Joint Constrained Learning for Causal Event-Event Relation Extraction of Brain Connectome

Abstract

Brain science research has entered the era of con-nectome. The era of connectome has brought new challenges and opportunities for brain science research. Many studies have reported the structural and functional connections of the brain, but extracting scientific evidences from them is not easy. Traditional neuroimaging text mining methods based on terms are not suitable for the complex experimental designs and results analysis of brain connectome studies. Therefore, this paper proposes a novel method for event-level neuroimaging text mining, which aims to extract causal event-event relations of brain connectome. The method uses a deep learning model that combines BiLSTM and MLP, and incorporates constraints learning to enhance the model's performance on few-shot datasets. The constraints include common sense constraints and domain constraints, which help the model to learn from prior knowledge and domain expertise. The experiments on a brain connectome article dataset show that the proposed method can effectively extract the causal event-event relations of brain connectome with low resource requirements.