wcag_AI_validation/scripts/esercitazione_12_2025

Folder structure

• analisi_esercitazione_12_2025 first notebook that includes similarities calculation and some basic EDA analysis
• analisi_esercitazione_12_2025_advanced the notebook contains the analysis on language switch
• analisi_esercitazione_12_2025_embedding the notebook builds a classifier on the semantic representation of the text generated by the LLM
• analisi_esercitazione_12_2025_classificatore the notebook builds a classifier based on some features related to user-LLM alt text similarities, text readibility, etc.
• analisi_esercitazione_ricostruzione_associazioni the notebook rebuilds the exercise dataset based on the DB dumbs from UI and backend microservices
• analisi_esercitazione_12_2025_models_comparisons starting from the rebuild exercise dataset, the notebook performs some test switching LLM model and/or prompt (also english translation)
• analisi_esercitazione_12_2025_build_full_dataset rerun all the features building using pandas apply
• analisi_esercitazione_12_2025_clip run CLIP score calculation between image and alt-text
• analisi_esercitazione_12_2025_inter_user_agreement calculate inter-user agreements and inter LLM runs agreements
• analisi_esercitazione_12_2025_distributions_comparison perform some indicator calculations to compare two candidates distrubutions with a reference one
• analisi_esercitazione_12_2025_classificatore_LLM Evaluate LLM classifier performance on the original alt-text assessment (0-1 classification problem)