| In recent years,with the fast rise of big data and social platforms,intelligent government platforms have also attracted more and more attention.In particular,the rapid expansion of the number of government texts,on the one hand,leads to the failure of various government departments to timely and accurately obtain the complaint information of Internet users,on the other hand,leads to the failure of government departments to provide all-round and comprehensive government functions to Internet users.Therefore,in the face of a large number of government information texts,how to effectively improve the efficiency of government services and optimize the working experience of Internet users has become one of the important topics in the research and development of e-government service platform.This paper first introduces the research status of multi-label text classification methods and personalized recommendation methods,and then analyzes and introduces the related work and theoretical knowledge of both methods in detail.According to the application scenarios and data characteristics in the field of government affairs,this paper further studies the multi-label text classification method and personalized recommendation method,and obtains the following research results.Aiming at the problem that various government departments cannot obtain the complaint information of Internet users in time and accurately,this paper combines convolution neural network,recurrent neural network and attention mechanism,and proposes a multi label text classification method by multi-level features.Both local and global features of government text can be extracted by our method,and finally the important information of government text is extracted by attention mechanism.Due to the lack of government text data with labels,this paper uses data expansion technology to expand government text information,which not only expands the data set,but also solves the problem of low efficiency of manual labeling.In order to enhance the accuracy of multi-label text classification,the word representation method is used to obtain the hidden features of government information text.Aiming at the problem that government departments cannot provide government functions to Internet users in an all-round and comprehensive way,this paper makes a deeper interaction between user information and government function information,and proposes a government function recommendation method based on BERT and neural network collaborative filtering.On the one hand,this method uses BERT word embedding technology to make users and government functions meet the relationship in semantic space,and then obtains negative samples according to the generated similarity threshold to solve the problem of matrix sparsity.On the other hand,the use of neural collaborative filtering technology enables users and government functions to further provide linear and nonlinear interaction in semantic space.Experiments show that the government function recommendation method based on BERT and neural network collaborative filtering not only has high recommendation performance and good robustness,but also can better learn the hidden interaction characteristics between users and government functions.Finally,using PyCharm as the development platform,this paper designs and develops a user-oriented government function recommendation system,which is based on BERT and neural network collaborative filtering.The system mainly includes several modules such as user registration and login,extracting user information,calculating similarity,obtaining negative samples and government function recommendation.Through this system,the government service function can be inversely recommended to Internet users. |
