Pith. sign in

REVIEW 3 major objections 58 references

DaDaDa: A Dataset for Data Pricing in Data Marketplaces

T0 review · 3 major / 0 minor · reviewed 2026-07-13 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read DaDaDa supplies the first cross-marketplace catalog of 16,147 data products so listed prices can be predicted from standardized metadata.

desk verdict Solid released multi-marketplace dataset with honest baselines; useful infrastructure, but the sales-comparison framing overreaches listed-price prediction. read the letter →

arxiv 2607.08785 v1 pith:KRKDJNLG submitted 2026-06-13 cs.LG cs.AI

Qiheng Sun , Hongwei Zhang , Junxu Liu , Xiaokai Mao , Jinfei Liu , Kui Ren , Haibo Hu This is my paper
classification cs.LGcs.AI
keywords datapricingmarketplacessalescomparisonapproachmetadatabenchmarkpricepredictionproductclassificationretrieval
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Data marketplaces need prices, but cost-based pricing collapses because data copies for almost nothing, and income-based pricing collapses because future revenue from a dataset is hard to forecast. The remaining practical route is sales comparison: price a new listing from what similar listings already ask. Until now that route has been blocked by fragmented taxonomies and uneven metadata across platforms. DaDaDa is a cleaned, unified table of 16,147 products from nine major marketplaces, with fields such as title, description, category, coverage, update frequency, volume, size, pricing mode, and normalized USD price. Models trained on it recover listed prices with competitive accuracy, classify products into a shared category scheme, and power a vertical search prototype that returns comparable candidates with prices attached. The point is to turn scattered public listings into a reusable benchmark for automated data-product pricing and discovery.

What carries the argument

DaDaDa: a unified 14-field metadata schema for 16,147 products, built by marketplace selection, web scraping, cleaning, currency and mode normalization, and manual mapping of heterogeneous categories onto an AWS-style taxonomy.

What would settle it

Train the same pricing protocol on the released explicit-price subset, then check whether predicted log-prices systematically fail on a held-out marketplace or against any newly disclosed actual transaction prices, private discounts, or renegotiated contracts that diverge from catalog listings.

Watch

Extended reading notes

Core claim

The paper establishes that a single standardized metadata schema spanning nine real marketplaces is enough to train supervised models that estimate public listed prices for data products and to support classification and cross-marketplace retrieval. Free and negotiation listings stay in the release for marketplace analysis, while explicit nonzero subscription, one-off, and usage-based prices form the supervised pricing target after mode-specific normalization.

Load-bearing premise

The central claim rests on treating seller-written public catalog prices and metadata, after manual alignment and mode-specific normalization, as a usable sales-comparison signal for pricing new data products rather than private transaction prices or buyer value.

Editorial extensions

If this is right

  • New data products can receive automated listed-price benchmarks by matching their metadata against DaDaDa instead of relying only on seller guesswork.
  • Heterogeneous marketplace taxonomies can be reduced to a shared category layer for comparable classification and search.
  • A vertical retrieval layer can surface cross-platform alternatives with price, coverage, and update constraints in one interface.
  • Future marketplace studies can retrain pricing, classification, and retrieval methods under one public schema and refresh protocol.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If catalog prices remain the main observable signal, pricing research will likely stay text-heavy, because descriptions dominate feature importance when structured fields are sparse.
  • Separating pricing-mode prediction from continuous price regression would better match real marketplaces where negotiation and free listings dominate many catalogs.
  • Periodic public snapshots of the same schema would turn one-shot benchmarks into a way to track how listed prices and disclosure practices drift over time.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 0 minor

Summary. The paper introduces DaDaDa, a standardized metadata corpus of 16,147 data products from nine public marketplaces (Datarade, AWS Data Exchange, Snowflake, Databricks, and five Chinese exchanges). After marketplace selection, scraping, cleaning, currency normalization, and manual alignment of categories to the AWS taxonomy, the authors release a unified schema (title, description, volume, size, dimension, coverage, update frequency, price mode, price, etc.) together with crawlers, preprocessing scripts, and samples. They treat data pricing as supervised listed-price regression on the ~4,929 products with explicit nonzero prices (log(price+1) target), report XGBoost R² ≈ 0.833 under 5-fold CV, ablations of text vs. structured features, SHAP feature groups by pricing mode, and leave-one-marketplace generalization to Snowflake (R² ≈ 0.744). Complementary experiments fine-tune multilingual transformers for category classification (weighted F1 ≈ 0.82) and demonstrate an Elasticsearch retrieval prototype with structured filters.

