Sentiment analysis to determine employee job satisfaction using machine learning techniques

Abstract

Over the past three years, the COVID-19 epidemic had a significant impact on the labor market. Employees have been laid off and the majority of them have changed careers. If they can collect more datasets in the future, the researchers will be able to apply fine-tuning approaches to achieve perfect accuracy and precision. Incorporating hybrid models such as optimization techniques, multi-modal models, transfer learning models, hybrid deep learning models, sentiment models, etc. also broadens the scope of this study. These models can employ a variety of learning approaches, such as deep learning or traditional machine learning, and they can use many different types of data, such as text, images, or audio. The corpus was an additional strategy for improvement. These models consider lengthier texts in addition. 10% of US workers who keep their existing jobs are dissatisfied with them. Employee happiness is mostly influenced by business culture, but there are also cer tain economic and social elements that are interconnected. To ascertain the level of employee satisfaction and associated factors, significant study has been conducted. One of the most popular channels for opinion expression is social media. People now discuss the advantages and disadvantages of their work on the US-based social media site Glassdoor. For this study, total 1,56,428 data has been collected from Glassdoor.First, the data is correctly pre-processed after collection. The under standing of employee work satisfaction is provided by user ratings. For the purpose of making future predictions, the data was divided into binary class dataset and multiclass dataset. Moreover, this data is subjected to machine learning algorithms and deep learning algorithms. The best way to reach the ultimate conclusion is to use Bi-GRU for binary class dataset which has an overall accuracy of 97% and Bert model for multiclass dataset which has an accuracy of 95%.