PharosAIFactory/Wiki_Greek_Personalities_Multiple_Choice

Wiki Greek Personalities Multiple Choice Dataset

Developed by NCSR-D, this dedicated multiple-choice evaluation dataset focuses on expanding LLM knowledge regarding Greek public figures, enhancing domain knowledge, and evaluating the ability of language models to select accurate biographical data on prominent Greek personalities.

The dataset benchmarks multi-choice reasoning tasks by offering a structured setup containing target questions, accurate answers, and hard, synthetically generated incorrect distractors.

---

Dataset Statistics & Structure

• Size: 1,650 multiple-choice questions.
• Coverage: 170 unique Greek public figures.
• Format: Uploaded from a line-by-line JSONL structure.
• Domains Covered: Politics, science, academia, culture, arts, sports, business, and public life.

Schema Blueprint

Each entry contains the following features:

• title (string): Target subject's identity name.
• question (string): Natural language multiple-choice query in Greek.
• choices (list of strings): A 4-option array containing three incorrect distractors and one correct option.
• correct_index (int): The zero-indexed position (0, 1, 2, or 3) corresponding to the true answer inside the choices list.
• answer (string): The explicit, long-form true answer text.

---

Dataset Curation Methodology

1. Source Context & Extraction

NCSR-D identified and targeted 16,075 unique Wikipedia pages matching Greek individuals. The system extracted specific Wikipedia person properties directly from a recent dump to ensure up-to-date and robust context mappings. A highly curated subset focused on 170 Greek public figures was preserved to construct this benchmark.

2. Multi-Choice Generation

The language model google/gemma-3-12b-it (Gemma Team, 2025) was leveraged to process the core question-answer data. For each question, the model generated three incorrect Greek answers (distractors) that differ contextually from each other and from the correct answer. This systematically constructed a robust multiple-choice layout with 4 choices.

3. Dual-LLM Quality Control & Filtering

To ensure exceptional data quality and linguistic consistency, a validation execution was performed using google/gemma-4-12B (Gemma Team, 2026). The evaluation system assessed both factual correctness against source text and linguistic compliance (measuring grammar, syntax, phrasing clarity, and overall Greek structural correctness). Pairs that did not satisfy these high standards were removed prior to generating the final evaluation configurations.

---

Loading and Usage

You can load this dataset natively using the Hugging Face datasets package:

from datasets import load_dataset

# Load the multiple-choice training/evaluation split
dataset = load_dataset("PharosAIFactory/Wiki_Greek_Personalities_Multiple_Choice")
print(dataset["train"][0])

Parsing an Instance Example

{
  "title": "Person",
  "question": "Πότε και πού γεννήθηκε ο Person;",
  "choices": [
    "Σπούδασε Πολιτικές Επιστήμες και Διεθνείς Σχέσεις στο Πανεπιστήμιο του Κεντ στις 10 Νοεμβρίου 1968, όπου και γεννήθηκε.",
    "Ο Person γεννήθηκε στις 10 Νοεμβρίου 1968 στον Βύρωνα.",
    "Έγινε δήμαρχος Βύρωνα το 2003, έχοντας γεννηθεί στον Βύρωνα στις 10 Νοεμβρίου 1968.",
    "Γεννήθηκε στην Γρανίτσα Ευρυτανίας, από όπου κατάγεται, στις 10 Νοεμβρίου 1968."
  ],
  "correct_index": 1,
  "answer": "Ο Person γεννήθηκε στις 10 Νοεμβρίου 1968 στον Βύρωνα."
}

Additional Information

• Curator Organization: NCSR-D / PharosAIFactory