برمجية
DTI Coder: bilingual measurement of digital technology integration in annual-report narrative (v1.1.0)
Ahmad Alomari, Anwar Allah Pitchay
الملخص
Adds a firm-grouped 5-fold cross-validation of the presence model over all 1,648 coded, non-artifact, non-duplicate sentences (88 firms, 259 positives), with the same settings as the held-out run. New files: training/cv_grouped.py results/results_cv_grouped.json (per-fold results; pooled out-of-fold F1 with 2,000-resample bootstrap CIs) results/pred_cv_oof.csv (out-of-fold predictions for all 1,648 sentences) Pooled out-of-fold F1 (95% bootstrap CI): English (n 829, 149 positives): hybrid 0.857 [0.811, 0.898]; model 0.828 [0.777, 0.872]; dictionary 0.734 [0.684, 0.777] Arabic (n 819, 110 positives): hybrid 0.833 [0.777, 0.880]; model 0.815 [0.757, 0.864]; dictionary 0.712 [0.651, 0.768] Hybrid minus dictionary: 0.123 [0.071, 0.181] English; 0.121 [0.067, 0.176] Arabic. All v1.0.0 files are unchanged. The released model weights are unchanged; the fold models are for evaluation only and are not released. Model weights: https://huggingface.co/ahmadomari/dti-coder-xlmr Live app: https://dti-coder.streamlit.app Ahmad Alomari and Anwar Allah Pitchay, School of Management, Universiti Sains Malaysia.
التوثيق
Alomari, A., & Pitchay, A. A. (2026). DTI Coder: bilingual measurement of digital technology integration in annual-report narrative (v1.1.0).. https://doi.org/10.5281/zenodo.22918276