End-to-end dialogue method and system integrated with external knowledge

Abstract

The invention discloses an end-to-end dialogue method and system integrated with external knowledge, wherein the method comprises the steps of obtaining the input text information; obtaining the dialogue historical information corresponding to the input text information, and encoding the dialogue historical information to generate a historical information vector; acquiring the database information, encoding the database information through a multi-head key value of an external database access module by using a memory network, and generating a database vector; and encoding the input text information, and generating a text reply according to the historical information vector and the database vector. According to the method, a dialogue system completely based on an attention mechanism is established, the model training time is shortened, the knowledge base information is extracted from multiple dimensions, and the knowledge base retrieval capability is enhanced.

Description

technical field [0001] The invention relates to the technical field of data interaction, in particular to an end-to-end dialogue method and system incorporating external knowledge. Background technique [0002] As a research hotspot of artificial intelligence, dialogue system has attracted extensive attention from academia and industry due to its potential intelligence convenience and attractive commercial value. The traditional task-based dialogue system adopts the pipeline method and consists of four modules: language understanding, state tracking, database query and language generation. The field needs to retrain the models of each module according to the new data. Similarly, when determining the error source, it is necessary to carry out cumbersome error analysis in each module, which consumes a lot of manpower. In such an environment, how to simplify the dialogue model and realize a transferable task-based dialogue system with the least manpower is the key topic of cur...

AI Technical Summary

Problems solved by technology

Although the pipeline method has mature technology and is widely used in the industry, there are problems such as interdependence of models, failure to correct errors, and labor-intensive problems; although the end-to-end method solves the problem of module dependence and labor-intensive in the pipeline method, but because the technology is not Mature, lacking the support of external knowledge and reasoning ability in practical application scenarios, affecting the quality of dialogue and the completion of tasks

[0006] Most of the end-to-end task-based dialog systems in related technologies are not supported by databases, and all use the traditional pipeline RNN method to encode the dialog process, which consumes a lot of time for training, and there is no universal, efficient, practical and accessible database dialogue system

Images