Article
A Transformer-Based Multi-Label NLP Architecture for Extracting Workforce Experience Indicators from Amazon Employees Data
In contemporary organizations, employee reviews shared on platforms such as Amazon represent a valuable source of insight into workplace dynamics; however, deriving meaningful information from this largely unstructured textual data remains a significant challenge. Conventional approaches, including manual inspection, heuristic-based analysis, and simple keyword-driven sentiment techniques, are not only time-intensive and inconsistent but also inadequate for capturing the nuanced context and multiple dimensions embedded within employee feedback. As a result, organizations face difficulty in accurately assessing critical workforce factors such as work-life balance, career progression, and job security when dealing with large volumes of reviews. These traditional methods lack scalability, struggle with complex sentiment interpretation, and often yield suboptimal accuracy, underscoring the necessity for a more advanced and intelligent solution. To overcome these limitations, the proposed system employs a transformer-based Natural Language Processing (NLP) framework integrated within a real-time Flask web application, utilizing Sentence-BERT (SBERT) embeddings for deep semantic feature extraction alongside a Stochastic Gradient Descent (SGD) based multi-label classification model, with a Hashing Vectorizer combined with Bernoulli Naive Bayes (BNB) serving as a comparative baseline. The framework executes a comprehensive pipeline encompassing data preprocessing, feature engineering, model training, prediction, and evaluation, enabling the simultaneous identification of multiple workplace experience indicators from a single review. By leveraging contextual understanding and scalable automation, this approach enhances prediction accuracy, minimizes manual intervention, and delivers actionable insights, thereby supporting datadriven human resource analytics and informed organizational decision-making
Full Text Attachment
