Sentiment Analysis of P2P Lending Fintech Service User Comments Using CNN-ROS-NCL on Imbalanced Data

Fintech P2P lending services have grown rapidly and thousands of comments from Fintech P2P application users appear every day, which is important to note as material for improving the system and its services. The purpose of this study is to explore machine learning and deep learning models to classify Fintech P2P Lending user comments based on sentiment. Classifying user comments efficiently is increasingly important to improve user experience and operational effectiveness. Data imbalance poses a challenge, which often hinders the performance of classification models. This paper uses Lexicon to determine sentiment polarity, CNN algorithm to perform classification, and ROS-NCL algorithm to handle imbalanced data. The ROS-NCL preprocessing strategy is able to balance the dataset and improve the quality of training data, so that CNN can effectively learn more general and representative features. We also compare the imbalanced data handling algorithms SMOTE, SMOTEENN and ADASYN. The experimental results show a significant improvement in classification performance with an accuracy of 99.10% and a misclassification rate of only 0.90%. This level of accuracy can be used to build applications based on the classification of P2P lending user comments that are currently trending so that users can make good use of Fintech P2P lending.