Research On Person-job Fit Method Based On BiLSTM And XGBoost

Online recruitment is one of the mainstream recruitment forms at this stage,and it plays a role as a bridge connecting job seekers and recruiters.Person-job fit is a key research problem in online recruitment.Benefiting from the development of natural language processing technology,most of the existing researches on person-job fit methods are based on the two textual features of job description and job seeker’s work experience.On the one hand,these studies do not take into account other types of text data,such as expected industries and job browsing record.On the other hand,the job application process of job seekers is "job browsing→ resume delivery → whether the recruiter approves or not",and previous studies have not taken into account the influence of the prior behavior of the job applicant on the subsequent application results.Based on the perspective of online recruitment platform,this paper builds a person-job fit model,including two stages of job and resume text similarity calculation based on Bi LSTM and person-job fit modeling based on XGBoost.The main research contents of this paper are as follows:(1)This paper proposes a method for calculating the text similarity between job and resume based on Bi LSTM.Firstly,from the perspective of recruitment platform,this paper gives a formal description and definition of the computing problem of similarity between job and resume text.Then,this paper explains the construction method of the proposed job and resume text similarity calculation model,including the job description feature representation based on Bi LSTM,the work experience feature representation fused with job browsing records,the job sub-category feature representation,and the job and resume text similarity calculation.Different from previous studies,this paper also considers the features of job browsing records,job sub-categories,and desired job sub-categories,aiming to mine job seekers’ preferences for jobs and enrich job and resume representations.Then,the model is validated on the person-job fit dataset.The numerical results show that the proposed model is effective,and ablation experiments show the effectiveness of considering the features of job browsing records,job sub-categories,and desired job sub-categories.Finally,in order to illustrate that the model is helpful to improve the service efficiency of recruitment platform,a case study is carried out.(2)This paper proposes a person-job fit method based on XGBoost.Firstly,this paper formally describes and defines the problem of person-job fit.Then,this paper describes the construction method of the proposed person-job fit model,including three steps: original feature analysis of jobs and resumes,feature construction and person-job fit prediction.The step of original feature analysis analyzes and preprocesses the original features of jobs and resumes.The step of feature construction constructs 11 features from the perspectives of "similarity between job seekers and target jobs" and "job seekers’ competitiveness among candidates applying for the same job",and the text similarity between job and resume is one of the most important features.The step of person-job fit input features into XGboost to calculate personjob fit degree.Then,in order to verify the effectiveness of the person-job fit model proposed in this paper,the numerical experiment part compares the model proposed in this paper with the models using several other machine learning methods,and uses the ablation experiment method to test the effectiveness of the constructed features.Finally,in order to illustrate the effect of person-job fit using the model,a case study is carried out.