An entity matching method and system based on unlabeled corpus

Abstract

The present application relates to an entity matching method and system based on unlabeled corpus, wherein the method includes: obtaining several candidate entities by segmenting the target corpus, calculating the statistical information of the candidate entities, obtaining a set of seed entities, and obtaining a set of seed entities according to the following steps: Statistical information of the seed entity set and the candidate entity, judge and select the entity closest to the seed entity from the candidate entities, obtain several optimal candidate entities, add the optimal candidate entity to the seed entity set, and repeat the above judgment and selection until there is no most optimal candidate entity. The optimal candidate entity is generated. Based on the generated word vector of the optimal candidate entity and the seed entity, it is judged whether the optimal candidate entity is an entity, and the result of entity recognition is obtained. Through this application, the problems of strong dependence on labeled samples and low recognition accuracy in entity recognition are solved, and the entity recognition results of unlabeled target corpus can be obtained by using the list of domain entity words, and at the same time, the recognition of domain entity words can be achieved. The effect of expanding the list.

Description

technical field [0001] The present application relates to the field of data recognition, in particular to an entity matching method and system based on unlabeled corpus. Background technique [0002] In the application scenarios of text information extraction, due to the diverse and detailed scenarios, sample labeling becomes an important part of the text information extraction process. Industrial applications are faced with the current situation of lack of labeled samples and high cost of sample labeling. The current text information extraction Among the methods, the method based on model training requires a large number of labeled samples. Although some deep models show a trend of higher and higher accuracy and fewer and fewer labeled samples, they still need a certain amount of labeled samples to be trained. The available models cannot work until the marked samples are obtained. Such a process is equivalent to transferring the development cost to the sample labeling, and ...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present application provides an entity matching method and system based on unlabeled corpus to at least solve the problems of strong dependence on labeled samples and low recognition accuracy in related technologies

Images