Chinese named entity recognition model based on reinforcement learning and training method thereof

Abstract

The invention relates to a Chinese named entity recognition model based on reinforcement learning and a training method thereof. The model comprises a strategy network module, a word segmentation recombination network and a named entity recognition network module. The method comprises: firstly, a strategy network specifying an action sequence; then, the word segmentation recombination network executing actions in the action sequence one by one; obtaining a phrase through a 'termination' action, the phrase being used as auxiliary input information to perform lattice-LSTM modeling to obtain a hidden state sequence, inputting the hidden state into a named entity recognition network to obtain a label sequence of sentences, and a recognition result being used as update of a delay reward guidance strategy network module. According to the method, sentences are effectively divided through reinforcement learning, modeling of redundant interference words matched in the sentences is avoided, dependence on an external dictionary and influences on long texts are effectively avoided, correct word information can be better utilized, and the Chinese named entity recognition model is better helpedto improve the recognition effect.

Description

technical field [0001] The invention relates to the field of machine learning, and more specifically, to a Chinese named entity recognition model based on reinforcement learning and a training method thereof. Background technique [0002] Named Entity Recognition (NER) is a basic task in the field of natural language processing, which refers to identifying named referents from text, which can be used for relational extraction, question answering system, syntactic analysis, machine translation and other tasks. Foreshadowing plays an important role in the process of natural language processing technology becoming practical. Generally speaking, the task of named entity recognition is to identify the names of three categories (entity, time, and number) and seven subcategories (person name, organization name, place name, time, date, currency, and percentage) in the text to be processed. entity. [0003] An existing Chinese named entity recognition model is lattice-LSTM. In addi...

AI Technical Summary

Problems solved by technology

[0004] However, just because the lattice-LSTM model uses the information of all the words in the sentence, if the word composed of adjacent words in the sentence exists in the external dictionary, it will be entered into the model as the registered word granularity information, but the word In this sentence, it is not necessarily a correct division. For example: "Nanjing Yangtze River Bridge" According to the idea of ​​this model, the entry words composed of characters will be used as input in order. The entry words mean that the word is already in the external dictionary. Included nouns, then the model will input "Nanjing", "Nanjing City", "Mayor", "Yangtze River", "Bridge" and "Yangtze River Bridge" as login words, but obviously the word "Mayor" In this sentence, it is an interfering word, and its word information has a negative impact on entity recognition

In addition, the model usually needs to independently construct an external dictionary according to the data set used in the experiment, which has a serious dependence on the external dictionary

At the same time, when the length of the text increases, the number of potential words in the sentence will also increase, which will greatly increase the complexity of the model

Images