Research On Text Semantic Similarity Based On Deep Learning

With the advent of the information age,text as the most important carrier of information has shown explosive growth.How to extract effective information from massive text data has become a current research hotspot in natural language processing.In the actual application of text semantic similarity tasks,there are training samples and no training samples.For the training samples and no training samples,domestic and foreign researchers have carried out a lot of research and formed two technical routes.One is Use supervised learning to complete the text semantic similarity task in the case of training samples,and the other is to use semi-supervised learning to complete the text semantic similarity task in the absence of training samples.This paper tries on these two technical routes,and proposes a capsNet-BiGRU text semantic similarity analysis model and a WMD-LDA text semantic similarity calculation model.Among them,the capsNet-BiGRU text semantic similarity analysis model is more accurate,but it requires training samples and longer training time.The WMD-LDA text semantic similarity calculation model does not require training samples and the calculation is more efficient.The research results of the thesis are as follows:1.The capsNet-BiGRU text semantic similarity analysis model is proposed for the technical route of supervised learning.The model combines the capsule network(caps Net)with the bidirectional gated recurrent unit network(Bi GRU),and uses their respective advantages to extract text semantics in different dimensions.Features can extract the semantic features of the text more accurately.The model improves the traditional attention mechanism according to the characteristics of similarity tasks,and proposes a mutual attention mechanism to make it more suitable for processing similarity tasks.The traditional dynamic routing algorithm in the capsule network is improved,so that it can extract the local semantic features of the text more accurately.The model uses the extracted semantic features of the text to complete the similarity calculation task.2.The WMD-LDA text semantic similarity calculation model is proposed for the technical route of semi-supervised learning.The model is based on the WMD distance algorithm and the LDA model.The improved WMD distance algorithm is used to calculate the semantic differences between text keywords,and the LDA model is used to calculate the differences between the probability distributions of text topics.The model improves the WMD distance algorithm,adds a restriction,reduces the complexity of the model,makes it more computationally efficient while ensuring the accuracy of the calculation results,linearly fusing the results calculated by the two models to obtain a more comprehensive Accurate text semantic similarity.