Research And Implementation Of Deep Text Matching And Sorting

With the rapid development of the Internet,the world generates about 168 million CDs of information each day,and the total storage of information is exploding.The rapid search functions in major open search engines,various professional fields and information management systems are all designed to help users quickly match and recall text information.The key to text matching is reasonable and efficient matching algorithm.Text matching is a core and common problem in the field of natural language processing.Many natural language processing tasks,such as question answering systems,dialogue systems,and partial recommendation systems,can ultimately be abstracted into text matching problems.The attention mechanism is currently widely used in text matching task.The existing attention mechanism has two problems:on the one hand,the traditional attention mechanism pays more attention to extracting the relationship between words,ignoring the meaning of these words themselves.Therefore,feature mining of words is not sufficient,the model performance will be limited.On the other hand,there are many types of existing attention mechanism calculation methods,and the complexity of applied models is generally high.If the feature selection is unreasonable,or the feature fusion method is not appropriate,the performance of the model will not improve but will decrease.Therefore,this paper proposed a new word attention mechanism and designed a novel hierarchical feature fusion model.Firstly,this paper designed a new attention mechanism—word attention mechanism,based on the characteristics of words themselves.When enhancing the features,the word attention mechanism only considers the internal meanings and hidden features behind words.It is a method that really focuses on mining the characteristics of the words themselves.We could get a new textual expression with more comprehensive information.Moreover,this method had low computational complexity and consumes less computing resources.Secondly,as feature extraction and enhancement methods increase,this paper proposed a novel layered attention fusion application model,which classified the processing granularity(word-level,sentence-level)of text information,and added the information into the text matching model differently.After information extraction and enhancement of the multi-layer attention mechanism,different levels of information are used differently.This fusion method comprehensively improved the performance of the matching and ranking model.The experimental results of our model on multiple authoritative data sets have reached the top level in the current field.Paper which including the above innovations has been accepted for publication at international conference.Finally,this paper applied the algorithm model studied to the BUPT Education Intelligent System,which proves the effective application value of our model.