Unveiling employee job satisfaction: Harnessing deep learning for sentiment analysis

Abstract

After the COVID outbreak, the trend in job satisfaction changed drastically and many individuals lost their employment. Understanding the social movement requires careful consideration of the variables affecting job sentiment. This study is primarily interested in examining the sentiments of workers at well-known Information Technology organizations. In the beginning, 1,56,428 reviews from various timelines were scraped from Glassdoor. The remarks are then carefully inspected to realize a variety of elements, and the required preprocessing techniques have been used to facilitate a deeper comprehension. Later, various Deep Learning architectures are given the preprocessed comments. The Bidirectional Gated Recurrent Unit (Bi-GRU) has outperformed all other architectures in terms of many performance measures, even though some other architectures have performed noticeably admirably. The generated result shows a 97% accuracy rate and correctly distinguishes the two classes included in the dataset. Apart from this architecture, nine other architectures are also investigated. Among them, Bi-CuDNNLSTM, GRU, and CNN also performed quite well and produced 97%, 96%, and 90% accuracy respectively.