CB-SentiLex: Bangladesh Bank Monetary Policy Stance Detection Benchmark

CB-SentiLex: Bangladesh Bank Monetary Policy Stance Detection Benchmark Version: 1.0DOI: 10.5281/zenodo.21337423License: MITPaper: "CB-SentiLex: An Auditable Weak-Supervision Framework for Central Bank Stance Detection with a Bangladesh Bank Benchmark" OVERVIEW--------This release contains the complete reproducibility package for CB-SentiLex, including the classified corpus, curated lexicon, trained transformer models, and three-annotator calibration set for Bangladesh Bank Monetary Policy Statements (2006-2026). CONTENTS--------data/ bb_sentences_stance.csv 7,432 classified sentences bb_document_stance.csv 39 document-level stance scores calibration_300.csv 300-sentence, 3-annotator calibration set lexicon/ dcs_lexicon.json 75-term CB-SentiLex lexicon (34 hawkish, 41 dovish) models/ distilbert/ Fine-tuned DistilBERT checkpoint config.json model.safetensors tokenizer.json tokenizer_config.json scripts/ Reproducibility scripts dcs_labeling.py Core labeling function train_classifier.py Transformer fine-tuning error_analysis.py Error analysis policy_decision_validation.py Policy-decision correlation external_validation_bootstrap_cis.py Bootstrap confidence intervals falsification_tests.py Falsification checks monthly_fx_validation.py Monthly FX correlation forward_macro_validation.py Annual macro correlation temporal_stability.py Temporal split evaluation adversarial_testing.py Adversarial robustness publication_figures.py Figure generation LICENSEREADME.mdcitation.cff FILE DESCRIPTIONS----------------- Corpus bb_sentences_stance.csv: 7,432 sentence-level classifications with weak labels, neural model predictions, and document metadata. Columns: sentence_id, text, source_document, period, language, doc_type, page, stance_score, stance_label, pred_hawkish, pred_dovish, pred_neutral, predicted_label. bb_document_stance.csv: 39 document-level stance scores aggregated from sentence labels. Columns: bank, doc_id, date, year, total_sentences, hawkish_count, dovish_count, neutral_count, stance_score. Calibration Set calibration_300.csv: 300 sentences annotated by three independent annotators (Persons 1-3) with labels, confidence levels, notes, and consensus status. Includes weak labels for comparison. Blind two-annotator Fleiss' kappa = 0.204; full three-annotator kappa = 0.231. Lexicon dcs_lexicon.json: The CB-SentiLex-Bengali lexicon with 75 terms (34 hawkish, 41 dovish) organized into four semantic families: rate keywords, inflation keywords, stance keywords, and policy keywords. Format: JSON with hawkish and dovish arrays. Model distilbert/: Fine-tuned distilbert-base-uncased checkpoint for three-class stance classification (HAWKISH, DOVISH, NEUTRAL). Trained on the weakly labeled BB corpus. Benchmark performance: ~0.87 macro-F1 on the fixed test split. QUICK START-----------import pandas as pdimport jsonfrom transformers import AutoModelForSequenceClassification, AutoTokenizer # Load sentence-level datadf = pd.read_csv("data/bb_sentences_stance.csv") # Load lexiconwith open("lexicon/dcs_lexicon.json") as f: lexicon = json.load(f) # Load calibration setcal = pd.read_csv("data/calibration_300.csv") # Load trained modelmodel = AutoModelForSequenceClassification.from_pretrained("models/distilbert")tokenizer = AutoTokenizer.from_pretrained("models/distilbert") # Classify a sentenceinputs = tokenizer("BB has decided to reduce the policy rate by 50 basis points", return_tensors="pt", truncation=True, max_length=128)outputs = model(**inputs)predicted = ["HAWKISH", "DOVISH", "NEUTRAL"][outputs.logits.argmax(-1).item()] CITATION--------@misc{nabil2026cbsentilex, title={CB-SentiLex: An Auditable Weak-Supervision Framework for Central Bank Stance Detection with a Bangladesh Bank Benchmark}, author={Ann Naser Nabil and Umme Hafsa}, year={2026}, publisher={Zenodo}, doi={10.5281/zenodo.21337423}, url={https://zenodo.org/records/21337423}} LICENSE-------MIT License. CONTACT-------Ann Naser Nabil: [e-mail] Hafsa: [e-mail]