Causal Inference for Knowledge Graphs and Large Language Models
Time: 2025-09-25
Published By: He Liu
Speaker(s): Danlei Gu (BICMR)
Time: 15:00-16:00 September 26, 2025
Venue: Room 77201, Jingchunyuan 78, BICMR
Causal modeling and inference are key steps for artificial intelligence to move toward cognitive intelligence. Traditional knowledge graphs emphasize semantic connections between entities and relations but struggle to distinguish correlation from causation, which limits their value in reasoning and decision-making. Meanwhile, large language models (LLMs) demonstrate strong capabilities in knowledge acquisition and semantic understanding. Yet, their reasoning largely relies on correlations and lacks causal constraints, making it difficult to support interpretable decisions. This talk first introduces the fundamentals of knowledge graphs and causal inference. It then examines the construction and reasoning methods of causal knowledge graphs, analyzing how causal information can enrich knowledge representation and path reasoning. Finally, it focuses on LLMs, discussing how causal inference and large models can be combined.