Significance. If the resource is maintained and used as intended, it fills a genuine empirical gap: theoretical data-pricing mechanisms have lacked real marketplace benchmarks, while prior commercial studies have been largely descriptive. The public release of raw crawls, intermediate files, scripts, and a working retrieval layer is a concrete contribution that enables reproducible pricing, classification, and discovery research. The paper is appropriately scoped as a dataset-plus-baseline resource rather than a new pricing theory; its value lies in making sales-comparison-style experiments feasible across heterogeneous platforms.

major comments (3)
  1. §I and §V-A frame DaDaDa as enabling the sales-comparison approach, yet the supervised task (Table IV) is listed-price regression on the explicit nonzero subset after free/negotiation products are excluded (Table I: only 30.5% of products). Catalog offers are not observed transactions, private discounts, or buyer valuations. The manuscript should either (a) restate the claim as “listed-price estimation under public metadata” and discuss the gap to realized sales comparison, or (b) provide additional evidence (e.g., sensitivity to negotiation-mode products, platform-level price-disclosure bias) that the retained prices are valid comparables.
  2. Table IX leave-one-marketplace transfer to Snowflake drops to R² 0.744 with substantially higher MAE/MSE than the pooled 5-fold results in Table IV. Combined with SHAP results (Fig. 6, Table VIII) showing description as the dominant feature group across modes, this suggests models largely recover catalog-price and marketing-language patterns rather than transferable structural value. A stronger generalization protocol (multiple held-out platforms, or explicit platform-invariant features) is needed before claiming cross-marketplace price benchmarks.
  3. §IV-C and §III: category labels are manually mapped onto the AWS Marketplace taxonomy, and missing coverage/size fields are filled by reading descriptions or defaults. These curation steps are load-bearing for both pricing and classification, yet inter-annotator agreement, mapping rules, and the fraction of imputed fields are not reported. Without that documentation, reproducibility of the “standardized” labels is incomplete.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical dataset release plus standard supervised regression/classification on held-out listed prices, not a self-defining derivation.

full rationale

DaDaDa is a resource paper that scrapes, standardizes, and releases marketplace metadata, then trains ordinary ML models (XGBoost, Gradient Boosting, etc.) to regress log(price+1) or classify categories. Performance is measured by 5-fold CV and held-out marketplace transfer against the same observed listed prices that constitute the labels; free/negotiation items are explicitly excluded from the regression task rather than being redefined into the target. There are no equations that define a quantity in terms of itself, no fitted constant re-labeled as a first-principles prediction, no load-bearing uniqueness theorem imported from the authors’ prior work, and no ansatz smuggled via self-citation. Ordinary dependence of a supervised model on its training distribution is not circularity under the stated criteria. The derivation chain is therefore self-contained and non-circular.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central claim rests less on new physics-like postulates than on construction choices: which marketplaces count, how metadata are aligned, and that listed prices are the right supervised target for 'data pricing' benchmarks. Free parameters are ordinary ML/preprocessing knobs. The main domain assumptions are economic and measurement assumptions about sales comparison and public catalog prices.

free parameters (4)
  • PCA retained dimensions for text embeddings = 223 dims at 95% variance (12th layer)
    XLM-RoBERTa embeddings are reduced to the minimum components explaining 95% variance (reported as 223 for the 12th-layer setting); this choice shapes the pricing feature space.
  • log(price + 1) target transform = log1p(price)
    Chosen because listed prices span many orders of magnitude; all regression metrics are on this transformed target rather than raw USD.
  • Model hyperparameters via grid search
    RF/KNN/GB/XGBoost/DNN hyperparameters are selected by grid search with cross-validation; reported R² depends on these fitted choices.
  • Marketplace selection thresholds = ≥8 categories; ≥70% metadata coverage
    Candidates must be crawlable, cover at least eight categories under AWS-like criteria, and include at least 70% of target metadata fields; these hand-set thresholds determine the nine included platforms.
assumptions (5)
  • domain assumption Cost and income pricing approaches fail for data products, so sales comparison is the viable practical approach.
    Stated in the abstract and introduction as the economic motivation for building a cross-marketplace benchmark.
  • domain assumption Public listed prices, after mode-specific normalization to USD, are a useful supervised target for data-product pricing models.
    Defines the §V-A task; free/negotiation products are excluded because they lack observed listed prices.
  • ad hoc to paper AWS Marketplace category taxonomy is an adequate common label space for products from all included marketplaces.
    Manual mapping in §IV-C creates the unified category field used for analysis and classification.
  • domain assumption Title and description text plus partial structured metadata are sufficient features for pricing, classification, and retrieval benchmarks.
    Justified by metadata sparsity discussion in §III and used throughout §V experiments.
  • standard math Standard supervised learning evaluation (shuffled 5-fold CV, held-out marketplace test) measures pricing-model effectiveness.
    Ordinary ML evaluation assumptions; no formal pricing equilibrium derivation is claimed.
invented entities (1)
  • DaDaDa unified data-product schema and dataset independent evidence
    purpose: Provide a standardized multi-marketplace corpus for pricing, classification, retrieval, and marketplace analysis.
    The paper's primary contribution is this constructed resource rather than a new physical or theoretical object; independent evidence is the public release and downstream baselines.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DaDaDa: A Dataset for Data Pricing in Data Marketplaces." pith.science (2026). https://pith.science/paper/KRKDJNLG

