excited to see how our findings can contribute to the broader NLP and AI community.
A big thank you to the co-authors Nils Feldhus, Tatiana Anikina, Simon Ostermann and everyone who supported this research—your feedback and collaboration have been valuable!
đź“„ Read the full paper here: https://lnkd.in/eFMpJpjT
