Research On Automatic Classification Of Customer Demand Information Based On Deep Learning

With the continuous innovation and development of science and technology,science and technology sharing has become more and more important,but it is difficult for users of science and technology resource sharing service platforms to have a comprehensive understanding of it,and it is difficult to accurately match the corresponding science and technology resources according to their own needs.At the same time,scientific and technological resources cannot be well fed back to users,which hinders the sharing of scientific and technological resources and maximization of information value to a certain extent.Accurate classification and identification of user demand information can help users accurately locate their needs in the professional category,so as to quickly find the various resources they need in this professional category,making technology sharing more intelligent and efficient.Therefore,the automatic classification and processing of user demand information is an urgent problem to be solved in scientific and technological services.In response to the above problems,the multi-level structure classification framework is combined with the high-precision SC-MSNN classification algorithm to achieve the final multi-level large-scale classification task with high precision,finally,the automatic classification of user demand information is realized,so that users can accurately and efficiently match the professional subject category resources they need.The main research work is as follows:(1)SC-MSNN classifier design.In order to realize the automatic classification task of user demand information,the demand information text data is converted into dense word vectors through preprocessing such as word segmentation,stop words removal,and word vector generation,and then through the fusion of hollow convolutional structure and residual network,Weight normalization,redundancy mechanism and other technologies designed the SC-MSNN classifier,so as to achieve high-precision demand information classification task.Then through a series of experiments to verify the effectiveness of the classifier SC-MSNN algorithm,but also through the public news corpus data set to verify the generality of the algorithm.(2)The hierarchical classification framework design based on the combination of multiple SC-MSNN classifiers.First,through analysis,it is concluded that only a single classifier cannot ensure the classification accuracy in large-scale category classification tasks.Then,the advantages and disadvantages of different hierarchical classification strategies are compared,and the local classification strategy based on the parent node is determined.Based on the basic framework of,and then in view of the problems it faces,technical means based on the common feature layer of sibling nodes,cross classification strategy,classification result fusion layer and other technical means are proposed to optimize the performance of the framework.Then a series of experiments were compared to verify the effectiveness of the hierarchical classification framework based on the combination of multiple SC-MSNN classifiers,and the generality of the framework was also verified through the public news corpus data set.(3)Combine the hierarchical classification framework based on the combination of multiple SC-MSNN classifiers with the information system,design and implement the Beibu Gulf user demand information intelligent classification system,and empower the information system.