@misc{pith2026260708785,
  author       = {Pith},
  title        = {Pith review of: DaDaDa: A Dataset for Data Pricing in Data Marketplaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KRKDJNLG}},
  note         = {Machine review of arXiv:2607.08785}
}
read the original abstract

High-quality data drives machine learning advances across industries. Recognizing the value of data, data transactions are increasingly common, giving rise to many data marketplaces, e.g., AWS Marketplace, Databricks, and Datarade. However, determining the appropriate prices for data products remains a significant challenge due to the unique properties of data products. Traditional pricing methods in economics can be categorized into the cost approach, the income approach, and the sales comparison approach. The cost approach fails in data pricing due to near-zero marginal cost from data replication, and the income approach fails due to inherently unpredictable data revenue. The sales comparison approach remains viable, yet its application is hindered by the absence of standardized pricing benchmarks for data products across marketplaces. To address this challenge, we introduce \texttt{DaDaDa}, the first dataset for data product pricing, containing metadata for 16,147 data products from 9 major data marketplaces worldwide. \texttt{DaDaDa} enables the training of pricing models, thereby establishing price benchmarks for new data products. In addition, \texttt{DaDaDa} can be utilized for other important tasks in data markets, such as data product classification and retrieval. Experiments and a retrieval prototype demonstrate the effectiveness of \texttt{DaDaDa} for pricing, classification, and retrieval of data products. The dataset and code are available at https://github.com/ZJU-DIVER/DaDaDa.

Figures

Figures reproduced from arXiv: 2607.08785 by the authors.

Figure 1
Figure 1. Percentage of each category in category. F [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Percentage of each pricing mode in price_mode. • Title. The title of the data product. • Platform. The marketplace name hosting the data prod￾uct. • Provider. The name of the data provider as listed on the data marketplace. The released table contains 1,991 named data providers. Among them, “Techsalerator” is the leading provider, offering a total of 644 data products, making it the supplier with the highest volume … view at source ↗
Figure 3
Figure 3. Overview of the data collection pipeline. First, we iden [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Scraping example of Datarade. fields, they are considered duplicates, and the redundant entries are removed. 2. Rectifying errors. During the crawling process, non-UTF￾8 characters may be retrieved, leading to errors when reading files containing such characters. We us…
Figure 5
Figure 5. Figure 5: Histogram and CDF (Cumulative Distribution Function) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Results of SHAP calculation. We calculate the relative [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: A screenshot of the data product search engine. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 6 linked inside Pith

  1. [1]

    https://www.advaneo-datamarketplace.de/database/en, 2026

    Advaneo. https://www.advaneo-datamarketplace.de/database/en, 2026

    2026

  2. [2]

    https://aws.amazon.com/marketplace, 2026

    AWS Marketplace. https://aws.amazon.com/marketplace, 2026

    2026

  3. [3]

    https://azure.microsoft.com/en-us/products/open-datasets/ , 2026

    Azure. https://azure.microsoft.com/en-us/products/open-datasets/ , 2026

    2026

  4. [4]

    https://www.bbgdex.com/, 2026

    Beibu Gulf Data Exchange. https://www.bbgdex.com/, 2026

    2026

  5. [5]

    https://www.bjidex.com, 2026

    Beijing International Data Exchange. https://www.bjidex.com, 2026

    2026

  6. [6]

    https://www.cantonde.com, 2026

    Canton Data Exchange. https://www.cantonde.com, 2026

    2026

  7. [7]

    https://carto.com, 2026

    Carto. https://carto.com, 2026

    2026

  8. [8]

    https://www.databricks.com/product/marketplace, 2026

    Databricks. https://www.databricks.com/product/marketplace, 2026

    2026

Show all 58 references
  1. [9]

    https://www.databroker.global/data-marketplace, 2026

    DataBroker. https://www.databroker.global/data-marketplace, 2026

    2026

  2. [10]

