Research On Person-Job Matching Approaches For Intelligent Recruitment

Talents are not only the root of the prosperity of one country,but also the cornerstone of the survival and development of enterprises.The competition among enterprises is the competition for talents.Recruitment is an important channel for enterprises to acquire talents.With the advantages of low cost and convenience,online recruitment has promoted the extensive development and application of various online recruitment platforms.These platforms accumulated a large number of resumes of job seekers and job recruitment advertisements.They provide abundant job choices for job seekers and present sufficient candidate talents for employers.However,opportunities and challenges coexist.While online recruitment platforms bring great convenience,bilateral users of these platforms are confronted with the challenge of information overload.This gives birth to the application of intelligent recruitment system.The core step of intelligent recruitment is to solve the problem of Person-Job Matching automatically.At present,the problem of Person-Job Matching has been widely concerned by researchers in management science,computer science and their interdisciplinary fields.However,there are still some challenges,such as high cost of feature engineering,difficulty in semantic information mining and lack of multidimensional data fusion.Therefore,this paper conducts research on intelligent recruitment,focusing on Person-Job Matching problem based on different types of data.The main research contents and contributions can be summarized as follows.(1)Resume data of talents has the characteristics of time sequence,existing methods lack the effectively mining of this sequence data.Therefore,this paper proposes a Person-Job Matching model based on sequence data representation learning,and predict the career direction of talent in several views.Specifically,the model is composed of feature embedding module,sequence feature extraction module and classification module.In the feature embedding module,the resume data are grouped and sorted according to the attribute categories.We regard each resume as one sentence,all resumes are represented as two-dimensional vectors through pre-trained word embedding model.In the sequence feature extraction module,Convolutional neural networks and long short-term memory are used to extract sequence features from two-dimensional hidden vectors of resume.In the classification module,personalized prediction networks with different numbers of neurons are designed according to different tasks.Finally,a large number of experiments show that our proposed model can efficiently extract the resume attribute representations,mine the time-series relationship between attributes,and effectively predict the job title,salary level and company size of job seeker’s future job.(2)Resume and job data contain multi-dimensional attributes.Existing methods either highly rely on feature engineering,or ignore the impact of structural data on PersonJob Matching results.Considering both structural and un-structural data,this paper proposes a Person-Job Matching model based on hierarchical interactive learning of multidimensional data.Specifically,the model is composed of feature embedding module,multi-field feature fusion module and classification module.In the feature embedding module,different feature embedding strategies are developed according to different types of data.The entity embedding method is used to map all attribute values to the same lowdimensional vector space.ALBERT pre-trained model is used to learn representations for textual data and phrasal categorical data.In multi-field feature fusion module,a hierarchical feature interaction network is proposed,which includes field-inner interaction module and field-outer interaction module.In the field-inner interaction module,multiple vector interaction operators are used to learn the potential matching relationship between resume and job in a specific attribute field.In the field-outer interaction module,a feature interaction learning method based on multi-head self-attention and residual network is proposed.Multi-head self-attention combined with residual network is used to learn the interaction between different attribute fields.After that,the comprehensive representation of matching signals between resume and job in each field is mined.In the classification module,the matching features of each attribute field is fused,and the multi-layer perceptron is utilized to further predict the matching result between the resume and the job.Finally,a large number of experiments are carried out on the open data set,and the experimental results verify the validity and feasibility of the proposed model in solving the Person-Job Matching problem.(3)The desensitization data of talents and jobs in some recruitment scenarios are lack of information transparency,the Boolean matching method between attributes cannot effectively mine the potential matching signals between desensitization data.To this end,this paper proposes a Person-Job Matching model based on desensitized data interactive learning,which can independently and effectively predict the matching scores between resumes and jobs.This model can improve the information retrieval efficiency of both recruiters and job seekers.Specifically,the model is composed of feature embedding module,feature interaction module and classification module.In the feature embedding module,different embedding methods are utilized for different desensitization data.Linear normalization is performed for numeric attributes,and entity embedding is used to extract feature representation for categorical attributes and keyword attributes.In the feature interaction module,different interaction modules are designed for different types of attributes.A hybrid interaction method is adopted for categorical attribute vectors,and multiple vector interaction operators are designed to extract the potential matching relationship between resumes and jobs.The similarity matrix is constructed based on cosine distance for keyword attribute vectors.Then the convolutional neural network is applied to extract the interactive features between resume and job.In the classification module,we concatenate all local matching features of each module,and multi-layer perceptron network is applied to predict the matching degree between resumes and jobs.Finally,a large number of experiments are developed on real data sets that have been widely studied.The experimental results verify that the proposed model can effectively solve the Person-Job Matching problem based on desensitization data.