Online Labor Markets Hiring Decisions Based On Deep Learning

With the development of Internet technology and the rise of sharing business models,a growing number of workers tend to choose work as freelancers on project oriented remote labor platforms instead of being traditional full-time workers.This new type of labor business model that breaks through organizational boundaries and nationality boundaries not only promotes global economic development,but also reshapes the global labor-employment relationship.The online labors rely on the Internet platform to effectively match the supply and demand of the scattered labor,and gradually form a huge online labor market.Although the online labor market is booming and diverse,the research and application of the online labor field are still in the exploratory stage,and many users' behaviors and economic phenomena have not yet been systematically studied.Based on the one of the largest online labor markets in the world,Freelancer platform,this thesis uses data-driven research methods to explore the issue of employer's hiring decision-making and applicant's project recommendation in the online labor market.Specifically,from the perspective of employers and applicants,this thesis focus on three major issues faced by online labor platforms:employer's hiring decision-making issue,reputation evaluation issue,and applicant's project recommendation issue.This thesis takes Freelancer platform with rich data as an example,makes use of deep learning algorithms to design corresponding predictive models,explores the online behavior of employers and applicants,and provides a scientific basis for achieving efficient matching of supply and demand in the online labor market.The main work and contributions of the thesis are summarized as follows:(1)An employer hiring decision-making method based on the deep choice model is proposed.The deep choice model is based on a pointwise convolutional neural network to construct an applicant's nonlinear utility function and estimate its utility value,while considering the entire set of applicants to estimate the hiring probability of each applicant.From the perspective of employers in the online labor market,when faced with many applicants,it is difficult for employers to make hiring decisions.A deep choice model that considers the entire set of applicants and the attributes of applicants is proposed.This model combines the features of point-wise convolutional neural network and the conditional logic model.Compared with the traditional conditional logic model,the purposed model can capture more complicated decision-making process and provide an effective solution for employers' hiring decision.(2)An applicant reputation evaluation model based on multiplicative long short-term memory recurrent neural networks is proposed.This model takes characters as the unit to segment text and conduct model training.Through the idea of transfer learning,the trained model will be transferred from the massive evaluation text to evlaute the reputation textual information in the online labor market platform.From the perspective of employers,the reputation ratings of applicants have inflated,which makes them hard to evaluate applicants' qualities.Focusing on review texts,a reputation evaluation model based on multiplicative long short-term memory recurrent neural networks is proposed.This unsupervised model is used for sentiment analysis of review text.Moreover,combined reputation evaluation model with deep choice model,an employer hiring decision-making method based on multiplicative long short-term reputation analysis is constructed.This method can predict employers' hiring decisions accurately and improve the applicability of employer hiring model.(3)A project recommendation model for applicants based on conditional variational autoencoders is proposed,which can simultaneously consider applicants' historical information and their own attributes.From the perspective of applicants,that applicants and projects are hard to accurately match,an applicant-project correlation matrix is constructed.The matrix is severely sparse and is difficult to directly optimize the objective function.In order to solve these problems,a deep hidden variable generation model is adopted.The neural network fits the probability distribution.The project recommendation model based on conditional variational autoencoding is puprosed.Compared with the collaborative filtering recommendation model based on variational autoencoder,the collaborative filtering model based on linear variational autoencoder and the hybrid recommendation model based on linear variational autoencoder,the purposed model is more suitable for application in the online labor market applicant project recommendation system.(4)Expriments of proposed models are conducted on Freelancer data set.The proposed models are employer hiring decision-making method based on the deep choice model,the applicant reputation evaluation model based on multiplicative long short-term memory recurrent neural networks,and the project recommendation model based on conditional variational autoencoding.The experimental results show that purposed models are more reliable than the existing models.The research results have achieved more efficient matching of labor supply and demand in the online labor market,and provided technical support for the online labor market platform to achieve personalized,intelligent,and precise services for both employers and applicants.