Short Text Classification And Information Extraction Research Based On Deep Learning

The development of the Internet and the explosive growth of network information bring us more comprehensive,more timely information,while it is hard to find the information we need quickly and accurately.Information extraction can search and return to the user more accurate and concise data from the mass of information,which can meet the user's need.Text classification can reduce the choice space of information extraction and we can develop different strategies for different types of information,which is an indispensable prerequisite for information extraction.At the present stage,the full understanding of natural language syntax and semantics is the key to text classification and information extraction.The syntactic and semantic features of artificial extraction for natural language are difficult and subjective,while deep learning can be self learning,and it is feasible for understanding natural language.We utilize the idea of deep learning to learn the syntactic and semantic features of the text,and then to learn the depth features of the extracted information,reduce the difficulty of the development for artificial features,and have good objectivity.With the help of CNN model,LSTM model and the advantages of traditional syntax tree,we improve the depth model,construct the depth neural network model,and explore the text depth feature for short text classification and information extraction.Our main work is:For the short text classification,we improved the traditional convolution neural network model(CNN),propose the concept of multi granularity convolution kernel,and combine the long and short term memory neural network(LSTM)model to find a new learning model(L-MFCNN)under the advantage of the two models.The new learning model is good for word order semantic learning and depth feature mining.Experimental results show,Our method still has a good performance under the condition of not making the tedious manual feature rules.For information extraction,we use word vector to represent question sentence and candidate information sentence,use the long and short memory neural network(LSTM)to study the semantic relevance of question sentences and candidate information sentences,use dependency tree analysis to select syntactic structure features,combine surface features to construct deep neural networks,learn the internal relevance information of questions?the candidate information and candidate information.Experimental results show,The method can be used to learn the syntactic and semantic features of sentences,and has good performance of information extraction?At last,this thesis design and implement question answering system application of information extraction,and apply the proposed method to the question answering system.Experimental results show,There is no more complex syntactic and semantic features,and the question answering system has a better performance.