SS-GEN: A Social Story Generation Framework with Large Language Models

• DOI: 10.1609/aaai.v39i2.32119
• Publication Date: 2025-04-11T09:41:56Z
• Authors: Yi Feng; Mingyang Song; Jiaqi Wang; Zhuang Chen; Guanqun Bi; Minlie Huang; Liping Jing; Jian Yu

Children with Autism Spectrum Disorder (ASD) often misunderstand social situations and struggle to participate in daily routines. Social Stories™ are traditionally crafted by psychology experts under strict constraints address these challenges but costly limited diversity. As Large Language Models (LLMs) advance, there's an opportunity develop more automated, affordable, accessible methods generate Stories real-time broad coverage. However, adapting LLMs meet the unique of is a challenging issue. To this end, we propose SS-GEN, Story GENeration framework LLMs. Firstly, constraint-driven sophisticated strategy named StarSow hierarchically prompt at scale, followed rigorous human filtering build high-quality dataset. Additionally, introduce quality assessment criteria evaluate effectiveness generated stories. Considering that powerful closed-source large models require very complex instructions expensive API fees, finally fine-tune smaller language our curated dataset, achieving comparable results lower costs simpler instruction deployment. This work marks significant step leveraging AI personalize cost-effectively for autistic children which hope can encourage future research on special groups.