Seed-mediated AuNR Synthesis Extraction with GPT-3

The files were used as follows:

Empty synthesis template:

 clean_template_formatted.json 

Training sets for fine-tuning GPT-3 to extract structured seed-mediated AuNR growth procedures from unstructured text:

aunr_synth_v6_ucfx_static_0_240_0_corrected_publishable.json

Predictions and corrected predictions for evaluating the performance of GPT-3 recipe extraction completions:

aunr_synth_v6_ucfx_static_0_40_6_predict_publishable.json
aunr_synth_v6_ucfx_static_0_40_6_corrected_publishable.json

GPT-3 recipe extraction completion prediction over the complete dataset:

aunr_synth_v6_ucfx_static_0_1137_0_predict_publishable.json

GPT-3 recipe extraction completion prediction over the complete dataset merged by paper with empty fields removed:

aunr_synth_v6_ucfx_static_dynamic_0_1137_0_predict_paperwise_merged_publishable.json

GPT-3 recipe extraction completion prediction over the complete dataset merged by paper with empty fields removed and post-processing applied:

aunr_synth_v6_ucfx_static_dynamic_0_1137_0_predict_paperwise_merged_postprocessed_n_recipes_publishable.json

Performance evaluation scripts:

prediction_statistics_complete.py
paperwise_prediction_statistics_complete.py
paragraphwise_prediction_statistics_complete.py