Multi-service scene dialogue method and system

Abstract

The invention relates to a multi-service scene dialogue method and system. The method specifically comprises the steps of judging whether there is a post-complement dialogue intention in a database or not after a to-be-replied statement is received; if no post-complement dialogue intention exists, inputting the to-be-replied statement into the trained BERT model for dialogue intention classification, and replying according to the obtained dialogue intention and the associated initial reply content; if the post-complementdialogue intention exists, comparing all post-complement dialogue contents with the to-be-replied statement; if the obtained similarity is not less than the preset value when the obtained similarity is the maximum, replying according to the post-complement dialogue intention associated with the post-complement dialogue content most similar to the statement to be replied and the post-complement reply content associated with the post-complement dialogue content; and if the obtained similarity is still less than the preset value when the obtained similarity is maximum, skipping to the step of inputting the BERT model. The problem of multi-round intelligent dialogues in different business scenes can be solved, quick response is realized, the defects of time consumption and response delay of BERT model training are overcome, and meanwhile, different dialogue processes share the same intention classification model.

Description

Technical field[0001]The present invention relates to the field of communication technologies, and more particularly to a multi-service scene conversation method and system.Background technique[0002]AI dialogue robots can replace automatic reply. At present, in order to implement intent, mostly use the classification algorithm, the first fixed dialogue category, then identify; the shortcomings of this method include: unable to apply to multi-service scene, flexible, if needed to increase the conversation in real time, must be used Quick response server; BERT model and dialog process is a corresponding, poor genericity, and the difficulty of the conversation process is difficult, and the operation is cumbersome; the training cost of the BERT model is high, the data volume is large, the training consumption, and the reaction is hysteresis.[0003]Therefore, there is still a need to improve the existing business scene dialogue method to solve the above deficiencies.Inventive content[0004...

AI Technical Summary

Problems solved by technology

At present, in order to realize intent recognition, classification algorithms are mostly used, that is, to fix the category of dialog intent first, and then to identify; the shortcomings of this method include: it cannot be applied to multi-service scenarios, and the flexibility is poor. If you need to increase the dialog intent in real time, you must use Fast response server; BERT model and dialogue process are one-to-one correspondence, poor versatility, and the construction of dialogue process is difficult and cumbersome to operate; BERT model has high training cost, large amount of data, time-consuming training, and slow response

Images