Named entity recognition model applied to science and technology documents in manufacturing industry

Abstract

The invention discloses a named entity recognition model applied to science and technology documents in the manufacturing industry, and the network structure of the model comprises a word embedding layer SciBERT which is used for converting an input word into a vector with a fixed length; the BiLSTM layer is used for mining hidden features by utilizing context information of the text sequence and is used for encoding the text; the attention layer is used for reducing weights of irrelevant modifiers in the entities, defining boundaries of the entities and avoiding extraction omission of important entities; and the CRF layer serves as an output layer of the network structure and is used for preventing the entities in the text sequence from being mistakenly labeled. The named entity recognition model can extract information and generate knowledge from a text, and can analyze various documents equivalent to manufacturing science and technology, such as product design text data, engineering test text data, supplier data, maintenance record data, product use data and the like in the manufacturing field; a technical basis can be provided for an enterprise to realize interconnection of various data assets, and the method is a key for promoting digital transformation of the enterprise.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to a named entity recognition model applied to manufacturing scientific and technical documents. Background technique [0002] With the exponential growth of manufacturing-related scientific literature and the number of digital resources available on the Internet, it is challenging to search and extract valuable information from manufacturing scientific and technical documents. Existing named entity recognition models have relevant research and applications in specific fields such as materials science, biomedicine, chemical science, network security, maintenance practice, and forensics science. However, for manufacturing scientific and technical documents, research uses named entity recognition models and extracts The valuable information is still in the embryonic stage of technical language processing research and application. [0003] The shortcomings of exist...

AI Technical Summary

Problems solved by technology

[0004] 1) Existing standard named entity recognition models cannot recognize domain-specific terms and concepts

[0005] 2) Due to the different characteristics of technical documents in different fields, the named entity recognition model developed for a specific field cannot be transplanted to the manufacturing field

[0006] 3) The existing technology center lacks research on automatic classification of manufacturing science and technology documents using the constructed named entity recognition model in the manufacturing field

Images