PhD Dissertation Proposal by LIANG Jinggui | Conversational Understanding in the Open World: Intention Discovery, Structuring, and Defense

ABOUT THE TALK

Conversational understanding—inferring the intentions behind user utterances—constitutes the interpretive foundation of conversational AI. However, existing research predominantly operates under a closed-world assumption, where user needs are drawn from a predefined ontology and expressed through rigid semantic representations. This assumption proves increasingly restrictive in real-world conversations, where new user needs continually emerge, complex intentions involve rich contextual and affective semantics, and malicious intentions may be strategically concealed and refined over multiple turns.

This dissertation advances conversational understanding in the open world along three interrelated directions: (1) intention discovery, which recognizes and characterizes emerging intents beyond predefined ontologies; (2) intention structuring, which represents fine-grained intentions through expressive yet consistent multi-aspect structures; and (3) intention defense, which safeguards conversational systems against malicious intentions that are concealed and adaptively refined across multi-turn interactions. These studies aim to establish more adaptive, expressive, and reliable conversational understanding for open-world interactions.

ABOUT THE SPEAKER

LIANG Jinggui is a PhD candidate in Computer Science at the School of Computing and Information Systems, Singapore Management University, supervised by Prof. LIAO Lizi. His research focuses on conversational understanding, LLM safety, and multi-agent systems.