Research On Reputation Evaluation Of Crowdsourcing Participants Based On Machine Learning Techniques

Crowdsourcing is an online activity through which employers openly call crowdsourcing participants to complete tasks through crowdsourcing platform,which is an effective way to solve problems remotely.Crowdsourcing breaks through the traditional business model,can effectively aggregate the wisdom of the crowd,and gather wisdom from all fields to participate in technological innovation and value creation.Crowdsourcing participants take part-time or full-time join in crowdsourcing activities in free time,provide valuable labor achievements,release social human resources potential,promote open innovation of enterprises,and promote social division of labor and economic development.At present,the reputation evaluation mechanism of crowdsourcing platform in China is not perfect,the evaluation index of crowdsourcing participants is single,the evaluation method is simple,and the differentiation ability is poor.It is difficult to effectively regulate and restrict the behaviors of crowdsourcing participants,which increases the transaction risk and affects the normal operation of crowdsourcing activities.The reputation evaluation mechanism is an effective way to establish the trust of both sides of the transaction,restrain the illegal behaviors of crowdsourcing participants,assist the employer in transaction decision-making,maintain the trading order of the platform,reduce the transaction risk,and ensure the successful completion of crowdsourcing tasks.It is also the cornerstone to solve the dynamic integration of human resources,team cooperation,task matching,task pricing,task quality control and other problems faced by crowdsourcing.Using scientific and reasonable reputation evaluation method to evaluate the reputation of crowdsourcing participants objectively,comprehensively and accurately is the key problem to be solved for the healthy,orderly and rapid development of crowdsourcing.Therefore,it has realistic significance and practical significance to study the reputation evaluation of crowdsourcing participants in this doctoral dissertation.The paper puts forward the selection approach of reputation evaluation index of crowdsourcing participants,classification approach of reputation,and reputation scoring approach,which are of theoretical significance and academic value.This paper puts forward a new way of reputation evaluation of crowdsourcing participants.Based on the selection of reputation evaluation indexes,this paper studies the reputation classification and reputation scoring of crowdsourcing participants,which has the following three innovations:1.A hybrid two-stage crowdsourcing participant reputation evaluation index selection approach(Relief F-SVM)based on Relief F is proposed.The hybrid two-stage crowdsourcing participant reputation evaluation index selection approach based on Relief F is described as follows: by expounding the reputation characteristics of crowdsourcing participants,systematically analyzing the reputation evaluation indexes of crowdsourcing platform,combing the relevant literature,following the selection principle,from the initial reputation dimension,transaction dimension,evaluation dimension,punishment dimension,28 initially selected reputation evaluation indexes of crowdsourcing participants are proposed.Taking the data set of crowdsourcing participants of Zhubajie platform as an example,the reputation evaluation indexes of crowdsourcing participants are selected in two stages.In the first stage,the best feature selection approach is selected from four dimensionality reduction methods: Relief F,mean impact value(MIV),principal component analysis(PCA)and linear discriminant analysis(LDA).In the second stage,the sequential backward selection(SBS)strategy is adopted,and the classifier is used as the evaluation function of feature selection.The Relief F-SVM algorithm with the best classification performance is choosed to select the reputation evaluation index of crowdsourcing participants.The reputation evaluation index selection approach(Relief F-SVM)of crowdsourcing participants based on the feature selection of Relief F can select the comprehensive,objective and effective evaluation indexes to identify the reputation status of crowdsourcing participants,which makes up for the single existing evaluation index of crowdsourcing platform,and is difficult to reflect the whole reputation of crowdsourcing participants.2.A heterogeneous ensemble classification approach(QGA-Hstacking)for crowdsourcing participants’ reputation is proposedAccording to the research results of the reputation evaluation index selection,taking the decision tree(DT),support vector machine(SVM),nearest neighbor classifier(KNN)and naive bayes(NB)as the base classifiers,and single classifier SVM with the best classification performance as the meta model,a heterogeneous ensemble classification approach(QGA-Hstacking)based on stacking classifier combination strategy and quantum genetic algorithm(QGA)hyper-parameter optimization is proposed.Taking three data sets as an example,using generative adversarial networks(GAN)to deal with unbalanced data sets,the QGA-Hstacking heterogeneous ensemble classification approach is compared with four homogeneous ensemble algorithms of random forest(RF),Ada Boost,GBDT and XGboost.The results show that the QGA-Hstacking significantly improves the classification accuracy of the base classifier,and has higher classification precision,stronger generalization ability and stability than homogeneous ensembles methods,is more suitable for crowdsourcing participants’ reputation classification evaluation.3.A reputation scoring approach(RSM-SA)combined with text sentiment analysis of crowdsourcing participant is proposed.The reputation scoring approach of crowdsourcing participants combined with text sentiment analysis is described as follows: firstly,a supervised sentiment analysis classification approaches(LDA-GBDT)combined with latent dirichlet allocation is proposed.By expanding the extraction features of employer review text,the ability of classifier to distinguish review text with fuzzy emotional boundary is improved.Then taking the emotional classification results of review text as one of the reputation evaluation indexes,considering the selected reputation evaluation indexes of crowdsourcing participants,by using mathematical method the RSM-SA scoring model is constructed from the initial reputation dimension,transaction dimension,evaluation dimension and punishment dimension.Finally,the rationality and validity of the model are verified by empirical analysis.The reputation scoring classifier of crowdsourcing participants proposed in this paper is to solve the shortcomings of WCRSM scoring model of crowdsourcing platform.Combined with structured and unstructured data generated by crowdsourcing participants’ trading activities,using machine learning technology and mathematical method,a crowdsourcing participants’ reputation scoring model(RSM-SA)is constructed to comprehensively quantify the reputation score of crowdsourcing participants.The empirical analysis shows that this approach improves the WCRSM reputation scoring method of crowdsourcing platform,and can more truly,comprehensively,accurately and dynamically feedback the reputation status of crowdsourcing participants.