Research And Application Of Text Similarity Algorithm Based On Deep Learning

With the rapid development of the Internet and deep learning technology,the field of natural language processing has made unprecedented progress.Natural language inference has achieved fruitful results in the age of big data.Text similarity analysis is a basic and critical task in natural language inference,and plays an irreplaceable role in many natural language processing tasks,such as: information retrieval,automatic question and answer,machine translation,automatic summarization and intelligent customer service.We can basically solve many text-related problems by improving the accuracy of Chinese text similarity calculation in the field of natural language processing.Therefore,we have done a lot of related work and research in order to improve the accuracy of text similarity algorithm.We mainly study the Chinese text similarity algorithm based on deep learning in this thesis.We train Chinese text similarity models with word embedding and character vector as input respectively,and analyze the influence of different granularity of input on the models.We analyzed the shortcomings of the traditional Siamese LSTM model and improved the model.By introducing bi-directional LSTM into the model,we can make full use of the information of each time step of bi-directional LSTM to capture multi-dimensional semantic information of text.Moreover,we introduce attention mechanism into the model to enrich the semantic information contained in sentence coding.In order to solve the problem of polysemy and get better semantic coding of sentences,we use transfer learning technology to introduce Bert model into text similarity computing task.Based on the Bert model,we combine the convolutional neural network to learn the deep text representation of the sentence.We designed and implemented the Bert-based representation model and achieved very good results.We also implemented two interaction-based models and merge them with other models trained in this thesis.We further improve the accuracy of Chinese text similarity calculation algorithm by using multi model fusion method.We designed and implemented an intelligent customer service system based on FAQ.We apply our semantic similarity model of Chinese text to intelligent customer service system.The system mainly involves key technologies such as text processing,and deep fusion model.The intelligent customer service system mainly responds to the problems related to the restricted field automatically.The system can automatically responds with high accuracy through the system test,which further illustrates the effectiveness of the deep learning model and text similarity fusion model implemented in this thesis.