A sequence labeling method, device and equipment

Abstract

The present invention relates to a sequence labeling method, device and equipment. The method includes: obtaining an embedded representation corresponding to a target sentence in a target domain; inputting the embedded representation into a language model in the target domain to obtain a sentence feature representation; according to the sentence feature representation, The pre-trained sequence labeling model parameters of each source domain and the preset source domain parameter learning algorithm can obtain the sequence labeling model parameters of the target domain; input the embedded representation into the sequence labeling model using the sequence labeling model parameters of the target field to obtain the target The hidden state information corresponding to the sentence; according to the hidden state information corresponding to the target sentence and the preset conditional random field algorithm, the labels corresponding to each word in the target sentence are obtained. Compared with the prior art, the present application realizes the transfer learning of multiple source domain parameters, obtains the most suitable sequence labeling model parameters for the target domain, and improves the accuracy of sequence labeling in the target domain.

Description

technical field [0001] The present invention relates to the technical field of natural language processing, in particular to a sequence labeling method, device and equipment. Background technique [0002] Sequence tagging is an important task in Natural Language Processing (NLP), and is widely used in text processing and other related fields, such as word segmentation tagging, part-of-speech tagging, named entity recognition tagging, dependency syntax analysis tagging, time series analysis, etc. . [0003] In traditional sequence labeling methods, it usually relies on high-quality annotation data for supervised learning. This supervised learning method can achieve effective labeling of texts. However, due to the lack of relevant annotation data in many fields, Therefore, it is difficult to realize the sequence labeling of texts in this field, and the way of labeling texts in unknown fields by learning knowledge in other fields through knowledge transfer, if there are large ...

AI Technical Summary

Problems solved by technology

[0003] In traditional sequence labeling methods, it usually relies on high-quality annotation data for supervised learning. This supervised learning method can achieve effective labeling of texts. However, due to the lack of relevant annotation data in many fields, Therefore, it is difficult to realize the sequence labeling of texts in this field, and the way of labeling texts in unknown fields by learning knowledge in other fields through knowledge transfer, if there are large differences between fields and less common information, it will also affect the transfer The effect of learning, resulting in lower accuracy of sequence labeling

Images