    https://datarade.ai/, 2026

    Datarade. https://datarade.ai/, 2026

    2026

  3. [11]

    https://www.datatang.com/, 2026

    Datatang. https://www.datatang.com/, 2026

    2026

  4. [12]

    https://www.elastic.co/, 2026

    Elasticsearch. https://www.elastic.co/, 2026

    2026

  5. [13]

    https://console.cloud.google.com/marketplace, 2026

    Google Cloud. https://console.cloud.google.com/marketplace, 2026

    2026

  6. [14]

    https://www.gzdex.com.cn, 2026

    Guiyang Global Big Data Exchange. https://www.gzdex.com.cn, 2026

    2026

  7. [15]

    https://mobilithek.info, 2026

    Mobilithek. https://mobilithek.info, 2026

    2026

  8. [16]

    https://data.nasdaq.com, 2026

    Nasdaq data link. https://data.nasdaq.com, 2026

    2026

  9. [17]

    https://www.selenium.dev/, 2026

    Selenium. https://www.selenium.dev/, 2026

    2026

  10. [18]

    https://dtxp.chinadep.com, 2026

    Shanghai Data Exchange. https://dtxp.chinadep.com, 2026

    2026

  11. [19]

    https://www.szdex.com, 2026

    Shenzhen Data Exchange. https://www.szdex.com, 2026

    2026

  12. [20]

    https://www.snowflake.com/en/data-cloud/marketplace/, 2026

    Snowflake . https://www.snowflake.com/en/data-cloud/marketplace/, 2026

    2026

  13. [21]

    https://www.veracity.com, 2026

    V eracity. https://www.veracity.com, 2026

    2026

  14. [22]

    https://www.westdex.com.cn, 2026

    Western China Data Exchange. https://www.westdex.com.cn, 2026

    2026

  15. [23]

    https: //www.irs.gov/individuals/international-taxpayers/ yearly-average-currency-exchange-rates , 2026

    Y early Average Currency Exchange Rates. https: //www.irs.gov/individuals/international-taxpayers/ yearly-average-currency-exchange-rates , 2026

    2026

  16. [24]

    https://ditm.zjdex.com, 2026

    Zhejiang Big Data Exchange. https://ditm.zjdex.com, 2026

    2026

  17. [25]

    A. F. Agarap. Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375 , 2018

    2018 arXiv

  18. [26]

    Agarwal, M

    A. Agarwal, M. Dahleh, and T. Sarkar. A marketplace for data: An algorithmic solution. In Proceedings of the 2019 ACM Conference on Economics and Computation , pages 701–726, 2019

    2019

  19. [27]

    S. A. Azcoitia, C. Iordanou, and N. Laoutaris. What is the price of data? a measurement study of commercial data marketplaces. arXiv preprint arXiv:2111.04427, 2021

    2021 arXiv

  20. [28]

    S. A. Azcoitia, C. Iordanou, and N. Laoutaris. Understanding the price of data in commercial data marketplaces. In Proceedings-International Conference on Data Engineering , 2023

    2023

  21. [29]

    S. A. Azcoitia and N. Laoutaris. A survey of data marketplaces and their business models. ACM SIGMOD Record , 51(3):18–29, 2022

    2022

  22. [30]

    A. E. Baum, N. Crosby, and S. Devaney. Property investment appraisal . John Wiley & Sons, 2021

    2021

  23. [31]

    L. Breiman. Random forests. Machine learning , 45:5–32, 2001

    2001

  24. [32]

    C. Chai, J. Wang, Y . Luo, Z. Niu, and G. Li. Data management for machine learning: A survey. IEEE Transactions on Knowledge and Data Engineering , 35(5):4646–4667, 2022

    2022

  25. [33]

    Chawla, S

    S. Chawla, S. Deep, P . Koutris, and Y . Teng. Revenue maximization for query pricing. Proc. VLDB Endow. , 13(1):1–14, 2019

    2019

  26. [34]

    L. Chen, P . Koutris, and A. Kumar. Towards model-based pricing for machine learning in a data marketplace. In P . A. Boncz, S. Manegold, A. Ailamaki, A. Deshpande, and T. Kraska, editors, Proceedings of the 2019 International Conference on Management of Data, SIGMOD Conferenc...

    2019

  27. [35]

    Chen and C

    T. Chen and C. Guestrin. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages 785–794, 2016

    2016

  28. [36]

    Y . Chen, Y . Shen, and S. Zheng. Truthful data acquisition via peer prediction. In H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, editors, Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, Neu...

