System and method for automatically expanding recognized text

Abstract

The invention provides a system for automatically expanding a recognized text. The system comprises a skill configuration module, a product management module, a skill synchronization module and a corpus generation module. The skill configuration module is responsible for creating an extension type and configuring a corresponding extension template. The product management module provides expansioncapability for a product through a product configuration skill mode, the product is configured with a plurality of skills at the same time to improve the expansion capability, and the skill synchronization module is responsible for integrating all expansion templates under all skills configured by the current product to serve as a knowledge base of the product expansion capability. The corpus generation module performs automatic extension on the corpus set imported into the product through the skill capability of the current product and the knowledge base, so that automatic extension and replacement can be realized, and the execution efficiency is effectively improved.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a system and method for automatic extension of acquaintance text. Background technique [0002] Today, with the rapid development of the field of intelligence, natural language understanding is a very important development direction. Having a good semantic understanding will give a product (such as a robot) a more intelligent image. Taking the most common customer service robot as an example, all the answers answered by the robot are based on pre-set content (except chat robots), and at the same time, the text content that the robot can understand is also manually configured. In order for the robot to recognize different questions, the configuration personnel need to give the robot enough corpus collections, and the testers also need to write enough corpus to test the correctness of the robot's replies. For example, the original demand corpus has a sentence: "How to apply for...

AI Technical Summary

Problems solved by technology

[0004] Disadvantage 1. At present, in order for the robot to recognize enough questions, the corpus will be expanded manually, and then given to the robot

However, operators have been performing synonym replacement through synonym query for a long time. As time goes by, it is easy to bring boredom to the personnel who perform the work (example of synonym replacement: where is the address of the library? The synonym of address: location , location, replace synonym query: Where is the location of the library, the program should also be able to identify)

[0006] Disadvantage 3: Although the current Chinese text can be replaced and extended by recommending synonyms, as described in Disadvantage 2, it is necessary to replace synonyms to expand the recommended synonyms, but there are cases where synonyms cannot be completely replaced.

Therefore, the main judgment finally falls to manual processing, which leads to a very high cost of corpus expansion.

[0007] Disadvantage 4. After the configuration or test personnel get the dialogue requirements, they must expand the content of the dialogue requirements. The configuration personnel are to make the robot support different question methods, and the test personnel are to verify whether the program supports fancy questions. Law

But if all the corpus is independently expanded, both parties will take a lot of corpus expansion time, but they have to expand

[0008] Disadvantage 5. The intelligence of the customer service system is reflected in its accuracy and timeliness. Therefore, each product will continue to improve and supplement new corpus or update the latest answers. In this case, it is easy to have different answers to old and new questions. Happening

However, the test found that it will lead to the need to readjust the configuration, which will increase the number of rounds of test rounds. In addition, the randomness of the answer may also appear, which will lead to missed tests by testers.

Images