Transforming Hindi Text Summarization: A PGGNN Approach with Enhanced Preprocessing and Comparative Evaluation

This article introduces a novel methodology for the summarization of Hindi text, leveraging the advanced capabilities of the Pointer-Generator Gated Neural Network (PGGNN) model. Prior to the implementation of the model, a comprehensive series of preprocessing steps is executed to significantly enhance the quality of the input text. These preprocessing measures involve the removal of emojis, thorough text cleaning, handling of hashtags, filtering out special characters, minimizing gaps, and overall improvement of the Hindi text's linguistic structure. The culmination of these protocols results in the transformation of the original 'article' within the data frame into a refined 'cleaned article' column, poised for subsequent in-depth analysis.The summarization process is meticulously organized into two distinct phases. In Phase 1, unsupervised methods are employed, including stemming, suffix stripping, the creation of a Document Term Matrix (DTM), sentence scoring, sorting, selection of the top 30% of sentences, and the compilation of extractive summaries. Phase 2 introduces the proposed PGGNN paradigm, encompassing initialization, architectural considerations, and the execution of a forward pass.To evaluate the performance of the summarization models, a comparative analysis is conducted, specifically pitting the PGGNN against the VGG 16 model. Metrics such as Rouge-1, Rouge-2, Rouge-L, and BLEU Score are employed in this assessment. Notably, the PGGNN approach outshines existing methods, achieving a Rouge-1 score of 76.89, a Rouge-2 score of 59.24, a Rouge-L score of 49.61, and an outstanding BLEU Score of 81.53.This research signifies a significant advancement in the domain of Hindi text summarization, offering a meticulous approach that not only involves cutting-edge neural network models but also places emphasis on robust preprocessing techniques, ultimately resulting in superior summarization performance compared to established models.