BERT-Based Transfer Learning Model to Enhance Human Resource Performance Appraisal System

The appraisal system in organizations has been plagued by several challenges, including inefficiency, discrimination, perceived unfairness, and an operational-focused approach. These issues hinder the effectiveness of performance appraisal processes, leading to suboptimal outcomes. Previous studies have highlighted these problems and proposed alternative approaches like coaching-oriented system which focuses on employee behavior and leadership development, while also advocating for the separation of appraisal from rewards to mitigate biases. However, there are gaps remain in addressing the limitations of automated systems such as the crucial role played by the quality and diversity of the training data in determining the accuracy and efficiency of the NLP model, along with acknowledging the irreplaceable role of human involvement. Our study aimed to address these issues by automating the feedback review process using a transfer learning model, enhancing accuracy, efficiency, objectivity, and scalability. This study implements the BERT model from Hugging Face and achieves promising validation metrics of better performance when compared to other language models such as Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) models. It is proved that biases and decision-making process in the traditional appraisal system can be improved with the deployment of advanced NLP models such as BERT. Furthermore, this model promotes a more equitable and transparent workplace environment by scaling and consistency in performance review.