Man-machine conversation method based on knowledge tracking and transferring

Abstract

The invention discloses a man-machine conversation method based on knowledge tracking and transferring. The man-machine conversation method comprises the following steps: 1, constructing a model with knowledge tracking and transferring functions, wherein the model adopts a coding-decoding framework based on deep learning and comprises a coding layer, a knowledge tracking transfer layer and a decoding layer. 2, according to the constructed model, using a prior-posteriori dual learning mechanism to train model parameters; and 3, after training is completed, fixing all model parameters, and then carrying out actual conversation application. According to the method disclosed by the invention, the suitability of knowledge selection is further improved, and the model is helped to generate replies with higher user experience; and secondly, an unsupervised prior-posterior dual learning mechanism not only enhances interaction between knowledge tracking and transfer and ensures that the prediction precision of the knowledge tracking and transfer is improved at the same time, but also significantly reduces the dependence of the model on manual annotation data.

Description

technical field [0001] The invention belongs to the field of intelligent man-machine dialogue, in particular to a man-machine dialogue method based on knowledge tracking and transfer. Background technique [0002] Human-computer dialogue, that is, humans can interact naturally with machines in the form of natural language (that is, human language). The intelligence level of the human-computer dialogue system can often be used to measure the development level of today's artificial intelligence technology, so building a sufficiently intelligent human-computer dialogue model is a long-term goal in the era of artificial intelligence. At present, the related products of human-computer dialogue have been gradually applied in the real life of human beings, bringing great convenience to human life. [0003] Dialogue models have many unresolved challenges, and ensuring the informativeness of the responses generated by the model is one of them. At present, many research methods incr...

AI Technical Summary

Problems solved by technology

[0005] Secondly, in terms of model training, the current mainstream methods are data-driven. Whether it is knowledge selection or reply generation, they all rely heavily on large-scale manually labeled data for supervised learning, which makes the cost of data acquisition very high.

However, few studies have explored the use of unsupervised learning methods to improve knowledge selection and reduce the dependence on labeled data.

Images