Research On Text Relation Extraction Based On Deep Learning

Relation extraction serving as a key link in natural language processing and knowledge map construction has important theoretical significance and broad application prospects.With the rapid development of computing power resources and the rapid expansion of data resources in recent years,natural language processing has been greatly promoted in all aspects so there are higher requirements for the accuracy and flexibility of relation extraction.While the current research on relationship extraction involves the traditional machine learning method which has the problem of error propagation in the model.As knowledge graphs and deep learning models develops,the research methods of deep learning and neural networks have become more and more popular among researchers.So the paper proposes a convolutional neural network relationship extraction model based on the attention mechanism and ON-LSTM neural network.According to the characteristics of the relationship extraction task combined with syntactic structural features and context semantics,the relationship extraction problem is converted into a relationship classification prediction problem while introducing an external resource library as an auxiliary feature extraction to extract high-level word-level feature vectors as part of subsequent classification input.Syntactic corresponding feature extraction is to embed the feature vector of word obtaining feature vectors as well as and the embedding vector of the position relative to the entity.Then obtain vanced syntactic feature vectors through the convolutional neural network layer and the ON-LSTM neural network layer.The model combines the extracted high-level syntactic features and word-level features and finally obtains the prediction vector for relationship classification via the Softmax layer.The item with the highest element score corresponds to the result of predicting the entity pair.The paper carried out a relation extraction task experiment on the Sem Eval-2010 Task 8 dataset and compared it with traditional machine learning-based methods as well as more mainstream deep learning model methods.The results show the method proposed in the paper is more effective.