Integrating Textual Features with Survival Analysis for Predicting Employee Turnover.

Area of Science:

• Human Resources Management
• Data Science
• Organizational Behavior

Background:

• Employee turnover poses significant costs to organizations.
• Traditional turnover prediction models often lack nuanced insights into employee sentiment and temporal dynamics.
• Professional networking platforms offer rich textual data that can be leveraged for predictive modeling.

Purpose of the Study:

• To develop and validate a novel methodology for predicting employee turnover.
• To integrate Transformer-based textual analysis with demographic variables using survival analysis.
• To enhance the accuracy and interpretability of turnover predictions for HR decision-making.

Main Methods:

• Utilized a dataset of 4087 work events from Maimai (a Chinese professional networking platform) from 2020-2022.
• Employed a hybrid feature extraction strategy combining sentiment analysis, TF-IDF, and Transformer-based deep learning semantic representations.
• Applied survival analysis to model time-dependent turnover risks and compared various predictive models.

Main Results:

• Integrating textual and demographic features significantly improved prediction performance, increasing the C-index by 3.38% and cumulative/dynamic AUC by 3.43%.
• Transformer-based text analysis demonstrated superior performance in capturing subtle employee sentiments compared to traditional methods.
• Survival analysis enhanced model adaptability by incorporating temporal dynamics and identified interpretable turnover risk factors.

Conclusions:

• The novel methodology effectively combines advanced text analysis with survival modeling for superior turnover prediction.
• This approach offers small and medium-sized enterprises a practical, data-informed tool for workforce planning and talent retention.
• Findings contribute to labor market insights, informing organizational strategies and policy-making for talent management.