Knowledge element extraction method for aviation field

Abstract

The invention provides a knowledge element extraction method for the aviation field, and the method comprises the specific implementation steps: inputting structured annotation data of the aviation field into a Bert model, and outputting a feature vector of the structured annotation data; the output feature vector and the feature vector learned by the Word2Vec model are fused, and a Concat superposition step is carried out; inputting the obtained word vector into a hierarchical normalization layer to obtain a standardized word vector; decoding each obtained character vector by utilizing a high-level reinforcement learning process, and identifying a relation trigger word in the character vector according to sentences; establishing an aviation domain long entity-oriented head-tail pointer model to obtain a predicted relationship and a tail entity start-stop position sequence; and after the predicted entities are output, matching is carried out according to label information of the entities according to a proximity principle and a matching mode. The method is oriented to the aviation field, and is combined with the vector output by the Bert based on a mode of fusing the characteristics of Zheng codes, Wubi, Pinyin, strokes and the like, so that the accuracy rate and the recall rate are improved.

Description

technical field [0001] The invention relates to the technical field of open relationship extraction, in particular to a knowledge element extraction method used in the aviation field. Background technique [0002] Information extraction (IE) is an important area of ​​natural language processing (NLP) that aims to extract structural information from unstructured text. As an important task in information extraction, relation extraction focuses on relation extraction between entities. A complete relation extraction system consists of a named entity recognizer for identifying named entities (e.g. people, organizations, locations) from text, an entity linker for linking entities to existing knowledge graphs, and a relation classifier for It is used to determine the relationship between entities based on the given context. [0003] In the existing relation extraction technology, some valuable triples are extracted after annotating a relation extraction dataset for some relations...

AI Technical Summary

Problems solved by technology

[0006] Most of the existing technologies for open relationship extraction are limited by the lack of high-quality labeled data, insufficient feature extraction capabilities of the model or generated relationships, difficulty in solving one-to-many problems, and entity matching problems

However, in practical applications, high-quality labeled data usually takes a lot of time. These technical defects make the existing technology not well meet the actual needs.

Images