Aspect-Based Sentiment Analysis of Job Reviews

Online job reviews provide critical insights into employee experiences, aiding organizations in improving workplace dynamics and informing job seekers about aspects like worklife balance, career growth, compensation, and company culture. This study proposes an aspect-based sentiment analysis framework for Glassdoor. We preprocess a Kaggle dataset of 75,012 reviews, extract contextual embeddings with BERT, and capture sequential dependencies with LSTM to generate sentiment scores for four aspects. The hybrid BERT-LSTM model helped to achieve an accuracy of 93.18%. SARIMAX forecasting models predict sentiment trends over time, with a Mean Squared Error of 0.01-0.03. The developed web application includes various features such as company-wise aspect-based scores, spatial and temporal analysis to present aspect scores based on location and time. The web application also integrates job opening alerts, enhancing its utility for career exploration. This approach offers organizations actionable insights into employee feedback and supports job seekers in decision making.