TISEA: a scalable deep learning framework for multi-faceted text analytics with topic modeling, summarization, and emotion classification
H. Sabireen, Riddhirup Bera, Abdul Quadir, Christy Jackson Joshua, A. Shahina · Engineering Research Express · 2025
Abstract In today’s fast-paced, information-driven world, the vast amount of readily available data poses significant challenges, particularly when individuals have limited time to process and absorb it. The need for efficient systems that can analyze, classify, and summarize textual information has therefore become increasingly crucial. To address this challenge, we propose the Topic Identification, Summarization, and Emotion Analysis (TISEA) system. This architecture streamlines the extraction of key topics, generates concise summaries, and performs emotion analysis on textual data, making information more accessible and manageable. TISEA employs a hybrid approach to topic identification, combining a supervised model based on BERT (Bidirectional Encoder Representations from Transformers) with an unsupervised method, Latent Dirichlet Allocation (LDA), for effective topic modeling. Text segregation is achieved through a cosine similarity algorithm applied to BERT embeddings to identify and group the most relevant sentences. Emotion analysis is integrated using a custom BERT-based framework, which provides insights into the emotional tone of the text. The system is trained on the Text Summarization Data (TSD) dataset from Kaggle, which contains news articles and their corresponding headlines as summaries. The TISEA system demonstrates strong performance in text summarization, achieving an accuracy of 75.86% and an average Bilingual Evaluation Understudy (BLEU) score of 0.622. For topic identification, the supervised model achieves 99.24% accuracy, while the emotion analysis model attains 99.62% accuracy, both outperforming state-of-the-art baselines such as Support Vector Machine (SVM), Decision Tree (DT), and Gaussian Naive Bayes (GNB) for topic identification, and K-Nearest Neighbors (KNN), SVM, and Naive Bayes (NB) for emotion analysis. Overall, the TISEA system offers a powerful, efficient, and highly accurate solution for analyzing and summarizing large-scale textual data. Its applications span multiple domains, including news aggregation, academic research, and sentiment analysis in social media, thereby providing individuals and organizations with an effective tool to manage and utilize information more effectively.