Research On Natural Language Processing Technology Based On Convolution Neural Network

Natural language processing is a very important branch of computer science and artificial intelligence.It becomes an essential prerequisite for machine to communicate with people and understand human intentions.It presents itself as a science that integrates computer science,linguistics and mathematical algorithms,which also contains many sub-tasks.The main content of this thesis consists of two parts in natural language processing: the short text emotion analysis and the Chinese entity relation extraction.Short text has been widely used in many areas of life,such as evaluations of goods in shopping sites,film reviews in video sites and blogs in social networking,which generates the core value for precise analysis of the emotion in short text.Nonetheless,short text inherits the following flaws: 1)the lack of context information;2)emotional polarity ambiguity.Taking the flaws above into account,traditional model can not fulfill the features extraction merely from word level.Therefore,this paper proposes a convolutional neural network model(SWcNN)which integrates word and sentence features to analyze short text.Compared with several traditional machine learning methods on the specified dataset,results indicate that the accuracy of the emotion polarity prediction is significantly improved.Chinese entity relation extraction provides a solid foundation for numerous key tasks of Chinese processing.At present,all of the main related work is built upon the traditional machine learning methods.Seeing that Chinese possesses unique features such as flexible grammar,complex expression,etc,conventional machine learning methods can not acquire desirable results.This paper proposes a convolutional neural network model which combines word and sentence features to extract the relationship of Chinese entities,and obtains better results than the existing models on ACE2005 dataset.