An extensible field man-machine conversation system state tracking method and device

Abstract

The invention provides an extensible field man-machine conversation system state tracking method and device, and the method comprises the steps of building a state tracking model, and employing a slot-value model to determine a current conversation state; splitting the training data feature set, collecting each round of conversation, collecting feature items in the conversation, and splitting eachfeature item; preparing context information; and performing model training based on a machine learning method. According to the method, each feature is independently predicted by the state tracking model without limiting the feature value range, and the dynamic change of the feature value range can be supported, so that the language understanding model does not need to be trained again in the business upgrading process, and the expansibility of the field man-machine conversation system is stronger.

Description

technical field [0001] The present invention relates to the field of artificial intelligence, specifically, to the field of man-machine dialogue systems. Background technique [0002] With the development of machine learning, especially speech recognition and understanding technology, many artificial dialogue systems are gradually replaced by human-machine dialogue systems. The existing human-computer dialogue systems are mainly divided into two categories, one is Xiaodu smart speakers and the like, based on the chat mode system, this type of system achieves the purpose of normal dialogue with people by learning the daily dialogue of human beings; The other category is a system similar to voice customer service, which achieves certain types of business services for users through man-machine dialogue. The design purposes of the two types of human-computer dialogue systems are different, the models used and the constraints they face are also different. The human-computer dia...

AI Technical Summary

Problems solved by technology

Since the use of feature values ​​in state tracking in existing domain human-computer dialogue systems is jointly used and interacts with each other, in order to cope with the change of feature values ​​in state tracking, existing domain human-computer dialogue systems need to The language understanding model needs to be retrained, which will obviously cause complexity in system upgrades and poor scalability

Images