Acknowledging Valued Perspectives! User-Assisted Summarization by Leveraging Opinions and LLMs

In today’s data-rich world, leveraging user comments for summarization is increasingly valuable. However, classifying these comments specifically for summarization remains a novel focus of this work. To address this, we propose theOpinion-AugmentedClassification andSummarizationFramework (OACSF), which categorizes comments into three types: complementary, supplementary, and detached, based on their stance and alignment with the article. The framework employs BART for comment classification, with cosine similarity and GPT-assisted relevance scoring used as baseline methods.We conduct experiments with five Large Language Models (LLMs): two models with expansive output token windows and three with constrained capacities requiring chunk-based segmentation. This diverse setup enables systematic analysis of how model architecture and token limitations influence opinion-based summarization. The generated summaries are evaluated using automatic readability metrics and validated through detailed human assessment. Our implementation and dataset processing code are available at the following anonymous GitHub repository:https://github.com/Raghvendra-14/Comments-Categorization-and-Summarization.