Social Media Summarization at Scale: How Far Small and Open LLMs Match-Up Against Commercial LLMs?

ACL ARR 2025 February Submission2178 Authors

Abstract: Summarizing Social media content is complex due to its unique language specifics, and the scale and speed of content generation. Recent developments in Large Language Models (LLMs) enable text summarization with unprecedented accuracy; however, their high computational cost and input length constraints remain major obstacles for real-world applications at scale. Several approaches based on Small Language Models (SLMs) and Open source alternatives exist that present affordable alternatives for managing computational complexity in practice. In this paper, we explore popular SLMs and open LLMs for long-context summarization applied to Social media, and evaluate their performance using two datasets comprising of long social media discussions collected online. The key findings show that fine-tuning smaller models and optimizing input selection can achieve high-quality summarization at significantly lower computational costs.

Paper Type: Short

Research Area: Summarization

Research Area Keywords: long-form summarization, benchmarking, evaluation, quantitative analyses of news and/or social media;

Contribution Types: NLP engineering experiment, Approaches low compute settings-efficiency

Languages Studied: English

Submission Number: 2178