Electric power entity chain finger method based on graph attention mechanism

The invention discloses an electric power entity chain finger method based on a graph attention mechanism. The method comprises the following steps: extracting a to-be-chain-finger entity set and initial high-dimensional vector representation of each entity from an input text; obtaining a reference item entity set most similar to each entity to be linked in the electric power knowledge graph; obtaining a one-hop entity and a relation of each reference item entity in the graph to form a reference item entity connection graph; performing high-dimensional vector representation on each reference item entity by using a graph attention mechanism network; calculating a confidence coefficient score of the entity to be linked and the reference item entity by using the high-dimensional vector representation of each reference item entity and the high-order vector representation of each entity to be linked; and sorting the confidence scores, and selecting the reference item entity with the highest score as a chain refere