Named entity recognition method based on hybrid transfer learning

Abstract

The invention discloses a named entity recognition method based on hybrid transfer learning. The named entity recognition method comprises the following steps: (1) data preprocessing; (2) instance migration; (3) pre-training the model: continuously training the BiLSTM-CRF model by using the source domain data set, and selecting and reserving the parameter with the best performance; and (4) model migration. According to the named entity recognition method based on hybrid transfer learning, sample transfer and model transfer are mixed, so that the problem of insufficient samples in the field of Chinese named entity recognition is solved, and meanwhile, the problem of inconsistent data distribution is solved by constructing an adaptive layer by using the maximum mean value difference; in the experiment process, the negative migration phenomenon of the algorithm can be reduced to a certain extent by dynamically selecting hyper-parameters.

Description

technical field [0001] The invention relates to the technical field of named entity recognition, in particular to a named entity recognition method based on hybrid transfer learning. Background technique [0002] As deep learning becomes a new field of machine learning, many scholars try to use deep learning technology to solve the problem of named entity recognition. Although the named entity recognition method based on deep learning has achieved good results, it is very difficult to obtain enough training data in practical applications. The lack of training data will lead to poor learning effect of deep learning. [0003] Migration learning can transfer and refer the learned knowledge to new problems. The purpose is to use the knowledge learned from a large amount of data to improve the performance of the target task. It has become a solution to the problem of small data sets. important method. Traditional transfer learning methods are divided into three categories: inst...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a named entity recognition method based on hybrid transfer learning to solve the problem that the learning effect of deep learning is not good due to the lack of training data, and negative transfer will occur to a certain extent when using traditional transfer learning Phenomenon, it cannot adapt well to changes in the text domain, and there will be problems with inconsistent distribution of training sets and test sets in the data set

Images