Intention recognition model training method, intention recognition method, and related apparatus

Abstract

The invention provides an intention recognition model training method, an intention recognition method and a related device. The intention recognition model training method comprises the following steps: performing preliminary training on a convolutional neural network by utilizing labeled training sample data; performing data enhancement on the labeled training sample data by adopting a conditional variation auto-encoder to obtain labeled extended sample data; and retraining the preliminarily trained convolutional neural network by using the labeled extended sample data to obtain an intentionrecognition model. Through carrying out data enhancement on training sample data with labels by adopting a conditional variation auto-encoder to obtain expanded sample data with labels; and then, theconvolutional neural network subjected to preliminary training is retrained by utilizing the extension sample data with the labels, so that the model precision can be improved and the labor cost canbe reduced on the basis of utilizing a small amount of accurately labeled training corpora.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to an intention recognition model training method, an intention recognition method and related devices. Background technique [0002] In recent years, data resources have shown explosive growth, which contains various needs expressed by users through natural language. These needs are "subjective" within a certain range. For example, users query through Google, Baidu and other search engines The demand for information, the query information that people need to know from the Internet search is recorded in the query log, which can be called the query intention; under the control of a certain purchase motivation, the user expresses the purchase intention of the product or service through the text content, It can be called consumption intention; consumers are willing to use part of their daily expenses to purchase investment and wealth management products, and the...

AI Technical Summary

Problems solved by technology

[0006] At present, researchers usually use the method of pattern matching, for example, based on the results of dependency parsing to construct templates to identify explicit intentions, the method of pattern matching determines whether a sentence has intentions by mining whether there are intention templates with explicit intentions in the sentence. Non-colloquial text has a high recognition accuracy rate, and the recognition accuracy for informal text is poor, and the customer's intention is often expressed in colloquial form

[0007] In order to solve the above technical problems, the prior art provides a guided classification method to identify customer intentions. Specifically, templates, n-grams, etc. are used as classification features, and other features are used to train a classifier to complete the explicit sentence classification. Intent recognition, but this method relies on a large amount of accurately labeled training corpus, and manual labeling corpus requires a lot of labor costs

Images