Design And Implementation Of Causal Reasoning Graph Construction Algorithm Based On Deep Learning

Events are one of the core features of human society,driving the activities of human society.However,current research objects such as knowledge graphs have not taken events as their main research object.Among the relationships such as sequence,causality,and conditions between events,causal relationships can more profoundly reveal the interactions and influences between events,and are particularly important for depicting the context of event development.However,mining related events from complex unstructured text and constructing causal relationships between events is a challenge in the construction of causal reasoning graphs.In response to this,this project has carried out work related to the construction of deep learning-based causal reasoning graph algorithms,mainly including the following four points:1.A generative event extraction algorithm PET,which incorporates prompt information,is proposed and implemented.To improve the model’s ability to extract and filter text features,a newinitialized T5 encoding layer(Multi-head Self-attention Layer)and a sparse attention layer are introduced after the output layer of the original encoder to enhance the model’s performance in the financial domain event extraction task.The effectiveness of the event extraction model proposed in this project is demonstrated through experimental analysis.2.A causal relationship identification model for events based on prompt learning,MPRO,is proposed and implemented.To reduce the dependency of the event causality extraction model on labeled data,this project introduces the method of prompt learning into the event causality relationship identification task for the first time.A shallow neural network is proposed to replace the verbalizer’s design for the relationship between vocabulary and labels.To verify the effectiveness of the proposed model,experimental results on multiple public datasets outperform comparison algorithms.3.According to the practical needs of the financial domain and to verify the effectiveness of the algorithm,a financial domain causal event dataset is constructed,which includes 2000 financial texts and 3956<CAUSEEVENT,TRIGGER WORD,RESULTEVENT>triplets.A financial domain causal reasoning graph automatic construction system is developed and built,implementing data management,automatic construction of financial domain causal reasoning graphs and visualization functions.At the same time,multiple algorithm components are integrated to conduct risk prediction and analysis in the financial domain and provide a user-friendly interactive operation interface.