Named entity recognition method based on multistage context feature extraction

Abstract

The invention discloses a named entity recognition method based on multistage context feature extraction, and belongs to the technical field of named entity recognition methods. The technical problem to be solved is to provide the improvement of the named entity recognition method based on multistage context feature extraction. According to the technical scheme for solving the technical problem, the method comprises the following steps of: extracting character-level features through a character sequence encoder; extracting word-level features by a word sequence encoder; based on context information extracted by Bi-LSTM, extracting attention distribution between words by a self-attention mechanism, and performing attention distribution normalization by using an SOFTMAX function; calculating a contribution coefficient of each word to a sentence based on normalized attention distribution, and calculating sentence-level features by an attention mechanism; acquiring internal relations among words, sentences and documents based on the 3D CNN, and extracting document levels are extracted according to the internal relations; the invention is applied to named entity recognition.

Description

technical field [0001] The invention discloses a named entity recognition method based on multi-level context feature extraction, which belongs to the technical field of named entity recognition methods. Background technique [0002] Named Entity Recognition (NER) is to find related entities from a piece of unstructured text and mark their location and type. Traditional named entity recognition methods mainly include rule-based, unsupervised and feature-based supervised methods. Studies have shown that named entity recognition methods based on traditional machine learning have the disadvantages of heavily relying on expert features and weak model generalization ability. [0003] In recent years, deep learning has made some breakthroughs in the field of named entity recognition and obtained the latest results, because deep learning not only saves the time of feature engineering, but also learns more useful abstract expressions. Researchers use neural networks to train chara...

AI Technical Summary

Problems solved by technology

However, Bi-LSTM needs to take the memory information in the previous memory and the current word embedding as input, making it unable to extract global information

Moreover, in order to extract richer features, stacking models is also one of the effective strategies, but the problem of model degradation is also inevitable

Images