Text relation automatic labeling method based on similar glossary

The invention provides a text relation automatic labeling method based on similar glossaries. The method comprises the following steps: determining all extracted relation names and corresponding entity types, and obtaining a corpus containing the relation names; generating a similar glossary capable of replacing the relation name in each place where the relation name appears in the corpus, and sorting out a relation type glossary according to the similar glossaries; performing named body recognition and entity relation triple extraction on sentences; judging whether entity pairs of entity relation triples meet conditions or not according to results of the named body recognition, and generating a candidate relation set; and generating a similar glossary for each vocabulary of relation phrases in the entity relation triples, judging candidate relations expressed by the vocabularies in combination with the relation type glossary, taking the candidate relation which is expressed most frequently in all the vocabularies as a labeling relation, and completing automatic labeling. According to the method, the problems that a traditional remote supervision method is high in knowledge base construction difficulty and low in corpus labeling quality are solved, and a new strategy is provided for constructing a data set needed by a relation extraction model.