    2020

  29. [37]

    Conneau, K

    A. Conneau, K. Khandelwal, N. Goyal, V . Chaudhary, G. Wenzek, F. Guzmán, E. Grave, M. Ott, L. Zettlemoyer, and V . Stoyanov. Unsupervised cross-lingual representation learning at scale. CoRR, abs/1911.02116, 2019

    1911 arXiv

  30. [38]

    Damodaran

    A. Damodaran. Investment valuation: Tools and techniques for deter- mining the value of any asset . John Wiley & Sons, 2012

    2012

  31. [39]

    Deep and P

    S. Deep and P . Koutris. QIRANA: A framework for scalable query pricing. In S. Salihoglu, W. Zhou, R. Chirkova, J. Y ang, and D. Suciu, editors, Proceedings of the 2017 ACM International Conference on Management of Data, SIGMOD Conference 2017, Chicago, IL, USA, May 14-19, 201...

    2017

  32. [40]

    Devlin, M

    J. Devlin, M. Chang, K. Lee, and K. Toutanova. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805, 2018

    2018 arXiv

  33. [41]

    R. Dolan. How do you know when the price is right? Harvard Business Review, 73(5):174–180, 1995

    1995

  34. [42]

    R. C. Fernandez, Z. Abedjan, F. Koko, G. Y uan, S. Madden, and M. Stonebraker. Aurum: A data discovery system. In 34th IEEE International Conference on Data Engineering, ICDE 2018, Paris, France, April 16-19, 2018 , pages 1001–1012. IEEE Computer Society, 2018

    2018

  35. [43]

    R. C. Fernandez, P . Subramaniam, and M. J. Franklin. Data market platforms: Trading data assets to solve data problems. Proceedings of the VLDB Endowment , 13(11):1933–1947, 2020

    1933

  36. [44]

    J. H. Friedman. Greedy function approximation: a gradient boosting machine. Annals of statistics , pages 1189–1232, 2001

    2001

  37. [45]

    Ghosh and A

    A. Ghosh and A. Roth. Selling privacy at auction. In Proceedings of the 12th ACM conference on Electronic commerce , pages 199–208, 2011

    2011

  38. [46]

    Hanspach, G

    P . Hanspach, G. Sapi, and M. Wieting. Algorithms in the marketplace: An empirical analysis of automated pricing in e-commerce. Information Economics and Policy , 69:101111, 2024

    2024

  39. [47]

    N. Jones. The AI revolution is running out of data. what can researchers do? Nature, 636(8042):290–292, 2024

    2024

  40. [48]

    D. P . Kingma and J. Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

  41. [49]

    Koutris, P

    P . Koutris, P . Upadhyaya, M. Balazinska, B. Howe, and D. Suciu. Query- based data pricing. Journal of the ACM (JACM) , 62(5):1–44, 2015

    2015

  42. [50]

    O. Kramer. Unsupervised k-nearest neighbor regression. arXiv preprint arXiv:1107.3600, 2011

    2011 arXiv

  43. [51]

    C. Li, D. Y . Li, G. Miklau, and D. Suciu. A theory of pricing private data. ACM Transactions on Database Systems (TODS) , 39(4):1–28, 2014

    2014

  44. [52]

    J. Liu, J. Lou, J. Liu, L. Xiong, J. Pei, and J. Sun. Dealer: an end-to-end model marketplace with differential privacy. Proceedings of the VLDB Endowment, 14(6), 2021

    2021

  45. [53]

    S. M. Lundberg and S.-I. Lee. A unified approach to interpreting model predictions. Advances in neural information processing systems , 30, 2017

    2017

  46. [54]

    Ma ´ckiewicz and W

    A. Ma ´ckiewicz and W. Ratajczak. Principal components analysis (pca). Computers & Geosciences , 19(3):303–342, 1993

    1993

  47. [55]

    Mehta, M

    S. Mehta, M. Dawande, G. Janakiraman, and V . S. Mookerjee. How to sell a data set? pricing policies for data monetization. Inf. Syst. Res. , 32(4):1281–1297, 2021

    2021

  48. [56]

    J. Pei, R. C. Fernandez, and X. Y u. Data and ai model markets: Opportunities for data and model sharing, discovery, and integration. Proceedings of the VLDB Endowment , 16(12):3872–3873, 2023

    2023

  49. [57]

    Srivastava, G

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdi- nov. Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research , 15(1):1929–1958, 2014

    1929

  50. [58]

    Wieting and G

    M. Wieting and G. Sapi. Algorithms in the marketplace: An empirical analysis of automated pricing in e-commerce. Available at SSRN , 3945137, 2021

    2021

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.