Law text multi-hop reading understanding method based on asynchronous hierarchical graph neural network

The invention provides a legal text multi-hop reading understanding method based on an asynchronous hierarchical graph neural network, and the method comprises the following steps: carrying out the fragmentation of a legal text, inputting the fragmented legal text into a pre-training model, and carrying out the coding; performing global information enhancement on the embedded vectors of the fragmented text segments through a Memory Attention module; after the problem is also sent to the coding layer, the enhanced context information and the embedding vector of the problem are subjected to bidirectional attention to obtain a context and problem bidirectional coding vector; the context and problem bidirectional coding vectors are asynchronously updated in each layer through an asynchronous hierarchical graph network multi-hop reasoning module; and obtaining answers and clue sentences of the target and a whole reasoning path through a multi-task module. According to the method, the accuracy of a machine reading