XLNet-based intelligent voice conversation intention recognition method

Abstract

The invention provides an XLNet-based intelligent voice conversation intention recognition method. The method comprises the following steps of S1, marking and sorting a standard question of a conversation node and a plurality of extended similar questions as corpora into a text sample; S2, splitting the text sample into a training set and a verification set; S3, inputting the training set into anXLNet model, initializing an original correlation weight, setting the iteration number and the step length, adding Triplet loss in a loss function, and fixing other layers of a network; S4, verifyingthe XLNet model on the verification set by the offline accuracy; S5, pre-loading the XLNet model, providing an interface for receiving a to-be-recognized voice and converting the to-be-recognized voice into text input, and outputting a classification category related to a conversation text; S6, adding corresponding threshold calculation and classification probability calculation in an online service; and S7, obtaining a text title of relevant classification through a configuration file during classification activation. According to the intention recognition method, the intention recognition rate is greatly improved; and a demo script is rewritten, so that the intention recognition rate is improved.

Description

technical field [0001] The invention relates to the field of speech recognition, in particular to an XLNet-based intelligent speech dialogue intention recognition method. Background technique [0002] With the rapid development of big data and machine computing power, deep learning technology has made many major breakthroughs in vision and speech. In the field of speech recognition, more and more intelligent voice robots are put into practical use. Whether a voice robot is intelligent or not depends on whether the intention recognition during the dialogue is accurate. The introduction of the XLNet model has refreshed the achievements of nlp technology in various data sets. However, there are still many problems in the use of XLNet technology in the field of speech recognition. [0003] First of all, based on the XLNet model, the script and model parameters in the demo are completely reused, and the demo script and model parameters are completely reused. In the actual scene,...

AI Technical Summary

Problems solved by technology

The introduction of the XLNet model has refreshed the achievements of nlp technology in various data sets. However, there are still many problems in the use of XLNet technology in the field of speech recognition.

[0003] First of all, based on the XLNet model, the script and model parameters in the demo are completely reused, and the demo script and model parameters are completely reused. In the actual scene, the model parameters are not adjusted according to the data distribution of the own, resulting in poor results; secondly, because of retraining the entire The corpus and computing resources required by the network are too large, causing the entire model to fail to converge or overfit when individuals or companies choose to retrain the entire network

Images