ConvINT: A Semi-Structured Intention Framework for Conversational Understanding

Jinggui Liang, Dung Vo, Lizi Liao

Keywords: Conversational Understanding, Weakly-supervised Generation, Large Language Models, Fine-tuning

Abstract: Understanding user intentions is critical for conversational AI, especially with the rise of large language models (LLMs) that demand a more nuanced comprehension of dialogue. Existing approaches, relying on rigid slot-value structures or unstructured representations, often miss the complexity of human intentions. In this work, we propose ConvINT, a novel semi-structured intention framework that offers a more holistic and fine-grained understanding of user intentions by organizing them into four key aspects: situation, emotion, action, and knowledge. Grounded in psychological and cognitive intention theories, ConvINT provides LLMs with a richer context for understanding user inputs while offering a semi-structured format that seamlessly integrates with prompt-based intention learning. To enable the efficient adoption of this framework, we introduce a Weakly-supervised Reinforced Generation (WeRG) method that scales ConvINT annotations across large datasets with high quality. By combining a small set of human-annotated instances with coarsely labeled data as weak supervision signals, WeRG effectively learns to generate ConvINT annotations, ensuring both scalability and precision. Experimental results demonstrate that integrating ConvINT with WeRG markedly improves LLMs’ ability to comprehend user intentions, yielding significant gains in downstream tasks such as response generation and task completion, as validated by both automatic metrics and human evaluations. These findings highlight ConvINT's potential as a comprehensive and adaptable framework for advancing intention understanding in conversational AI.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9195