DysLexLens Analyzes Dyslexic Learners' AI Experiences from Forums

While dyslexic learners increasingly use AI tools for support, their actual experiences and insights with these technologies remain largely unexplored. This paper introduces DysLexLens, a specialized low-resource large language model (LLM) framework developed to systematically analyze discussions from online forums, specifically Reddit, to understand how dyslexic individuals interact with and perceive AI tools. DysLexLens is an end-to-end architecture designed for evidence traceability. It begins by employing a dictionary-driven filtering method to create a focused corpus from noisy social media data, ensuring higher relevance for low-resource contexts. It then integrates LLM-assisted semantic analysis with knowledge-graph (KG)-based query reasoning to identify meaningful patterns and generate verifiable responses to specific questions. The framework includes quantitative evaluation metrics like RAGAS and Query Robustness to measure LLM response performance, alongside structured qualitative validation guidelines to assess response quality, particularly focusing on hallucination and evidence alignment. Demonstrated on dyslexia-related Reddit data with 30 questions, DysLexLens showed effectiveness and potential generalizability to other low-resource forum contexts, offering a valuable tool for understanding user experiences in niche communities.