INNOVATIONS IN NATURAL LANGUAGE PROCESSING FOR SMART ENTERPRISE SOLUTIONS

Abstract

Natural Language Processing (NLP) is fast revolutionizing enterprise activities through automated processing, classification, and management of text-based information at large scale. The proposed study is concerned with the issue of multi-class text classification in the context of enterprise news by enjoying the opportunities of transformer-based modeling to obtain high precision and reliability. The methodology included the utilization of a pre-processed and stratified Kaggle news dataset comprising 10,000 instances and 22 classes, as well as the DistilBERT architecture that is effective and supports contextual learning. The experiment process involved the use of advanced tokenization, optimized hyperparameters and stringent evaluation measures such as accuracy, macro and weighted F1-scores, per-class examination and visualization of the embedding through t-SNE and PCA. The findings showed that the DistilBERT classifier performed with an overall accuracy of 80.2%, and macro F1-score of 80.0%, and that it was strong even in infrequent and overlapping categories, which were well above the traditional and earlier deep learning baselines. The model has a nuanced understanding that was confirmed by error analysis and visualization, but there were still slight difficulties with the separation of semantically close classes. Finally, this work demonstrates that transformer-based NLP models can be effective in enterprise-scale text classification and points to the potential to build on innovation by adapting to a domain, providing interpretability, and scalable deployment, heralding smarter, context-aware enterprise solutions in the emerging digital environment.