Automatic Text Processing and Sentiment Classification System for Digital Application Reviews Based on Bidirectional Language Model

Abstract

ABSTRACT This invention provides an automatic text processing and sentiment classification system for handling large-scale informal reviews on digital platforms characterized by the use of informal language and specialized jargon. The main problem of digital review analysis lies in the high complexity of casual linguistics that degrades standard computing performance. To overcome this obstacle, this system integrates an automatic text pre-processing module that is specifically configured to clean text noise, perform non-standard word normalization, and reduce words to their basic forms. Furthermore, the system implements a main processing module based on the Indonesian bidirectional Deep Learning Transformers BERT (IndoBERT) architecture that has been refined through domain-specific fine-tuning techniques. This technique allows the artificial intelligence model to deeply understand the nuances of contextual meaning.All of these data flows are presented through an interactive web interface module for comprehensive visualization of analysis results. Through the integrated architectural configuration, this invention is able to significantly increase the accuracy value of the sentiment classification system to 83.2% when compared to the standard base model that has not been optimized. This system can be widely implemented as a practical computing tool for application developers and digital industry players in evaluating user opinions, complaints, and preferences accurately, objectively, and in real-time. Keywords: text processing, automatic sentiment classification, digital platforms, special jargon.