---
pretty_name: LLM-Profiling
tags:
- evaluation
- benchmark
- facts
- datasets
license: mit
thumbnail: llm_prof.jpg
size_categories:
- 10K<n<100K
---
<p align="center">
  <img src="./assets/llm_prof.jpg" width="420">
</p>

<h1 align="center">Knowledge Graph-Based Dynamic Factuality Evaluation</h1>
<h3 align="center">Worldview Benchmark for Large Language Models</h3>


> <sub>The authors do not endorse any political, ideological, or moral position implied by the items or by model outputs.
> All entries are probes for factual and value-related reasoning and must not be used to profile real people or to justify harmful actions or decisions.</sub>

## Dataset Description

Profiling: Knowledge Graph-Based Dynamic Factuality Evaluation is a research benchmark for studying how large language models handle facts in a dynamic knowledge graph, including retrieving, updating, and verifying them under changing context.

- **Goal:** evaluate how well the LLM can reproduce verifiable facts and whether it maintains the integrity of its "beliefs" when the data set changes.
- **Data source:** Wiki, Gre
- **Size:** ≈19K multiple-choice questions
- **Format:** JSON

<h1 align="left">Category:</h1>
<img src="./assets/stat.jpg" width="500">

### Supported Tasks
Tasks this dataset is suitable for:
- 🧠 **Factual Knowledge Evaluation:** Measuring factual accuracy and knowledge retrieval capabilities across domains (geography, history, science, culture).
- 🧩 **Knowledge Graph Reasoning:** Evaluating ability to infer relationships between entities and navigate structured knowledge.

### Languages
Primary language of instructions/prompts: `ru` (Russian), `en` (English).

---

## Dataset Structure

Example of a single instance:

```json
{
    "task": "What body of water is near the birthplace of Marcelo Romero?",
    "option_1": "River Gironde",
    "option_2": "Hamoaze",
    "option_3": "River Plate",
    "option_4": "Haring river",
    "correct": 3,
    "meta_difficulty": "medium",
    "meta_type": "single",
    "meta_origin": "wikidata"
}
```


## 📚 Citation

If you use LLM-Profiling in your work, please cite it as:

```bibtex
@misc{stonic_worldview_benchmark_2025,
  title        = {LLM-Profiling: A Worldview Benchmark for Large Language Models},
  author       = {Andrey Chetvergov — chetvergov-as@ranepa.ru
                  Rinat Sharafetdinov — sharafetdinov-rs@ranepa.ru
                  Stepan Ukolov — ukolov-sd@ranepa.ru
                  Timofei Sivoraksha — sivoraksha-ta@ranepa.ru
                  Alexander Evseev — aevseev-23-01@ranepa.ru
                  Danil Sazanakov — hdystasyfibkv@gmail.com
                  Sergey Bolovtsov — bolovtsov-sv@ranepa.ru},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/llmpass-ai/stonic_dataset}},
}
```