Study On The Application Of Deep Learning Technology In Chinese Personal Relation Extraction

With the advent of the Internet age,there are lots of text resources in the network.How to effectively use the computer to process these texts to get useful information has become a focus in the field of artificial intelligence.Relation Extraction,as a research hotspot in natural language process,has attracted the attention of researchers.Chinese personal Relation Extraction which can be regarded as a special kind of Relation Extraction has important research significance.Its results can be used in Chinese personal Knowledge map,Chinese personal Relation Question&Answer and other fields.Nowadays,the research on the Chinese personal Relation Extraction is mainly focused on traditional method of Relation Extraction,including feature-based method,kernel-based method and relation pattern based method.Traditional method of Relation Extraction has the following shortcomings.Feature-based method needs predefined features by manually and its effect depends directly on the feature selection.Kernel-based method doesn't apply to Relation Extraction which has a amount of data.It's difficult to extract Relation pattern comprehensively by relation pattern based method,which affects the effect of Relation Extraction.Deep learning Relation Extraction method makes up the deficiency of traditional method of Relation Extraction.It has the advantages of not requiring predefined features,can be applied to large data sets and good extraction effect.Therefore,how to build a Chinese personal Relation Extraction with deep learning technology is a topic worthy of study.In this paper,we study the application of deep learning technology in Chinese personal Relation Extraction in detail.The major research work of this paper are as follows:1)We propose to construct the Chinese personal Relation Extraction system with deep learning technology.In order to automatically generate training data for Chinese personal Relation Extraction task,we use distant supervision to align Chinese knowledge database "Hudong Baike" and Chinese corpus "SogouCS 2008".2)In order to combine the advantages of two mainly model in Deep learning Relation Extraction:CNN(Convolutional Neural Network)and RNN(Recurrent Neural Networks),we propose a model named LSTM_PCNN and apply it to Chinese personal Relation Extraction.The LSTM PCNN model combines the advantages that PCNN model(CNN derivative model,Piecewise Convolutional Neural Network)can extract local text features and Bi-LSTM model(RNN derivative model,Bi-directional long short term memory network)can model sequence data.Experiments show that the LSTM PCNN model is better than the Bi-LSTM model and PCNN model in Chinese character relationship extraction task.3)Based on the LSTM_PCNN model and Attention mechanism,we propose a model named LSTM PCNN_ATTE which can take the influencing factor of entity pair and context information into account and describe the effect of words in sentence to judge the relation between the persons with Attention mechanism and apply it to Chinese personal Relation Extraction.Attention mechanism force LSTM PCNN ATTE model to focus on those words that are more useful for Chinese personal Relation Extraction.Experimental results show that the LSTM PCNN_ATTE model has better performance than the LSTM_PCNN model in Chinese personal Relation Extraction.At last,we do an experiment to compare the effect of LSTM PCNN ATTE model and feature-based model on Chinese personal Relation Extraction.We have found that LSTM PCNN ATTE achieve higher precious,recall and F-value,which further proves the effectiveness of Deep learning method in the Chinese personal Relation Extraction.