Model Analytic in Fintech User Comment Features Using LDA-CNN on Imbalanced Data
International journal of intelligent engineering and systems · 2024
Peer-to-peer (P2P) lending platforms are growing significantly, and users always leave comments on the application to provide ratings.User comments are important to analyze to see the needs and constraints of fintech users.The Purpose of the research is to create an analytical model that effectively addresses the problem of limited accuracy in classification due to data imbalance in P2P Lending platforms.The research aims to improve feature detection and overall model quality by effectively managing imbalanced data.The design involves a combination of techniques.First, the Latent Dirichlet Allocation (LDA) method is used to organize topics and label data.To address the data imbalance, the study employs Random Over Sampling (ROS) and Neighborhood Cleaning Rule (NCL).The final classification is performed using Convolutional Neural Networks (CNN).Additionally, a comparative analysis with other algorithms like LSTM and CNN-LSTM is carried out to validate the effectiveness of the proposed approach.The Findings reveal that the CNN-ROS-NCL model is capable of managing imbalanced data, which improves class distribution and enhances the model's quality by reducing noise and misleading samples.The CNN model achieved a classification accuracy of 94.66% on 10 feature classes, suggesting a significant improvement in feature detection and classification performance on the P2P Lending platform.The Originality of this research lies in the innovative integration of LDA for topic analysis with CNN for classification a novel approach in the context of fintech feature development.This combination has not previously been used in fintech and offers a new way to automatically detect features in Fintech P2P Lending user comment datasets by identifying key topics.The research contributes to the enhancement of fintech applications and services by providing a model that improves the understanding and processing of user comments on P2P platforms.