Research On Talent-post Matching And Resignation Warning For Talent Computing

Talent computing is a new method system that combines various computing-related technologies with various application problems in the talent field,such as talent selection,evaluation,matching,and prediction.The talent computing research is of great value to enterprises,social economy,and even national strategic talent deployment.With the digital development of talent management,new opportunities and challenges have been brought to the research of talent-oriented computing,and the related research has become a hot topic of current research.Previous studies demonstrate that performance prediction,ability-aware matching,personality trait recognition,and turnover prediction have an important impact on improving the degree of Person-Job matching and alleviating the brain drain.To this end,based on deep learning technology,this paper studies a series of key issues of performance prediction,ability-aware matching,personality trait recognition,and turnover prediction for talent computing,in order to accurately identify talents and risk leavers,aiming to provide support for reducing enterprise costs and improving enterprise efficiency.The main work and contributions in this paper can be summarized as follows:(1)We propose a two-stage method for personnel performance prediction.Performance prediction can help managers to identify the suitable candidates and improve the competitiveness of enterprises,but it faces the challenges of complex performance factors and low accuracy.To this end,we propose a two-stage performance prediction model.The performance prediction problem is essentially a classification problem.Firstly,a hybrid model is presented,which can automatically capture the complex dependence of performance characteristics.The contributions of various performance factors to the final decision is different,and attention mechanism is adopted.Then the final performance prediction judgment is made in the classification layer.Finally,experiments on a performance data are conducted to evaluate the performance of the model.The results show that the proposed method significantly improves the performance of personnel performance prediction.(2)We propose a performance prediction neural network model based on capsule network.Performance decision-making problem can be broken into a hierarchical structure related to the decision elements,and capsule network just simulates the idea of hierarchical classification of personnel performance prediction task,which makes it suitable to use capsule network to solve performance prediction task.Specifically,for modeling the personnel performance,we first devise gated recurrent unit to represent performance characteristics in the first layer.The second layer clusters low-level features into high-level capsule representations.The last layer predicts personnel performance category for each capsule representation.The performance prediction task also has the class imbalance issue,and the corresponding objective function is designed in the performance prediction layer.Experimental results show that the proposed model achieves better classification performance than other benchmark models.(3)A job resume fit neural network based attention model is proposed.Person job fit is to judge whether the candidate's skills fit the job objectives of the enterprises,which can help managers to identify the suitable candidates,improve the competitiveness of enterprises,and reduce labor costs.However,the existing methods do not fully consider the semantic features of job resumes.Based on this,first,a word-level job resume text representation method is designed to learn the semantic features of job requirements and resume experience.Second,the ability perception recognition based on attention mechanism is used to measure the different importance of the company's focus on the ability in the job requirements,as well as the different contributions of the job seeker's ability in completing the task.Third,to measure the fit degree between the position and the resume,semantic similarity is used.In the last module of the model,the vector that represents the semantics of the job resume is transferred to the fully connected layer with sigmoid function for the judgment of person job fit.Finally,experiments on real data verify the effectiveness of the proposed model.(4)We propose a semantic-enhanced multi-labeled personality trait recognition model.Personality trait recognition is to judge whether the candidate's personality characteristics fit the job objectives of the enterprises,but it faces the challenge of inaccurate expression of semantic information in the word embedding layer and failing to achieve the best performance on each label when using text for personality trait recognition.Aiming at these challenges,based on the recurrent neural network,a semantic-enhanced multi-labeled personality trait recognition model is proposed.First,context learning method is used to capture text information.Then,a fully connected layer is used to fuse the semantic information of personality traits,and further capture high-level semantics of texts.Finally,we adopt binary cross entropy loss function to identify the multi-label personality traits.The proposed model avoids dependence on feature engineering,and allows the same model to adapt to detecting five different personality traits labels without modifying the model itself.Experimental results show that the proposed model achieves the best classification performance on each label compared with other methods.(5)We propose a profile-aware personnel turnover prediction neural network model.Turnover prediction is helpful for decision makers to take preventive measures to retain the key talents who are expected to be at risk of turnover,and reduce costs,but it faces some challenges in practice.On the one hand,the results of different turnover behavior characteristics are different,and the performance of existing turnover prediction methods is not high.On the other hand,the turnover data are imbalanced,and turnover possibility has not been considered.In view of these challenges,we construct a deep learning model based on turnover profile to solve the above challenges.Specifically,we use gated recurrent unit to learn profile-aware features representations of the personnel samples.To evaluate the contribution of various turnover factors,we further introduce an attention mechanism.Along this line,we creatively design a weighted-based probability loss objective function.The performance of the proposed model is verified on two turnover data,and the experimental results show that the proposed model can effectively improve the F1 in terms of turnover prediction.