Research On Relation Extraction Based On Attention Recurrent Convolutional Network

With the advent of the big data era,a large number of unstructured text data containing rich semantic information has appeared.In order to cope with the challenge of massive data,relation extraction and knowledge graphs have become important research topics in the field of natural language processing.The construction of knowledge graphs is assisted by relation extraction to realize the reconstruction of massive data,which has important practical significance.The article focuses on the problems in relation extraction network,and conducts related research work.First,due to the complexity and variety of sentence structures,existing relation extraction networks are obviously insufficient in extracting sentence features.Therefore,how to fully learn the potential relationship features of sentences in the feature extraction stage is the current research focus.Second,although the method of automatically constructing a data set through distant supervision greatly reduces the cost of manual labeling,there is a problem of incorrect labeling of relationship labels,so it is of important practical significance to alleviate the influence of distant supervision mislabeling on relation extraction.In view of the above problems,this paper has conducted research on constructing new relation extraction network model,and has achieved some beneficial results.The main research content and innovation of this paper are as follows:(1)Aiming at the problem of low information utilization rate of sentence dependency tree and poor feature extraction effect in relation extraction task,an self-attention-guided gate perceptual graph convolutional network model was proposed.Firstly,a soft pruning strategy based on the attention mechanism was used to assign weights to the edges in the dependency tree through the attention mechanism,thus mining the effective information in the dependency tree and filtering the useless information at the same time.Secondly,a gate perceptual graph convolutional network structure was constructed,thus increasing the feature perception ability through the gating mechanism to obtain more robust relationship features,and combining the local and non-local dependency features in the dependency tree to further extract key information.Finally,the key information was input into the classifier,then the relationship category label was got.Experimental results indicate that,compared with the original graph convolutional network relation extraction model,the proposed model has the F1 score increased by 2.2 percentage points and 3.8 percentage points on Sem Eval2010-Task8 dataset and KBP37 dataset respectively,which makes full use of effective information,and improves the relation extraction ability of the model.(2)Aiming at the problem of incorrect labels in large-scale training data caused by distant supervision methods in relation extraction tasks,a recurrent convolutional network relation extraction model with multi-level attention is proposed.First,use the recurrent convolutional neural network to encode the sentence features,and obtain the global temporal correlation features and local correlation features in the sentence,so as to fully learn the latent semantic features in the sentence.Secondly,a segmented attention mechanism is constructed after the pooling operation,and the corresponding weight information is assigned to each part of the sentence by calculating the correlation between each part of the sentence and the relation vector,so as to obtain the attention-weighted sentence feature representation.Finally,in view of the imbalance of the wrong label data in each package,a cross-bag-level sentence attention mechanism is applied to dynamically assign weights to instances under the same relationship label,and selectively focus on valid instances to achieve effective suppression of noise labels.In general,the proposed model can not only obtain more abundant and effective sentence semantic features,but also can make full use of effective examples in large-scale data,capture more comprehensive feature information in the data,and improve the relation extraction effect of the model.(3)In order to verify the usability of the relation extraction method proposed in this paper in Chinese text,this paper builds a network model for Chinese relation extraction task based on the proposed method.First,the preprocessing operation is performed on the Chinese data set obtained by distant supervision to realize the standardization of Chinese text.Then,the bidirectional long short-term memory network is used to learn the sequence features of Chinese text,and the gated perception structure is applied to the convolutional neural network as a feature extractor to capture a more comprehensive sentence feature representation.Finally,aiming at the problem of incorrect labeling in distant supervised Chinese data,the use of cross-bag-level sentence attention is used to realize the identification and extraction of key features of Chinese text.The network model was trained on the Chinese data set and achieved good results,and it tried to deploy the trained model to specific applications.The experimental results show that the relation extraction network model designed in this paper has a good performance in Chinese character relation recognition.