APP下载

融合知识图谱的多行为职位推荐

2025-08-09刘滨刘格格詹世源高歆杨晓艳

河北科技大学学报 2025年3期
关键词:三元组求职者职位

中图分类号:TP393 文献标识码:A DOI:10. 7535/hbkd. 2025yx03011

Multi-behavioral job recommendation integrating knowledge graph

LIU Bin 1,2 ,LEI Xiaoyu 1,2 ,LIU Gege 1,2 , ZHAN Shiyuan 1,2 ,GAO Xin 1,2, ,YANG Xiaoyan 1,2 (2 (1.School of Economicsand Management,Hebei Universityof Scienceand Technology,Shijiazhuang,Hebei O5oo18,China; 2. Research Center of Big Data and Social Computing,Hebei University of Science and Technology, Shijiazhuang,Hebei O50o18,China)

Abstract:Toimprove the accuracyof job recommendations,amulti-behavioral jobrecommendation integrating knowledge graph(MB-JRIKG)was proposed basedonthe implicit preference informationof both job seekers and recruiters inthe process of \"browsing job positions $$ submitting resumes $$ receiving feedback from recruiters\". This method constructed a knowledge graphof jobseeking basedonrealjobdata,andproposedamultibehaviorpreferencepropagationstrategybasedonpreference propagationtheory.\"Recruiterrecognition\"ineach stageof jobseekingwassetasthe targetbehavior,and theauxiliary behaviorsof job seekers browsing job positions and submiting resumes wereset tocomprehensively predict job seekers' preferences.Firstly,the user's historical records under diferent behaviors wereused as user-interested seed sets,and preference propagationwascarriedoutalong therelationshipsbetwee nodesinthe knowledgegraph toinfer theuser's potential preferences and enhanceuser representation;Then,the user representation vectorand positionrepresentatio vector were input intothe predictionfunction tocalculate the interactionprobabityof theuserundereach behaviortype,and the weightedsumwasusedastheinteractionprobabilityofthetargetbehavior.Finall,theclick-throughrateprediction experiment wasconductedusing thecompetition datasetof Alibaba'shuman job intellgent matching.Theresults showthat compared with the four benchmark models(MF,XGBost,KGCNandRippleNet),MB-JRIKG achievesa 0.014 5 improvementinAUCanda0.O288improvement in ACCover the sub optimal benchmark model RippleNet,verifying the effectivenessofthemodelandachievingfullutilizationofdata.Thismodelefectivelycombinestheinteractivebhaviorofboth parties seeking employmentfor recommendation,introduces atribute association of job knowledge graph,and has reference value for achieving personalized job recommendation.

Keywords:natural language procesing;jobrecommendation;knowledge graph;multiple behaviors;preference propagation

网络招聘因其成本低、效率高、不受时空限制等特点而得到快速发展,相关平台集聚了海量的简历和职位信息,解决信息过载问题、实现智能招聘是当前的主要挑战。推荐算法因其可以帮助用户在海量内容中进行高效遴选,提升了用户体验,从而广泛应用于电影[1]、音乐[2]、新闻资讯[3]、电商[4]等推荐场景,并成为产业界和学术界持续研究的重点领域。当前主流的推荐算法主要包括基于协同过滤的推荐算法[5]、基于内容的推荐算法[6]和混合推荐算法[7]。基于协同过滤的推荐算法能够有效捕捉用户偏好,并易于在多种场景下实现,应用较为广泛。然而,数据稀疏性和冷启动问题是推荐算法面临的普遍性挑战,相关工作将背景信息融合进推荐过程,如用户-项目属性[8]、社交网络[9]等,并获得了良好收益。知识图谱作为一种以图结构表示的语义知识库,可以为推荐系统提供项目间丰富的语义信息和结构关系,有助于挖掘其间的潜在联系,提升推荐项目的有效性。

此外,与传统的电商或社交推荐场景不同,求职者在网络招聘平台的求聘过程为“浏览岗位 -投递简历 -招聘者反馈”,求聘各环节行为存在递进关系,如该简历和职位配对得到招聘者认可的前提是求职者对该职位有浏览岗位并投递简历的行为。除了最终的目标行为(招聘者认可)外,其他辅助行为(浏览岗位、投递简历)的交互记录同样蕴含着大量的求聘双方偏好信息。……

登录APP查看全文

猜你喜欢

三元组求职者职位
基于扎根理论的大学生求职信息跨源搜寻行为过程模型研究
有限域 Fpn 上完全置换多项式的构造
宜都市:让民营企业与求职者“双向奔赴”
公共就业服务存在的问题与对策研究
基于句子转换和双注意力机制的归纳关系预测
“皇恩浩荡”中的司马池
美国副总统这差事
话题
招聘管理的数字化创新