Deep relation extraction method fusing theme knowledge in government government field

Abstract

The invention provides a deep relationship extraction method fusing theme knowledge in the government government field, which is used for extracting a relationship label matched with an entity pair for a sentence containing the entity pair based on the theme knowledge of the government field, and is characterized by comprising the following steps: S1, constructing a relation matching model used for judging the matching degree between the sentences and the relation labels; s2, extracting theme knowledge corresponding to the relation label through a preset theme model; s3, respectively inputting the sentences, the relation labels and the corresponding theme knowledge into a relation matching model so as to obtain matching scores between the sentences and the relation labels; and S4, matching the sentences with the corresponding relation labels based on the matching scores so as to complete relation extraction of the sentences.

Description

technical field [0001] The invention belongs to the field of knowledge base construction, relates to the construction of government affairs knowledge base, and in particular relates to a construction method integrating theme models and deep learning in the government field Background technique [0002] In the government data governance scenario, the construction of a knowledge base can effectively help applications such as data governance and association. Relation extraction in the government domain is one of the important tasks of knowledge base construction, which aims to identify the semantic relationship between entity pairs from unstructured text. Early relationship extraction mainly relied on manual labeling of training data, and then used machine learning algorithms such as support vector machines to achieve relationship classification. Supervised relations often face the problem of data sparsity and are difficult to scale to large-scale extraction tasks. In 2009, M...

AI Technical Summary

Problems solved by technology

These methods not only ignore the rich semantics implied by the relationship, but also do not consider label semantics as an additional supervision for the relationship extraction task, so it is difficult to really extract the correct or more matching relationship labels for each sentence.

Images