Research Of Chinese Personal Relation Extraction Based On Deep Learning

With the rapid development of Internet technology,text data flooding the network has also shown an exponential growth trend,and most of these text data are stored in unstructured forms,making it impossible for humans to use it,information extraction technology can extract valuable information such as relationship information,event information,etc.from these unstructured mass texts and convert them into structured information for storage.The results can be applied to intelligent question answering system and knowledge Library building and many other areas.Relation extraction as the core task in information extraction has very important research significance and application prospects,and it has been a hot research technology in the field of NLP in recent years.With the rapid development of deep learning technology,many researchers have applied deep learning technology to a variety of NLP tasks and have achieved very good results.Deep neural networks have powerful feature learning capabilities and avoid timeconsuming manual features definitions in traditional methods,and many other advantages,this article introduces deep learning technology into the research of Chinese personal relation extraction tasks.The main research contents are as follows:(1)Aiming at the lack of a standard corpus in the current Chinese personal relationship extraction task,this paper studies the method of remotely supervising the alignment and matching of the Chinese knowledge base "Baidu Encyclopedia" with free news text to automatically generate labeled datasets.(2)Research and explore the most commonly used convolutional neural networks and recurrent neural networks in relation extraction tasks,and found that although these two different types of networks all perform well in relation extraction tasks,the two neural network models has certain advantages and limitations,so this paper proposes a personal relationship extraction model based on convolutional recurrent neural network,combining PCNN and Bidirectional GRU.It has the ability of convolutional neural network to extract local important features,and has the advantage of recurrent neural network being good at processing sequential tasks.The experimental results prove that the convolutional recurrent neural network model proposed in this paper has better personal relationship extraction effect than the single convolutional neural network model and recurrent neural network model.(3)For the problem of mislabeling in remote supervision,a sentence-level attention mechanism is introduced in the model,and the impact of noise data is reduced by assigning different attention weights to sentences with unbelievable confidence,the neural network would be able to learn more accurate and richer feature information,and further improve the accuracy of the character relationship extraction model.