Research And Application Of Text Matching Based On Deep Learning Learning

The text matching problem occupies a core position in the field of natural language processing.Matching retrieval information,intelligent question and answering,machine translation,dialogue system,paraphrase recognition,etc.can be summarized as text matching problems.The essence is the degree of similarity between matching text.The traditional text matching model has the characteristics of simple structure and fast calculation speed.However,the traditional text matching model requires a lot of manpower and material resources to obtain very little useful information.This paper studies and analyzes the deep text matching model,and proposes a matching pyramid model based on residual network policy and a multi-granularity capture matching feature model to achieve the purpose of improving the accuracy of text matching.Finally,the research results are applied to the intelligence of the online exam platform.Review function.The main work of this paper has the following points:1.Introducing the residual network policy to improve the matching pyramid model.The construction of the matching pyramid model is difficult to determine its optimal network hierarchy.The shallow network structure cannot make the data get adequately trained and cannot reach the optimal level.The deep network structure makes the model training slow and easily causes network degradation problems.This paper proposes a matching pyramid model based on residual network strategy.The concept of residual network is introduced to enable the model to self-determine its optimal layer and perform identity mapping on redundant layers to ensure that the model can be trained in depth to avoid network degradation issues that increase the accuracy of text matching.The comparison of model matching effects is carried out through experiments.The experimental results show that the matching effect of the improved matching pyramid model has been significantly improved on the training data set.2.Proposing a multi-granularity capture matching features model.In order to improve the matching pyramid model considering the problem of single granularity and lack of high-level granular interactive information,a multi-granularity capture matching feature model was designed.This thinking process is modeled according to the way humans understand sentences,and a multi-granular representation of text is constructed.The matching process is divided into multiple granularities.It is no longer limited to single-grained thinking and improves the content contained in the text.Semantic information.The interaction information between texts is captured from each granularity and the captured matching features are combined to give the degree of matching between texts.The experimental results show that the matching effect of the multi-granularity capture matching feature model is better than that of the standard matching pyramid model,and the matching performance index is significantly improved.3.Using the multi-granularity capture matching feature model to solve the subjective intelligent review problem of the online exam platform.The multi-granularity capture matching feature model proposed in this paper is applied to the review module in the examination system,which is used to solve the intelligent review problems of subjective questions and provide a reference for review teachers.The experimental results show that the multi-granularity capture matching feature model has a better review effect than the traditional review model.