Convolutional neural network relation classification method combining forward and reverse instances

Abstract

The invention discloses a relation classification method based on a convolutional neural network and relates to the relation extraction and classification technology field, wherein a forward instance and a reverse instance are combined in the method. The method comprises the following steps of S1, for a sentence text entity to be classified, according to a front and back linear sequence of words in a sentence, dividing into the forward instance and the reverse instance; S2, using a CNN sentence encoder to code a forward instance entity and a reverse instance entity, and constructing a coding characteristic vector of the forward instance of the sentence and a coding characteristic vector of the reverse instance; and S3, according to the coding characteristic vector of the forward instance and the coding characteristic vector of the reverse instance, using a softmax layer to carry out relation classification and acquiring a classification result ri. Compared to the other methods, by using the method of the invention, on a CNN, forward and reverse conditions of a same entity pair are combined so as to carry out integration classification, and a final classification effect is increased.

Description

technical field [0001] The invention relates to the technical field of relation extraction and classification, in particular to a convolutional neural network-based relation classification method combined with forward examples and reverse examples. Background technique [0002] At present, the existing relation extraction technologies in the world can be mainly divided into three types: the method based on pattern matching, the method of relation extraction based on machine learning and the method of open domain information extraction. The method based on pattern matching uses the artificially constructed pattern matching relationship, which requires manual design of the pattern, which has poor transferability; the open domain information extraction method extracts a certain sentence predicate as the relationship string between the subject and the object, and then clusters the relationship string to obtain relationship, the extraction accuracy of this method is poor and it i...

AI Technical Summary

Problems solved by technology

[0006] The relationship classification methods in the prior art mainly have the following deficiencies: the feature-based relationship classification method needs artificially designed features, and the mobility is poor; the tree-kernel-based method can only obtain features by defining a kernel function, and the feature is single; while the neural network-based method Medium CNN (Convolutional Neural Network) is easy to implement and has high training efficiency and good classification effect, while other more complex methods have low training efficiency and it is difficult to achieve the same classification effect as CNN

But it still has the following problems: the entity pairs in the same sentence are input into the neural network in different order, and the classification results may be different.

", set the entity "stress" as e 1 , entity "divorce" as e 2 The result is Cause-Effect; use "stress" as e 2 , "divorce" as e 1 The result Effect-Cause should be obtained; but in the actual classification process, there are two situations where the results do not correspond

Images