Trigger word recognition method based on hybrid neural network and multistage attention mechanism

Abstract

The invention discloses a biomedical event trigger word recognition method based on a hybrid neural network and a multi-level attention mechanism, and belongs to the field of natural language processing. Firstly, a convolutional neural network and a bidirectional gating recurrent neural network are combined to construct a biomedical event trigger word recognition model based on a hybrid neural network, so that a large amount of labor cost consumed by excessively depending on a natural language processing tool in a feature extraction process is reduced, and the sentence semantic accuracy is improved; secondly, key information in sentences is reinforced through a word-level attention mechanism, mutual influence among elements is enhanced through a sentence-level attention mechanism, and semantic influence of context among sentences is reinforced through a chapter-level attention mechanism; experimental results show that the recognition efficiency of the cancer-related biomedical event trigger word is effectively improved.

Description

technical field [0001] The present invention relates to the field of natural language processing, especially a method that can effectively overcome long-distance dependence, explore multi-level global attention features, low training difficulty, few network parameters, fast convergence speed, high recognition accuracy, based on hybrid neural network and multiple Biomedical event trigger word recognition method based on level attention mechanism. Background technique [0002] In recent years, with the rapid development of biomedical science, the amount of experimental data and literature in the field of biomedicine has increased exponentially. Even if relevant researchers spend a lot of time carefully screening and combining search terms, it is still difficult to quickly obtain useful knowledge from a large number of documents. In this case, biomedical text mining technology came into being. Biomedical text mining, also known as biomedical natural language processing (BioNL...

AI Technical Summary

Problems solved by technology

In this case, most of the existing methods fail to fully investigate the multi-level global features in biological texts, which is not conducive to trigger word recognition and event extraction tasks

At the same time, the existing trigger recognition methods based on deep learning still face problems such as high training difficulty, many network parameters, and slow convergence speed.

[0006] To sum up, there is currently no medical event trigger word recognition method that can effectively explore multi-level global attention features, low training difficulty, few network parameters, fast convergence speed, and high recognition accuracy

Images