Fueled by enthusiasm of volunteers, government subsidies, and open data legislation, the Web of Data (WoD) has enjoyed a phenomenal growth. Commercial data, however, has been stuck in proprietary silos, as the monetization strategy for sharing data in the WoD is unclear. This contrasts with the traditional web where advertisement fueled a lot of the growth. This raises the question how the WoD can (i) maintain its success when government subsidies disappear and (ii) convince commercial entities to share their wealth of data. In this paper, we propose a marketplace for decentralized data following basic WoD principles. Our approach allows a customer to buy data from di erent, decentralized providers in a transparent way. As such, our marketplace presents a rst step towards an economically viable WoD beyond subsidies.
The Web of Data (WoD) is a machine-readable alternative to the traditional World Wide Web. In the WoD, data is exposed in a semantically annotated format which allows machines to easily access the information they need according to the task they are performing. Due to the ease of integration given by the underlying Semantic Web technologies, data sources can be queried in a federated fashion without agreeing on a common scheme beforehand. Hence, the WoD can be seen as one big, decentralized database which can be queried over the Web.
Without nancial incentives, many promising datasets will be poorly
maintained or be unavailable as relying on volunteers is not enough to keep the data
up-to-date and the endpoint running. Indeed, as [
From our opinion, one main reason is a lack in nancial incentives for people
to provide data in a semantic format. Unlike in the traditional Web,
semantic data is accessed primarily by automatic agents rather than human ones.
Therefore, advertisements are completely ignored while accessing the data. An
alternative to advertisement is to charge a fee for accessing the data. Such
strategies are already pursued in the traditional Web by companies like Bloomberg,
LexisNexis, and Thomson Reuters. Also, marketplaces like the Azure
DataMarketplace allow di erent publishers to sell data with di erent subscriptions. So far,
none of these implemented markets allow users to buy data in an integrated way
from decentralized data providers. Speci cally, it is not possible for a user to buy
data which constitutes of a join between di erent datasets from di erent sources.
Hence, applying the aforementioned subscription-based monetization strategies
to the WoD is not compatible with the idea of a decentralized Semantic Web.
How can we wean the WoD from government subsidies or federation-averse
centralization? Finding an answer to this question is crucial to ful ll the promise
of the data economy [
Our vision is to create a marketplace where decentralized data providers can o er their data and customers can buy answers to SPARQL queries. Such a marketplace is one possible way to make the Semantic Web independent from subsidies and make it nancially sustainable. Our marketplace di erentiates itself from existing marketplaces by these two distinctive features: 1. Customers can buy commercial data in an integrated way { meaning that a customer can buy data from multiple commercial datasets as if they would be one big, commercial database. 2. Customers are buying only the data which is needed to form the respective query answer instead of buying all data which is involved in query execution. 2
We propose to build a marketplace which allows combining data from di erent sources in an integrated way. Depending on the query and the providers involved, there might be di erent combination of providers' data that yield non-empty query answers. The market hence needs to (1) decide which datasets to include| a process that is akin to source selection but needs to consider the prices for the di erent results and (2) determine optimal payments to each of the providers ensuring their participation in the marketplace.
Most data in the Semantic Web is not located in a single endpoint but
distributed over several endpoints. Each endpoint can, potentially, contribute to
a given query answer. Searching for endpoints o ering the required data
becomes cumbersome if the number of endpoints increases. A customer may need
to access a lot of data out of which only a few (if any at all) end up in the
result. Given these problems, we argue for a marketplace which is able to
assess individual endpoints on their usefulness for a given query and which can
help the customer to decide which data should be bought. As we have shown in
[
Once a query is executed on promising endpoints, the result can be rated by the marketplace and either a buying decision can be made by the market on behalf of the customer, or a summary of the ndings can be given to the customer who can then make a buying decision. Only after the buying decision has been made and the involved payments have been completed, the customer will receive the actual data.
Besides the buying decision, the market has to determine (1) how much a customer has to pay for the query answer and (2) how much payment each provider's contribution to a query answer warrants.
