Research And Application Of Feature Calculus In Relation Extraction

The sentence-level relation extraction task aims to identify the semantic relation between two entities in a sentence.Due to the limited number of words in the sentence,there is a problem of feature sparsity in the relation extraction task.To solve the above problem,this paper proposes a feature calculus method for the relation extraction task to fully obtain the semantic and structural features of sentences.This method uses two entities in the relation instance to group the characteristics of the sentence into different sets.Then,use set and logical operations to combine the features in the feature set to generate combined features.The main research works of this paper are as follows:(1)Relation extraction method based on feature calculus.Most of the existing work is based on the greedy method or personal experience to study how to find more features,and rarely considers the language characteristics and grammatical functions of features in sentences.Therefore,this paper combines the features according to the two entities in the sentence to capture structural and semantic information,which helps to minimize the feature sparsity problem.This paper proposes a feature calculus method to operate formal features.At the same time,a maximum entropy model was constructed,and the influence of different feature combination methods on the relation extraction performance was systematically explored on the ACE2005 Chinese data set,which proved the importance of sentence structural information to the relation extraction task.(2)Relation extraction method combining neural networks and feature calculus.The neural network model can automatically obtain the semantic feature representation of the sentence,but it cannot fully model the sentence structural information.Therefore,this paper combines the advantages of neural networks and feature calculus and proposes a method of combining the two for relation extraction for relation extraction.This method combines the combined features generated by feature calculus with the convolutional neural network and the bidirectional long short-term memory network,then the model can obtain sentence structural information.On the ACE2005 Chinese data set,it is proved that the method proposed in this paper is significantly better than other relation extraction methods based on neural networks.Finally,based on feature calculus and neural networks,a friendly interface extraction system is designed and implemented.The system extracts the relations existing in sentences and forms structured data for storage and retrieval,providing technical reference for more downstream tasks such as knowledge graphs,intelligent question and answer,etc.