Unknown intention recognition method and device, equipment and storage medium

Abstract

The invention discloses an unknown intention recognition method and device, equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the steps of obtaining an intention unknown sample when an intention recognition model is called to fail in intention recognition of a to-be-recognized sample; performing anomaly detection on the semantic representation vector of the intention unknown sample to obtain a local outlier factor density in the sample; in response to the fact that the density is larger than a density threshold value, determining that an unknown intention exists in an intention unknown sample, and the intention unknown sample being used for training the intention recognition model again after a new intention is manually marked. According to the method, an intention unknown sample is confirmed again through anomaly detection of a local outlier factor, and a sample in which an unknown intention really exists is screened out, so that extension training is performed on an intention recognition type of an intention recognition model; especially for an intelligent customer service scene, the requirements that new user intentions emerge endlessly, the new user intentions need to be continuously mined, and the types of the user intentions capable of being recognized by the intelligent customer service robot are increased can be met.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, and in particular to an identification method, device, equipment and storage medium for unknown intentions. Background technique [0002] With the development of deep learning, classic learning tasks such as image classification and intent recognition have made breakthroughs in performance and accuracy. [0003] As one of the most basic supervised learning tasks in deep learning, classification tasks often rely on a large amount of training data. First, the training data set is manually labeled, that is, the data set is marked according to the given intention, and the intention of each training sample is marked. However, it is difficult to provide a relatively comprehensive set of labeled samples of intentions at the initial stage of the intention recognition model, which inevitably leads to some intention labels being ignored. Therefore, an unknown intent discovery me...

AI Technical Summary

Problems solved by technology

[0004] However, such methods have a great possibility of identifying noise samples as unknown intentions, resulting in a low accuracy rate for determining unknown intentions.

Images