Enhancing Talent Intelligence Evaluation with Improved Pre-training Models

Abstract

In the past few years, there has been a noticeable surge in the utilization of artificial intelligence techniques within the realm of talent evaluation, particularly in response to the exponential growth of big data. Conventional talent evaluation approaches, typically relying on the integration of expertise and system-based title appraisal, are plagued with challenges including errors in subjective judgment. The strategic importance of intelligent talent assessment cannot be overemphasized, especially for discovering and fostering high-potential talent across various domains, inclusive of research. This paper introduces an innovative method, termed TE-RCB, auxiliary applications in the evaluation of actual talent titles. The initial stage of this method’s execution involves the meticulous construction of a talent evaluation dataset. This data is sourced from actual talent title applications and standardized scores provided by simulation expert engineers. The subsequent step is the application of the RoBERTa-WWM-large model, employed for the vectorization of pertinent information indispensable for talent evaluation. Further, the TextCNN model is utilized for the extraction of critical features from the talent-related information, succeeded by the application of the BiLSTM model to delve into deeper semantic correlations within the talent attributes. Comprehensive experiments are conducted on both the talent evaluation dataset and the publicly available dataset. The empirical outcomes decisively underline the superior accuracy of the proposed method, asserting its efficacy in addressing the pertinent challenges in the domain of intelligent talent assessment.