Adaptive Semantic Intelligence for Workforce Experience Attribute Inference Using Transformer Representations

Abstract

The widespread adoption of digital workplace platforms has enabled employees to share detailed feedback regarding their professional experiences, creating valuable textual resources for understanding organizational performance and workforce satisfaction. Employee reviews provide important insights into workplace culture, career opportunities, compensation, job security, and work-life balance. However, manually analyzing these large volumes of unstructured feedback is inefficient, subjective, and incapable of supporting timely organizational decision-making. Furthermore, conventional text mining and machine learning approaches often struggle to capture contextual semantics, handle imbalanced datasets, and simultaneously predict multiple workforcerelated factors with high accuracy. To address these challenges, this research proposes an intelligent fine-grained opinion mining framework for employee review analysis. The proposed system initially performs comprehensive Natural Language Processing (NLP) preprocessing to improve textual quality, followed by contextual feature extraction using Google Pathways Language Model (PaLM). To mitigate class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is employed to generate representative samples for minority classes, thereby improving model generalization. The extracted semantic representations are subsequently processed using the proposed Transformer-Guided Adaptive Model (TGAM) for simultaneous prediction of multiple organizational factors, including work-life balance, skill development, salary and benefits, job security, career growth, and overall work satisfaction. Comparative performance analysis is conducted using Quadratic Discriminant Analysis (QDA), Linear Discriminant Analysis (LDA), and Histogram-Based Gradient Boosting (HGB) as baseline models. Experimental results demonstrate that the proposed TGAM consistently outperforms the existing approaches by achieving superior predictive accuracy across all target variables. The developed framework provides an effective decision-support system for workforce analytics, enabling organizations to understand employee perceptions, improve workplace policies, and support data-driven human resource management.