Knowledge Graphs encode valuable information that has been so far restricted to the companies that developed them, or dependent on public subsidies. Data marketplaces have emerged as platforms where data consumers meet providers to find the data they need and willing to pay for, as an answer to the need of enabling data transactions. In this position paper we examine the specifics of selling Knowledge Graphs with respect to general datasets, and how can we sell decentralized Knowledge Graphs in a decentralized way.
In this section we present the challenges and problems associated with selling KGs. For each theme we discuss related work on general datasets, identify key differences for KGs, and formulate research questions. When applicable, we also discuss how established techniques for KGs could be adapted to support the sale of a KG.
Data Pricing and Valuation
[
We posit that query pricing in KGs will follow the same general results than those found in databases. A potential research direction is to exploit the semantics of the KG to fine-tune the prices of different subsets to strike an appropriate balance between arbitrage and reasonable prices. Another interesting difference is to assess if the potential price of a de-reference to another KG. In terms of collaborative construction of a KG, we believe Shapley-value techniques as used for ML models will still be relevant, but target metrics will be much more complex than “more accurate”. What are general target metrics for a KG? If they depend on Semantics, what algorithms to derive them? 2.2.
The first step of a data owner that wants to sell in a data marketplace is to create a description to
publish, in a very similar way as when wanting to publish to join the Linked Data Cloud. We expect
existing tools for generating metadata [
Another important aspect when selling KGs is the definition of privacy and usage policies. This is
particularly important for KGs that include personal data. Several works have studied how to define
and evaluate usage policies for queries [
Transactions in a data marketplace are not restricted to datasets. The ecosystem is completed with services (aka Data Apps or Data Processing Operations) that can be used to analyze data, leading to the problem of given dataset descriptions, find services that could be used to process them according to certain desired characteristics. The problem relates to that of Web Service Discovery: matching formally described user requests with service functionality satisfying these requests [20,21]. As with datasets, the main blocker is the problem of generating the description as automatically as possible.
So far, we have assumed that a transaction occurs in a centralized data marketplace controlled by a
single operator trusted by the transaction participants. However, from an economics point of view we
expect multiple operators develop different marketplaces with different rules and value propositions.
Data owners may choose to advertise their datasets in one or many data marketplaces and data
consumers may need to use datasets. At a data level, we believe this is an iteration of the Linked Closed
Data vision from 2011 [
The simple assumption is that either the marketplace operator or one of the participants is entrusted to receive all data and algorithms to execute the agreed data processing workflow. Sadly, there are several scenarios where this is not desirable, the most evident being a dataset’ usage policies forbidding its transfer outside the owner’s infrastructure. The most active related area of research is Federated Learning, where multiple organisations train a Machine Learning Model without leaking data [22]. How to generalize this to any type of data operation or to a sequence of data operation? The problem is reminiscent to that of Workflow Management Systems, that has been revisited to support Big Data processing pipelines [23]. More generally, Trusted Execution Environments [26] have been proposed to deal with this problem, and have been demonstrated in “privacy-preserving” data marketplaces [27].
In a marketplace context, questions about payment arise. What protocols to authorise access and unlock payment based on Service Level Agreement conditions when data and algorithms are decentralized? Are Blockchains like in IoT marketplaces [24] the most efficient solution? Or are there more efficient protocols that only require a minimal amount of trust among them? 3.2.
In an ideal (Linked Data) world and in many IoT visions transactions automated agents consume and follow the published rules and disputes are avoided. When humans and money are involved, dishonest actors may try to breach usage conditions, or simply not deliver the promise of its specification. In a centralized scenario, the data marketplace operator may have the mechanisms to identify and punish dishonest members. But how to achieve the same in a decentralized scenario? If we assume it is not possible to agree on a centralised arbiter that monitors the behaviour of participants, what protocols the federated marketplace operators can implement? What is the minimum amount of information that needs to be exchanged or maintained?
In this paper, we examined the particularities of selling Knowledge Graphs on the light of the emergence of data marketplaces. We reviewed related work across two dimensions: (i) selling Knowledge Graphs (ii) decentralized sale of Knowledge Graphs. Our intent is this description will inspire researchers to coordinate efforts in this direction
Authors were supported by the European Union’s Horizon Europe research and innovation actions under Grant Agreement Nº 101093216. More information available at https://upcastproject.eu/
[20] M. Kifer, R. Lara, A. Polleres, C. Zhao, U. Keller, H. Lausen and D. Fensel, ‘A Logical Framework for Web Service Discovery’. Proceedings of the ISWC 2004 Workshop on Semantic Web Services: Preparing to Meet the World of Business Applications, 2004 [21] Cabral, L., Domingue, J., Motta, E., Payne, T., Hakimpour, F. (2004). Approaches to Semantic Web Services: an Overview and Comparisons. In: Bussler, C.J., Davies, J., Fensel, D., Studer, R. (eds) The Semantic Web: Research and Applications. ESWS 2004. [22] D. Roman et al., "Big Data Pipelines on the Computing Continuum: Tapping the Dark Data," in Computer, vol. 55, no. 11, pp. 74-84, 2022. [23] C. Zhang, Y. Xie, H. Bai, B. Yu, W. Li, Y. Gao, A survey on federated learning, Knowledge
Based Systems, Volume 216, 2021. [24] Christidis J, Karkazis PA, Papadopoulos P, Leligou HC. Decentralized Blockchain-Based IoT Data
Marketplaces. Journal of Sensor and Actuator Networks. 2022 [25] Sebrechts, M. et al. (2022). Solid Web Monetization. In: Di Noia, T., Ko, IY., Schedl, M., Ardito, C. (eds) Web Engineering. ICWE 2022. Lecture Notes in Computer Science, vol 13362. Springer, Cham. https://doi.org/10.1007/978-3-031-09917-5_40 [26] M. Sabt, M. Achemlal and A. Bouabdallah, "Trusted Execution Environment: What It is, and What It is Not," 2015 IEEE Trustcom/BigDataSE/ISPA, Helsinki, Finland, 2015, pp. 57-64, doi: 10.1109/Trustcom.2015.357 [27] Nick Hynes, David Dao, David Yan, Raymond Cheng, and Dawn Song. 2018. A demonstration of sterling: a privacy-preserving data marketplace. Proc. VLDB Endow. 11, 12 (August 2018), 2086– 2089. https://doi.org/10.14778/3229863.3236266