nithanthram/wireless-reuse-taxonomy

Wireless Dataset Reuse Taxonomy

This dataset accompanies the paper “Why Aren't We Reusing Datasets More in Wireless Research?” It provides an annotated catalog of papers and datasets used in wireless research. The release contains scholarly metadata and derived annotations rather than copies of the underlying wireless datasets or paper full text.

Contents

Configuration	File	Records	Description
pdf_backed_datasets	consolidated_datasets_pdf_only.csv	322	Primary canonical dataset catalog used for the paper's dataset-level analysis. Each row has at least one dataset mention extracted from full paper text.
papers	consolidated_papers.csv	6,560	Papers retained with a yes or maybe wireless-relevance classification.
bibliography	consolidated_bibtex.csv	6,560	Bibliographic records corresponding to the retained papers.
manual_validation_datasets	manual_datasets.csv	99 mentions representing 98 unique datasets	Manually curated dataset annotations in the validation scope.
manual_validation_papers	manual_papers.csv	79	Manually identified wireless papers in the validation scope.
ai_validation_datasets	ai_validation_datasets.csv	95	Canonical datasets recovered by the AI-assisted workflow in the matched validation scope.
ai_validation_papers	ai_validation_papers.csv	82	Wireless-relevant papers recovered by the AI-assisted workflow in the matched validation scope.

The manual and AI-assisted validation files cover the same publication scope. Validation-only reuse is calculated from dataset mentions associated with papers in that shared scope.

Primary dataset schema

consolidated_datasets_pdf_only.csv contains:

• Canonical Name: canonical dataset or data-source name after entity resolution.
• All Name Variants: alternate extracted names associated with the canonical record.
• Bibtex Citation Keys: distinct corpus papers associated with the dataset.
• Reuse Count: number of distinct associated papers in the full corpus.
• Modality(ies): extracted measurement or data types.
• OSI Layers: associated OSI-layer coverage.
• Collection Environment: extracted setting, such as real-world deployment, physical laboratory testbed, or simulation.
• Availability Status: practical public-access classification. open denotes verified public access; closed covers access that was restricted, proprietary, request-based, unknown, or otherwise unverified.
• Availability URL: public access location when identified and verified.

The Datasets field in consolidated_papers.csv contains per-paper extracted names. Use consolidated_datasets_pdf_only.csv as the authoritative canonical catalog. The Reuse Count field in ai_validation_datasets.csv refers to full-corpus reuse; validation-only reuse is computed by matching its citation keys to ai_validation_papers.csv.

Methodology

Paper records were collected from ACM SIGCOMM, ACM IMC, USENIX NSDI, IEEE ICC, and IEEE Transactions on Wireless Communications across the study period. An LLM classified papers according to whether wireless communication, networking, or sensing was central to the work. The retained paper catalog combines the yes and maybe classifications to favor recall.

For retained papers, an LLM extracted structured dataset metadata from paper text. The primary catalog consists of canonical dataset records with full-paper-text-backed extraction evidence.

Dataset mentions were consolidated using exact-name matching, normalized availability URLs, fuzzy similarity over names and metadata, and LLM confirmation of ambiguous candidate pairs. Reuse counts were computed from the distinct paper citation keys associated with each canonical dataset.

Availability was evaluated using evidence from paper text and associated public sources. Public labels were checked against identified access URLs, with targeted review of prominent datasets when automated verification was inconclusive.

Summary

• 322 unique datasets with full-paper-text-backed extraction evidence.
• 124 datasets (39%) with verified public access.
• 299 datasets appearing in one paper.
• 23 datasets appearing in more than one paper.
• 21 of 23 multi-use datasets (91%) with verified public access.
• 103 of 299 single-use datasets (34%) with verified public access.

In the manual validation set, 26 of 98 unique datasets (27%) were public and one dataset appeared in multiple papers. In the matched AI-assisted validation set, 39 of 95 datasets (41%) were public and two datasets appeared in multiple validation papers.

Limitations

The catalog is based on automated extraction and conservative entity resolution. Dataset mentions can be missed, and records describing the same artifact can remain separate when there is insufficient evidence to merge them. Reuse may therefore be under-counted. Availability reflects practical access at verification time and can be affected by link changes, registration requirements, repository migration, and incomplete availability statements. The annotations should not be interpreted as authoritative metadata supplied by the original authors or publishers.

Code and prompts

The analysis pipeline and prompt inventory are available at:

https://github.com/Nithanth/wireless-mnist-crawler

License

No license is granted with this release. The files are publicly viewable; contact the authors regarding redistribution or reuse of the annotations.

Citation

Please cite the accompanying paper and this dataset repository when using the annotations.