Information processing device, its man-machine dialogue method and storage medium

Abstract

The present invention provides an information processing apparatus capable of performing man-machine conversation processing, a man-machine conversation method and a storage medium thereof. The information processing apparatus includes an input information receiving unit that receives language information input by a user as input information; a feedback information and feature information obtaining unit that, in response to receiving the input information, invokes a domain service in the information processing apparatus, and generates feedback information of the input information in each candidate domain and feature information of each candidate domain for each candidate domain related to the invoked domain service; and an output domain specifying unit that sorts the candidate domains based on the feature information of the candidate domains extracted by the feedback information and the feature information acquiring unit, and determines the candidate domains with the best ranking as the output domains, wherein the feedback information includes feedback language information and / or feedback action information.

Description

technical field [0001] The invention relates to an information processing device capable of man-machine dialogue processing, the information processing device, its man-machine dialogue method and a storage medium. Background technique [0002] Human-computer dialogue technology is a research direction in the field of artificial intelligence. Generally speaking, human-computer dialogue is to enable users to interact with computers through human language (that is, natural language). Generally, human-computer dialogue can be divided into the following four categories according to the interaction mode: open-domain chat, task-driven dialogue (multiple rounds), question answering (single round), and recommendation. [0003] Among them, task-driven multi-round dialogue means that the user enters information with a clear purpose in order to expect information or services that meet specific constraints, such as: ordering meals, ordering tickets, looking for music, movies, or certain ...

AI Technical Summary

Problems solved by technology

However, when the semantics are ambiguous or the input semantics processed by the domain itself are relatively close, the traditional human-computer interaction system cannot accurately select the domain, resulting in the inability to obtain an appropriate reply.

In addition, when a new domain or skill is added, the model training needs to be re-trained for the intent classifier to increase the intent classifier of the new domain / to increase the expert system corresponding to the new domain, resulting in poor scalability of the system

Images