Relation extraction method and device for entities in unstructured sequence text

The invention provides a relation extraction method and device for entities in an unstructured sequence text, and the method comprises the following steps: (1) carrying out character-level embedding on an input text, and generating a unique vector representation for each character; (2) randomly selecting two different entities from the entity set, dividing the sequence text into five regions by taking the entity pair as a boundary, and generating corresponding feature vectors; (3) constructing an entity relationship feature map, and carrying out joint modeling on local features of each region and structural features between the regions based on an attention mechanism and a map convolution operation; (4) the entity features and the edge features in the feature map are combined to judge the relation categories existing between the entity pairs; and (5) repeating the steps (2)-(4), traversing all the entity pairs to complete the extraction target of the relationship information between the entities in the text,