Codebase & Experiments

Paper: CIKM Short Paper - Knowing When to Correct: Cost-Aware LLM Routing
Authors: Stergios Konstantinidis, Hayman Lotfy, Michalis Vlachos (University of Lausanne)
Venue: ACM International Conference on Information and Knowledge Management (CIKM) - Short Paper
GitHub: https://github.com/Stergios-Konstantinidis/CIKM_public
Overleaf: https://git.overleaf.com/6a1d78e803cbdf7def32a839

Codebase Guide: Optimizing-LLM-based-OCR-Corrections

1. Directory Structure

code/experiments/:

confbert_router.py: ConfBERT token-level scoring and routing pipeline.

experiment_linear_regression.py: Lasso and Ridge feature selection and cross-validation.

experiment_svm_classifier.py: Support Vector Machine routing.

experiment_nn_regression.py: Neural network router.

lazy_clf_scan.py: AutoML screening of 30+ regression algorithms.

code/plotting/:
plot_routing_frontier_paddle.py: Generates CER vs Cost Pareto curves. plot_threshold_sweep.py: Threshold sensitivity analysis. plot_error_confidence_cer.py: Correlation between OCR confidence and CER. code/evaluation/:
run_evaluations.py: Batch evaluation harness across Tesseract, EasyOCR, and PaddleOCR. data/:
evaluation_dataset/groundtruth.json: Ground truth transcriptions for 609 segments. raw_ocr_results.json: Output dumps from Tesseract, EasyOCR, and PaddleOCR.