Figure 1 shows all the steps in our marketplace: 1. The market receives a query from the customer and executes it on the available sources. 2. Only a certain number of solutions from the original query answer are selected to compose the nal query answer the customer will receive. Either the market does this buying decision on behalf of the customer or the customer decides based on some statistics which solutions to include into the nal query answer. 3. The customer pays the marketplace the indicated price and receives the query answer. 3
To discuss the costs involved in producing a query answer, we distinguish between two di erent roles on the sellers' side: Provider and Host.
A provider is the originator of data, which is used in the production of a
query answer. Providers are responsible for the quality of data, including
recentness, consistency and accuracy [
The separation between host and provider enables more exible business models for data provision, as some providers might have an initial budget to create data (e.g., government subsidies) but do not have the funds to cover the operating costs for running a SPARQL endpoint or may have other reasons to outsource the actual data provision. Providers can decide to act at the same time as a host for their own and/or other provider's data. Nevertheless, we will distinguish between these two di erent roles and treat them as separate entities.
Data providers might have large xed costs, which typically accrue whilst creating the data. The marginal costs of o ering data, however, is (e ectively) 1. 2. 3. Fig. 1. The three steps from a query to the zero for the provider. This is because, as discussed above, the data is not served by providers but by hosts. Any cost that might occur while o ering data is in icted on the host. It is important to note that even if a provider acts as its own host, the marginal costs are only in icted on the entity acting as a host, not as a provider.
Like cloud service providers, hosts incur the xed cost of operating the infrastructure, possibly some variable cost relative in the size of the data they store, and some marginal cost in form of the computational resources spent for each executed query. The host's marginal costs occur whenever the providers' data are queried, independently of whether any data will eventually be bought by a customer.
Data providers rely on the hosts to make their data available to the marketplace and thus, enable customers to buy their data. Similar to a Web host for traditional Web content, hosts in our market concept are paid by the provider, based on some service agreement. Hence, the providers have to include the hosting costs into their pricing decision. The hosts' costs are already compensated by the providers prior to query execution by the market. Thus, the hosts' costs become transparent to the market and, as a result, the market and customer do not have to take them into account. This facilitates the buying decision.
Figure 2 shows how the payments are distributed from the marketplace to the providers. The providers pay the hosts depending on some service agreement. Note that the payment from provider to the host can be independent from the payment from the marketplace to the providers. Also note that only provider which can actually contribute to the nal query answers are getting payed for their services. 4
To continue growing and being able to serve as a high-quality, decentralized data source, the WoD has to nd the means to fund the creation, serving, and maintenance of data sources. In this paper, we proposed a new vision for funding these activities in the form of a marketplace for Semantic Data.
As a precursor to our research, we conducted a pilot study simulating a
market platform for the WoD [
Based on our research, we foresee the following challenges in building a market place for Semantic Data: { Di erent market mechanisms have to be explored to understand their tradeo s under various market settings.
$
$ Data Data Data Data Provides Provides Provides $ Provides
Fig. 2. The customer pays the market which redirects the payment to the providers. The providers pay the hosts for their services.
{ Given a market mechanism, providers of Semantic Data have to decide how they will bundle and price the data they are selling. The challenge is to nd prices which will satisfy the customers and allow the providers to cover their costs. { Our market idea introduces a new metric for source selection, query optimization, and query execution: the nancial pro tability. Revisiting known techniques and developing new techniques with respect to this new metric will undoubtedly open interesting opportunities for research. { A customer might not know enough about the structure of the o ered data to compose a query in the rst place. Data providers might be required to o er some representative sample data for free or allowing to execute explorative queries (with possible limitations) for free on their datasets.
This paper is a rst step in the direction of nding stable nancing for the WoD. We plan to address the aforementioned challenges in future work and believe that our vision of a marketplace for Semantic Data is a promising way to ensure the nancial sustainability of decentralized providers of Semantic Data.
This work was partially supported by the Swiss National Science Foundation under grant #153598.