Construction And Application Of Government Response Quality Evaluation Mode

As more and more people use the smart government platform to consult business,the types and number of consulting are growing rapidly,but the traditional method is mainly online reply manually,and the capacity of different business departments and customs in different regions is not balanced,the traditional method is difficult to form effective supervision of the reply quality.there are severe challenges in ensuring the quality and efficiency of government replies which provided by the government affairs platform.This paper focuses on the reply quality of customs business consultation based on the smart government theories.By constructing a reply quality evaluation model,we realize the supervision of government replies,so as to further improve the customs service system in the business consultation scenario.The research content and innovation points of this paper mainly include the following two aspects:1.Combined with the pre-training language models,part-of-speech tagging and syntax parse,this paper evaluates the quality of reply from five aspects: timeliness,relevance,readability,information intensity and normalization.Wherein,the feature extraction method for message and reply text constructed by the ALBERT(A Lite Bidirectional Encoder Representations from Transformers)pre-training model can more effectively solve the problem that the traditional statistical model only reflects part of information,while ignoring the inherent semantic features of the text;The index of readability which based on dependency syntactic parse and constituent syntactic parse can analyze the internal logic and semantic composition of the text from grammatical structure and recursion;By using the part-of-speech tagging algorithm based on deep learning model,the part-of-speech tagging for the reply content is carried out and the proportion of nouns in the reply text is counted.Compared with the statistics of text length only,this method can better reflect the information contained in the reply text.In addition,combined with the kernel density estimation,model comparison and case test,the rationality of single index is evaluated.2.Based on autoencoder and entropy weight method,index weight determination and comprehensive quality evaluation of reply are realized.Firstly,the autoencoder algorithm is used to reconstruct each evaluation index into potential spatial representations with fewer dimensions and most information of the indexes.Secondly,the entropy weight method is used to determine the weight of each representation and calculate the weighted score.Compared with the direct use of the entropy weight method on original indexes,the entropy weight method based on the autoencoder can avoid the phenomenon that the weight does not match the reality due to ignoring the importance of the indexes themselves.In addition,the rationality of the quality evaluation model constructed in this paper is verified by multi-angle analysis of reply quality and models comparison.