Datasets:

ZhikaiXu24
/

DAS-Bench

Category stringclasses 4 values	Method stringclasses 10 values	BSC_Support float64 2.83 4	BSC_Attribution float64 2.6 4.03	BSC_Multi_Reference_Synthesis float64 1.93 4	BSC_Balance float64 2.03 4.07	BSC_Avg float64 2.75 3.85	TSQ_Coverage float64 3 4.23	TSQ_Boundary_Control float64 3 4.3	TSQ_Organization float64 3 4.4	TSQ_Insight float64 2.97 4.5	TSQ_Avg float64 2.99 4.29	HDQ_Alignment float64 3.43 4.47	HDQ_Progression float64 2.43 4.13	HDQ_Specificity float64 3.07 4.53	HDQ_Local_Synthesis float64 2.07 4.17	HDQ_Avg float64 2.75 4.28	MAR_References float64 3 5	MAR_Visuals float64 1.29 5	MAR_Layout float64 4.83 5	MAR_Completeness float64 3 5	MAR_Avg float64 3.07 5	Total_Avg float64 3.18 4.34
Reference	Human	3.87	3.9	3.53	4.07	3.84	4.23	4.23	4.4	4.3	4.29	4.47	4.07	4.37	4.07	4.24	5	5	5	5	5	4.34
General	Codex	3	3.43	2.73	2.03	2.8	3	3	3	2.97	2.99	3.43	2.43	3.07	2.07	2.75	5	1.77	5	5	4.19	3.18
General	GPT Deep Research	3.57	3.77	3.33	2.6	3.32	3.47	3.5	3.53	3.43	3.48	4.1	3.47	4	3.47	3.76	5	1.57	5	5	4.14	3.68
General	Gemini Deep Research	2.83	2.6	2.87	3.5	2.95	3.8	3.83	3.83	3.8	3.82	4.17	3.97	4.17	3.97	4.07	4.6	4.87	4.9	5	4.84	3.92
General	Naive RAG	4	4	3.33	3.6	3.73	3.97	4.3	4.2	3.77	4.06	4.47	3.93	4.53	3.93	4.22	5	1.37	5	5	4.09	4.03
Auto Survey Generation	AutoSurvey	4	4	3.29	3.95	3.81	4	3.86	3.1	4	3.74	4.1	3.14	4.24	3.29	3.69	5	1.29	5	3.38	3.67	3.73
Auto Survey Generation	SurveyForge	4	3.9	3.81	3.19	3.73	4	4	3.24	4	3.81	4.1	3.29	4.19	3.38	3.74	5	1.29	5	4.05	3.83	3.78
Auto Survey Generation	LiRA	3.97	3.97	3.57	3	3.63	4	3.33	3.53	4	3.72	3.97	3.9	4.43	3.93	4.06	3	1.4	4.87	3	3.07	3.62
Auto Survey Generation	InteractiveSurvey	3.1	3	1.93	2.97	2.75	4	3.7	3.83	4	3.88	3.8	3.93	4	4	3.93	4.13	4.77	4.83	5	4.68	3.81
Proposed	DAS	4	4.03	4	3.37	3.85	4	4	4.37	4.5	4.22	4.27	4.13	4.53	4.17	4.28	5	5	5	5	5	4.34

DAS-Bench

DAS-Bench is a 30-topic multi-domain benchmark for automatically generated academic surveys. It is paired with DAS-Eval, a 16-criterion evaluation suite for publication-oriented academic surveys covering scholarly citation, taxonomic synthesis, hierarchical discourse, and manuscript reliability.

Paper: Deep Academic Survey
Project page: DAS
Source and evaluation toolkit: ZhikaiXu24/DAS
Literature metadata lake: ZhikaiXu24/DAS-2M

This repository is the Hugging Face distribution of the current public DAS-Bench release. The GitHub repository remains the canonical source for the evaluation implementation and development history.

Load with `datasets`

from datasets import load_dataset

topics = load_dataset(
    "ZhikaiXu24/DAS-Bench",
    "topics",
    split="benchmark",
)

results_30 = load_dataset(
    "ZhikaiXu24/DAS-Bench",
    "results_30_topics",
    split="evaluation",
)

results_cs21 = load_dataset(
    "ZhikaiXu24/DAS-Bench",
    "results_cs21",
    split="evaluation",
)

Dataset configurations

Configuration	Rows	Contents
`topics`	30	Frozen benchmark topic identifiers and research topics.
`results_30_topics`	10	Criterion-level aggregate results for the full 30-topic benchmark.
`results_cs21`	10	Criterion-level aggregate results for the matched 21-topic CS subset.

The matched CS subset is not a separate benchmark split. It contains the 21 computer science topics shared by every evaluated system and supports comparison under identical topic coverage.

Current release scope

The current public release includes:

the frozen 30-topic list;
the benchmark task definition and metadata schema;
the BSC, MAR, TSQ, and HDQ metric definitions;
the DAS-Eval implementation and safe placeholder configuration;
the public aggregate results; and
qualitative comparison materials.

This release does not include the fixed 300-paper candidate pool for each topic, survey-generation code, formal submission-format documentation, private experiment artifacts, or model outputs. Candidate pools may be added in a future release without changing the current topics and results configurations.

Benchmark task

Each complete benchmark instance defines a research topic and a fixed pool of 300 candidate-paper metadata records. The 30 topics span language models, computer vision, robotics, scientific discovery, security, medicine, climate, remote sensing, finance, and uncertainty estimation.

See benchmark/task_specification.md for task boundaries and benchmark/metadata_schema.json for the candidate-paper metadata schema.

DAS-Eval

DAS-Eval defines four complementary metric families, each containing four criteria scored from 1 to 5:

Balanced Scholarly Citation Quality (BSC)
Manuscript Artifact Reliability (MAR)
Taxonomic Synthesis Quality (TSQ)
Hierarchical Drafting Quality (HDQ)

The reported Total Avg. is the arithmetic mean of all 16 criteria. Complete definitions and aggregation rules are provided in benchmark/evaluation_protocol.md.

Evaluation toolkit

The public toolkit is available under evaluation/:

eval_prepare.py: prepares Markdown and rendered page images from survey PDFs;
eval_bsc.py: evaluates BSC;
eval_mar.py: evaluates MAR;
eval_tsq_hdq.py: evaluates TSQ and HDQ; and
run_eval_all.sh: runs the evaluation stages serially.

Install the common dependencies with:

python -m pip install -r requirements.txt

Configure the hosted or local OpenAI-compatible judge in config.json. API credentials must be supplied through environment variables and must never be committed. PDF preprocessing uses MinerU, which must be installed separately.

Qualitative comparison

Download the original qualitative case study PDF.

License

The files in this repository are released under the Apache License 2.0. DAS-2M is distributed separately under the terms on its dataset page.

Citation

@article{xu2026deepacademicsurvey,
  title   = {Deep Academic Survey: Stateful Agentic Closed-Loop Paradigm for Academic Survey Automation},
  author  = {Xu, Zhikai and Xue, Zhucun and Hu, Teng and Wang, Yabiao and Liu, Yong and Zhang, Jiangning},
  journal = {arXiv preprint},
  year    = {2026}
}