Assessing the Feasibility of Processing a Paper-based Multilingual Social Needs Screening Questionnaire Using Artificial Intelligence v2
This repo contains the results from assessing GPT-3.5 Turbo on the training synthetic examples ("/evals/") and from comparing models ("/comparisons/") on the two testing datasets (synthetic and scanned). The LLM calls and responses are also included.
Links to notebooks (HTML) referrenced in Supplementary Appendix:
- Analyzing Synthetic Dataset: https://oekekezie.github.io/poc-ai-sdoh-questionnaire-v2/html/Analyzing%20Synthetic%20Dataset.html
- Analyzing Scanned Dataset: https://oekekezie.github.io/poc-ai-sdoh-questionnaire-v2/html/Analyzing%20Scanned%20Dataset.html
- Comparing Performance on Synthetic Testing Dataset: https://oekekezie.github.io/poc-ai-sdoh-questionnaire-v2/html/Comparing%20Performance%20on%20Test%20Dataset.html
- Comparing Performance on Scanned Testing Dataset: https://oekekezie.github.io/poc-ai-sdoh-questionnaire-v2/html/Comparing%20Performance%20on%20Scanned%20Dataset.html