Method and apparatus for identifying query intent based on hybrid strategy

Abstract

The embodiment of the invention provides a method and a device for identifying query intention based on a mixed strategy. The method comprises the following steps: acquiring a query sentence of a user; obtaining a first candidate query intention by using a semantic rule template according to a query statement of a user; according to the user's query statement, using the machine learning model to obtain the second candidate query intention; matching the first candidate query intent with the second candidate query intent; if the first candidate query intent and the second candidate query intentmatch successfully, outputting as semantic query intent. The first candidate query intent, the second candidate query intent, and the semantic query intent are at least one. The embodiment of the invention combines the semantic rule template and the machine learning model to carry out query intention identification, can better understand the query intention of a user's query sentence, and improvesthe semantic identification ability and the understanding effect.

Description

technical field [0001] The present invention relates to the technical field of speech recognition, in particular to a query intent recognition method and device based on a mixed strategy. Background technique [0002] At present, the recognition method for the user's semantic query intention is mainly to use a rule template or a machine learning model for recognition. However, although the use of rule templates has a good effect on identifying word slots in query sentences, it has a limited effect on the generalization of intent recognition, and cannot accurately identify the user's query intent. Although the use of machine learning models is better for intent classification and recognition, it is not ideal for word slot recognition. Moreover, the machine learning model needs hundreds of thousands of labeled samples to train a model with better recognition effect, but the average user cannot provide so much labeled data, and the cost of building a model in this way is relat...

AI Technical Summary

Problems solved by technology

However, although the use of rule templates has a better effect on identifying word slots in query sentences, it has limited effect on the generalization of intent recognition, and cannot accurately identify the user's query intent.

Although the use of machine learning models has a good effect on the recognition of intent classification, it is not ideal for the recognition of word slots.

Moreover, the machine learning model needs hundreds of thousands of labeled samples to train a model with better recognition effect, but general users cannot provide so much labeled data, and the cost of building a model in this way is relatively high

On the other hand, the user's query language is usually more colloquial, which is quite different from the language features of traditional search scenarios. If the semantic features cannot be accurately obtained, the user's query intention cannot be well recognized

Images