Logistic Regression Classification with TF-IDF and FastText for Sentiment Analysis of LinkedIn Reviews

Abstrak

Social media and professional networking platforms like LinkedIn have become crucial platforms for individuals to interact, share information, and build professional networks. Despite the significant benefits LinkedIn has provided to its users, there are still some limitations such as account restriction ambiguity, synchronization issues, and the emergence of spam and irrelevant content. Therefore, it is important to understand users' responses to the application. Previous research has shown that sentiment analysis can be an effective tool in understanding user reviews of applications. This study will continue previous research by analyzing the sentiment of user reviews of the LinkedIn application using the Logistic Regression method, taking into account the use of TF-IDF Feature Extraction and FastText Feature Expansion. Logistic Regression was chosen because it is effective in handling binary sentiment classification problems and has relatively high training speed. This method will be tested to address data imbalance and improve classification performance. This research demonstrates that this approach can provide optimal results in measuring accuracy, recall, precision, and F-Score. The research findings will provide valuable insights for LinkedIn application developers to enhance service quality. Based on the evaluation metrics, it can be observed that the first testing scheme with default parameters achieved an accuracy of 91.86%, a precision of 94.05%, a recall of 91.99%, and an F1-Score of 93.01%. The percentage values obtained already surpass 90%.