Training method, system and device of intention recognition model and readable storage medium

Abstract

The invention discloses a training method, system and device of an intention recognition model and a readable storage medium. The method includes the steps that a basic model is trained in advance according to a source task model and source task tagged corpuses; parameters of a network layer of the basic model are imported into a public network layer of a target task model, and the parameters of the public network layer are fixed; parameters of a specific network layer of the target task model are finely adjusted through target task tagged corpuses, wherein the specific network layer is a network layer, except the public network layer, in the target task model, and the number of the source task tagged corpuses is larger than that of the target task tagged corpuses. The target task model istrained through transfer learning of the source task tagged corpuses and the source task model, high accuracy of the target task model can be obtained by only finely adjusting the target task model by tagging a small number of target task corpuses, and the training cost of the target task model is reduced; meanwhile, the training time of the target task model can be effectively shortened.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, in particular to a training method, system, device and readable storage medium for an intention recognition model. Background technique [0002] With the development of network technology, AI (Artificial Intelligence, artificial intelligence) technology has been widely used. For example, the chatbot chatbot has been used in many fields such as intelligent customer service, personal assistant, and emotional companionship. The most important technology for chatbot is intent recognition technology. Only by accurately understanding the user's intent can it serve users better. Intent recognition technology is generally based on deep learning CNN (Convolutional Neural Networks, convolutional neural network) or RNN (Recurrent Neural Network, recurrent neural network) classification algorithm, but training the intention recognition model through deep learning CNN or RNN classification algor...

AI Technical Summary

Problems solved by technology

Intent recognition technology is generally based on deep learning CNN (Convolutional Neural Networks, convolutional neural network) or RNN (Recurrent Neural Network, recurrent neural network) classification algorithm, but training the intention recognition model through deep learning CNN or RNN classification algorithm requires a large number of If a small amount of artificially labeled corpus is used to train the intent recognition model in order to save costs, it will lead to underfitting of the intent recognition model, thereby reducing the accuracy of the prediction

Images