JRQi/DeepResearch-Bench-Multilingual

---
configs:
- config_name: all
default: true
data_files:
- split: test
path: data/deepresearch_bench_prompts_multilingual.csv
- config_name: source_prompt
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_source_prompt.csv
- config_name: en
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_en.csv
- config_name: zh
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_zh.csv
- config_name: es
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_es.csv
- config_name: it
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_it.csv
- config_name: ar
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_ar.csv
- config_name: bn
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_bn.csv
- config_name: ja
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_ja.csv
- config_name: el
data_files:
- split: test
path: data/subsets/deepresearch_bench_prompts_el.csv
license: apache-2.0
language:
- en
- zh
- es
- it
- ar
- bn
- ja
- el
task_categories:
- text-generation
- question-answering
- translation
size_categories:
- n<1K
tags:
- benchmark
- deep-research
- multilingual
- translation
pretty_name: DeepResearch Bench Multilingual Prompts
source_datasets:
- muset-ai/DeepResearch-Bench-Dataset
---

# DeepResearch Bench Multilingual Prompts

This dataset provides prompt-level multilingual translations for the 100 research tasks used in [muset-ai/DeepResearch-Bench-Dataset](https://huggingface.co/datasets/muset-ai/DeepResearch-Bench-Dataset).

The translations cover eight languages:

- `en`
- `zh`
- `es`
- `it`
- `ar`
- `bn`
- `ja`
- `el`

## What is included

This repository focuses on the benchmark prompts only.

On the Hugging Face Hub, the Dataset Viewer is configured with one default subset named `all` plus nine explicit subset configurations: `source_prompt`, `en`, `zh`, `es`, `it`, `ar`, `bn`, `ja`, and `el`. Each subset currently exposes a single split named `test`, which is expected because each configuration maps to one file.

Files:

- `data/deepresearch_bench_prompts_multilingual.jsonl`
- `data/deepresearch_bench_prompts_multilingual.csv`
- `data/subsets/deepresearch_bench_prompts_source_prompt.jsonl`
- `data/subsets/deepresearch_bench_prompts_source_prompt.csv`
- `data/subsets/deepresearch_bench_prompts_en.jsonl`
- `data/subsets/deepresearch_bench_prompts_en.csv`
- `data/subsets/deepresearch_bench_prompts_zh.jsonl`
- `data/subsets/deepresearch_bench_prompts_zh.csv`
- `data/subsets/deepresearch_bench_prompts_es.jsonl`
- `data/subsets/deepresearch_bench_prompts_es.csv`
- `data/subsets/deepresearch_bench_prompts_it.jsonl`
- `data/subsets/deepresearch_bench_prompts_it.csv`
- `data/subsets/deepresearch_bench_prompts_ar.jsonl`
- `data/subsets/deepresearch_bench_prompts_ar.csv`
- `data/subsets/deepresearch_bench_prompts_bn.jsonl`
- `data/subsets/deepresearch_bench_prompts_bn.csv`
- `data/subsets/deepresearch_bench_prompts_ja.jsonl`
- `data/subsets/deepresearch_bench_prompts_ja.csv`
- `data/subsets/deepresearch_bench_prompts_el.jsonl`
- `data/subsets/deepresearch_bench_prompts_el.csv`
- `build_multilingual_deepresearch_bench.py`

The combined multilingual files contain:

- `id`: benchmark prompt id
- `source_language`: original source language in the upstream benchmark (`zh` for ids 1-50, `en` for ids 51-100)
- `source_prompt`: the original upstream prompt text
- `en`, `zh`, `es`, `it`, `ar`, `bn`, `ja`, `el`: prompt translations

Each subset file contains exactly three columns:

- `id`
- `source_language`
- one text column only: `source_prompt` or one of `en`, `zh`, `es`, `it`, `ar`, `bn`, `ja`, `el`

This makes it easy to load per-language subsets independently while preserving prompt ids and source-language provenance.

## Translation method

The translations were written prompt-by-prompt by GPT-5.4, with the goal of preserving the research intent, technical terminology, and task structure of each benchmark query. No external machine translation service was used for the translated prompt fields in this repository.

## Provenance

Source benchmark:

- [muset-ai/DeepResearch-Bench-Dataset](https://huggingface.co/datasets/muset-ai/DeepResearch-Bench-Dataset)

This multilingual prompt set is a derived artifact built from the upstream prompt list in:

- `generated_reports/openai-deepresearch.jsonl`

The prompt ordering and ids match the canonical 100-task benchmark.

## Rebuild

To regenerate the local JSONL and CSV files:

```bash
python3 build_multilingual_deepresearch_bench.py
```

## License

This derived prompt translation set is released under Apache 2.0, consistent with the upstream dataset license.