Model training method, recognition method and device for named entity recognition

Abstract

The invention provides a model training method, recognition method and device for named entity recognition. The model training method includes: inputting predefined class data into the first prototype network constructed by the coding module and the classification module, and obtaining the learned Encoding module; input the predefined class data into the encoding module before and after learning respectively, and input the results to the group classifier for training; input the data of other classes into the encoding module before and after learning respectively, and input the results to the group classifier after training respectively , to obtain grouping results of other class data; input the predefined class data and other class data into the second prototype network, and train the second prototype network. The model training method, recognition method and device for named entity recognition provided by the present invention can effectively mine more undefined classes from other classes by using the weak supervision signals of predefined classes, thereby utilizing the rich Semantic information to enhance the ability of small sample named entity recognition.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a model training method, recognition method and device for named entity recognition. Background technique [0002] Named entity recognition aims to find named entities from sentences and classify them into predefined classes (Yadav and Bethard, 2019). For example "Newton is a mathematician. He was born in Lincolnshire", the named entity recognition task needs to identify "Newton" and "Lincolnshire" as named entities, and judge their types as people and places. In actual scenarios, new named entity types emerge in an endless stream, and it takes time and effort to label data for these new types. Therefore, how to quickly learn from a small number of labeled samples has attracted widespread attention. This task is also called small sample named entity recognition (Fritzler et al., 2019). [0003] Traditional named entity recognition models (e.g. LSTM+CRF (Lample et al., ...

Images