應用機器學習預測求職網站會員投遞履歷之行為

NCUIR > School of Management at National Central University > Executive Master of Information Management > Electronic Thesis & Dissertation > Item 987654321/89898

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/89898

Title:	應用機器學習預測求職網站會員投遞履歷之行為
Authors:	陳姵嘉;Chen, Pei-Chia
Contributors:	資訊管理學系在職專班
Keywords:	機器學習;投遞履歷預測;隨機森林;machine learning;applied jobs prediction;random forest
Date:	2022-09-15
Issue Date:	2022-10-04 12:04:04 (UTC+8)
Publisher:	國立中央大學
Abstract:	徵才廣告多數以無差異化方式在求職網站上進行曝光，為有足夠的曝光量，可能透過求職網站上的展示型廣告增加曝光量，許多徵才企業會以收到求職履歷數作為曝光效果的衡量依據，一旦收取的履歷數過低，就可能引起客訴、補償客戶等行為，而精準廣告是增加對目標族群曝光方式之一。本研究將以個案求職網站的角度，透過機器學習方式預測在職中求職會員在求職網站上投遞履歷之行為，並探討投遞履歷人數較多與較少的冷門、熱門時段之差異。本研究比較隨機森林、支援向量機、羅吉斯回歸與貝式分類器四種方法在預測分類上的有效與精確性，並結合主成分分析與資訊增益兩種特徵選取、萃取方式，比較兩種特徵選取在預測上的幫助。實驗結果中於冷熱門時段與各機器學習方法組合探討下，羅吉斯迴歸、隨機森林在冷熱門時段 Macro-F1 與召回率皆為 64%以上、AUC 為 70%，三種衡量指標皆較貝氏分類器與支援向量機佳;精確率不論冷熱區間，四種模型表現皆可達 60%。以 Macro-F1、召回率、AUC 表現較佳的羅吉斯迴歸、隨機森林更仔細地探討發現:兩種方法在冷門、與熱門時段表現趨勢皆較一致。 1. 在召回率表現中，隨機森林無投遞履歷召回率為 70%、有投遞履歷為 65%;羅吉斯迴歸無投遞履歷召回率則為 80%、有投遞履歷招回率約 40%以上，兩者差異較隨機森林大。 2. 隨機森林有投遞履歷的精確率 48%與無投遞 80%兩者差異最大，且在熱門時段下是本次研究中比較的四種模型中最低的;羅吉斯迴歸有投遞的精確率約為 50%、而無投遞履歷為 70%，表現與貝氏分類器、支援向量機相似。 ;Most recruitment advertisements are advertised on job search websites in an imprecise way. In order to have more exposure, it is possible to increase exposure through display advertisements on job search websites for expanding their reach. The number of job application resumes is used as the basis for measuring the exposure effect. When the customers get fewer job applications, it may cause customer complaints and compensate customers. Accurate advertising is one of the ways to increase exposure to target groups. This study uses machine learning to predict the behavior of job seekers applying for jobs on job search websites, and explores the differences between unpopular and popular times when there are more and less people applying. Besides, this study compares the results of four algorithms, namely Random Forest, Support Vector Machine, Logistic Regression and Bayesian classifier in the prediction classification, and combines two feature selection of principal component analysis and information gain to compare the two methods which could help in prediction. In the experimental results, the Macro-F1and the recall rate of Logistic regression and random forest in the cold and hot period are both over 64%, and the AUC are 70%. Except for the support vector machine, the precision of others is all over 60%. Furthermore, Logistic regression and random forest are consistent in the trend of unpopular and popular periods. The relevant descriptions are as follows. 1. Random forest has the recall of 70% without applied job, and 65% with applied job. Logistic regression has the recall of 80% without applied job , and 40% with applied job. 2. The precision of random forest with applied jobs is 48%, and without applied jobs is 80%. The precision of Logistic regression is similar to Bayesian classifier and support vector machine, which is 50% without applied jobs and 40%with applied jobs.
Appears in Collections:	[Executive Master of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	152	View/Open

社群 sharing

Loading...