Transfer learning system and method for natural language processing tasks based on domain adaptation

Abstract

The invention discloses a transfer learning system and method based on domain-adaptive natural language processing tasks, including: an open domain partial module and a specific domain partial module, wherein the open domain partial module is used for: utilizing existing open domain data Carry out training and modify the output layer according to different natural language processing tasks, specifically including: LSTM network trains the open domain to obtain the weight matrix of the open domain neural network; some domain-specific modules are used to: construct a domain-specific neural network, according to different The natural language processing task modifies the output layer, which is consistent with the output layer of the open domain; among them, the weight matrix of the repeated part of the neural network is kept the same as the weight matrix of the trained open domain neural network, and the training of the specific domain neural network is obtained according to the results of the output layer. Results of natural language processing tasks. It addresses the problem of domain adaptation in natural language processing tasks.

Description

technical field [0001] The invention is a migration learning system and method for natural language processing tasks based on domain adaptation, and solves the problem of domain adaptation in natural language processing tasks. Background technique [0002] Natural Language Processing (Natural Language Processing) is a sub-discipline in the field of artificial intelligence and linguistics. Discuss how to process and use natural language in this field, including natural language cognition, natural language generation and natural language understanding. Natural language cognition refers to allowing computers to "understand" human language. Natural language generation systems convert computer data into natural language. Natural language understanding systems convert natural language into a form that can be more easily processed by computer programs. The main tasks of natural language processing include part-of-speech tagging, sentiment analysis, and syntactic analysis. [00...

AI Technical Summary

Problems solved by technology

However, the ambiguity of natural language and the overfitting of the model in the embedding layer will lead to poor performance of these models in the application of specific fields.

At the same time, due to the lack of training data in specific fields, it is impossible to use specific field data for training

Images