mrince/CBRT-MPC-SummaryReports

---

license: mit task_categories: - text-classification language: - en tags: - central-bank - monetary-policy - sentiment-analysis - hawkish-dovish pretty_name: CBRT MPC Summary Reports (Hawkish/Dovish/Neutral Labels) size_categories: - 1K<n<10K

This dataset contains 7,200 sentences from the CBRT (Central Bank of the Republic of Türkiye) Monetary Policy Committee (MPC) Summary Reports. Each sentence has been labeled as "hawkish", "dovish", or "neutral" based on majority voting from three large language models (GPT-4, Gemini, DeepSeek).

The dataset is balanced across the three sentiment classes and is suitable for fine-tuning sentiment classifiers or monetary policy tone analysis.

Labels:

• neutral - [0]: informational or balanced tone without clear policy stance
• hawkish - [1]: tightening bias, upward rate signal
• dovish - [2]: easing bias, downward/inflation-tolerant tone

Used in: mrince/CBRT-RoBERTa-HawkishDovish-Classifier

The following common prompt was used in the artificial intelligence labeling phase.

prompt = f"""
You are an expert in monetary policy communication. Classify the following sentence from the Central Bank of the Republic of Turkey (CBRT) based on its policy tone.

Sentence: "{sentence}"

Choose only one of the following labels:

- "hawkish": if the sentence signals inflationary pressure, interest rate hikes, tightening, or economic risks requiring policy action.
- "dovish": if the sentence signals easing, financial support, rate cuts, or declining inflation.
- "neutral": if the sentence is purely descriptive with no policy implications.

   Return only the label as a single word: hawkish, dovish, or neutral.
"""