Machine Learning-based Questionnaire Credibility Audit System

The household survey is the main method of national census.Every year,the state finances huge expenditures for these activities,and expects to obtain accurate survey data to grasp the real situation and make reasonable decisions to serve the national economic life.However,due to various reasons in the investigation process,there are a large number of unqualified questionnaires in the questionnaire(such as questionnaire fraud),which affects the accuracy of the survey data,which in turn affects the scientific nature of decision-making.In order to identify unqualified documents,the current common practice is to conduct manual and one-by-one questionnaire review.There are problems such as low efficiency,high cost,and high subjectivity.As the number and scope of investigation activities increase year by year,such problems are becoming more and more serious.The machine audit system automatically screens questionnaires,frees manpower,improves audit efficiency,reduces costs,and improves the objectivity of audits.In the aspect of automatic audit,the existing research is lacking,and the audit work is still in a semi-automatic state.With the development of computer performance,machine learning and big data technology are widely used in image processing,speech processing,data classification and clustering,etc.The amount of audit data is huge,the types of features are diverse,and the feature structure is unified.The method of machine learning can be used to model the credibility of the questionnaire so that the data can be classified more quickly and accurately.Therefore,this research has important practical significance,improve the efficiency and quality of auditing,and expand the application space of machine learning.To this end,in order to realize the automatic audit of the questionnaire,this paper designed a questionnaire system based on machine learning.The key of this method is to extract the characteristics of the questionnaire and model the credibility of the questionnaire through various algorithms of machine learning.The challenge is that the feature dimensions in the questionnaire are too large,and the quality of the questionnaire images is poor and the audio resolution is low.Therefore,our idea is to fully exploit the effective features in the aspects of audio,image,GPS distance,investigator credibility,etc.,and at the same time,carry out feature engineering on the high-dimensional features of the questionnaire,reduce the dimension of the feature,and finally through machine learning.The classic algorithm,training the questionnaire credibility model and predicting.The main contribution of this paper is to design and implement an automated auditing system for the questionnaire review process of the popular science institute,and to implement the algorithm using the latest model.Based on the surface of a large number of actual data(70,000 copies),the results of the automatic audit system of the questionnaire presented in this paper are generally consistent with the results of the manual review given by China Science Popularization,and the evaluation dimension of the system is more,and the score results are more For the sake of fineness,subjectivity is very small,auditing is more efficient,and auditing costs are lower.Specifically,the contributions of this article are as follows:(1)In this paper,the questionnaire survey information,from the multi-dimensional evaluation questionnaire,including audio,image,distance,respondent sequence of respondents,investigator credibility and other dimensions,fully explore the effective characteristics of the questionnaire.At the same time,the feature engineering of the above multi-dimensional features is carried out,and the correlation analysis is carried out to realize the dimensionality reduction of high-dimensional features.(2)This paper designs and implements a machine learning-based model,which fully analyzes the characteristics of the questionnaire and models the credibility of the questionnaire.This paper compares several classic machine learning models and finds models that perform optimally on existing data sets.(3)Finally,based on the above work,this paper designs an automatic auditing system for questionnaires.Through the analysis of various modules such as audio and image,the initial diagnosis results of each module are formed,and the high-dimensional features of the questionnaire are combined with the characteristics of each module.The machine learning model is sent and the questionnaire credibility score is obtained.