Relation extracting method based on convolution neural network and distance supervision

Abstract

The invention discloses a relation extracting method based on a convolution neural network and distance supervision. The method comprises the steps of (1), mapping existing few relations into required relation types; (2), expanding different expression ways of entities in the existing relations; (3), obtaining a great number of relevant unstructured texts from the internet, and establishing indexes; (4), querying sentences relevant to the entities through the indexes, and classifying the sentences into positive samples and negative samples; (5), converting the samples into feature vectors based on the convolution neural network; and (6), classifying the texts by employing the obtained feature vectors, thereby obtaining new relation pairs. According to the method, on the basis of an assumption that one sentence may have one relation, starting from few known relations, a great deal of new structured information is obtained by employing a great number of unstructured texts from the internet, and namely, the new relations are discovered.

Description

technical field [0001] The invention relates to neural network, natural language processing, information extraction and relation extraction, in particular to a relation extraction method based on convolutional neural network and distance supervision. Background technique [0002] In recent years, with the rapid development of the Internet, there are more and more content and knowledge on the Internet, even exponentially doubling, including news, blogs, emails, government documents, chat records and so on. However, these data are all unstructured electronic texts. How can humans easily understand all this data? A very good idea is to convert these unstructured data into structured semantic information. However, the huge amount of data makes it very difficult or even impossible to manually annotate this information. Therefore, it is hoped that these data can be marked into a text structure that is easy for humans to understand and read through computers and computer technol...

AI Technical Summary

Problems solved by technology

Since the supervised method needs to manually mark the training data set in advance, and this work is relatively labor-intensive, the supervised method is not suitable for large-scale open domain information extraction tasks.

[0005] Second, unsupervised methods

However, after a large number of iterations, the accuracy rate usually drops more, which is caused by the accumulation of labeling errors. This phenomenon is called the "semantic drift" problem.

Images