Architecture for Low-Resource Data Analysis

DysLexLens addresses the challenge of extracting meaningful insights from sparse, noisy social media data regarding the lived experiences of dyslexic learners using AI. The framework operates as an end-to-end pipeline designed to transform unstructured forum posts into verifiable, knowledge-grounded insights.

Key components include:

• Dictionary-Driven Filtering: To combat noise in low-resource contexts, the system uses a specialized dictionary to filter Reddit posts, ensuring the corpus remains focused on the intersection of dyslexia and AI.
• KG-Based Reasoning: The framework integrates LLM-assisted semantic analysis with Knowledge Graphs (KG) to perform structured query reasoning, allowing for more reliable pattern discovery than standard retrieval alone.

Verification and Evaluation Rigor

Because the framework targets sensitive user experiences, it prioritizes evidence alignment and hallucination mitigation. It employs a dual-layered evaluation approach:

• Quantitative Metrics: The system utilizes RAGAS and Query Robustness metrics to benchmark the performance of LLM-generated responses against the source data.
• Qualitative Validation: The authors provide structured guidelines for human-grounded assessment, specifically focusing on whether the generated insights are accurately supported by the original forum evidence.

The authors demonstrated the framework's efficacy using 30 specific queries on dyslexia-related Reddit data, providing a reproducible baseline for researchers looking to apply similar architectures to other low-resource community datasets.