Refine
Document Type
- Part of a Book (1)
- Conference Proceeding (1)
Conference Type
- Konferenzartikel (1)
Language
- English (2) (remove)
Has Fulltext
- no (2) (remove)
Is part of the Bibliography
- yes (2) (remove)
Keywords
- Datenmanagement (2) (remove)
Institute
Open Access
- Closed (1)
- Closed Access (1)
Marketing and sales have high expectations of new methods such as Big Data, artificial intelligence, machine learning, and predictive analytics. But following the “garbage in—garbage out” principle, the results leave much to be desired. The reason is often insufficient quality in the underlying customer data. This article sheds light on this problem using the data quality and value pyramid as an example. The higher up the value-added pyramid the data is located, the higher its quality and the more value it generates for a company. In addition, we show how the use of monitoring systems, such as a data quality scorecard, makes data quality visible and improvements measurable. In this way, the actual value of data for companies becomes obvious and manageable.
We propose secure multi-party computation techniques for the distributed computation of the average using a privacy-preserving extension of gossip algorithms. While recently there has been mainly research on the side of gossip algorithms (GA) for data aggregation itself, to the best of our knowledge, the aforementioned research line does not take into consideration the privacy of the entities involved. More concretely, it is our objective to not reveal a node's private input value to any other node in the network, while still computing the average in a fully-decentralized fashion. Not revealing in our setting means that an attacker gains only minor advantage when guessing a node's private input value. We precisely quantify an attacker's advantage when guessing - as a mean for the level of data privacy leakage of a node's contribution. Our results show that by perturbing the input values of each participating node with pseudo-random noise with appropriate statistical properties (i) only a minor and configurable leakage of private information is revealed, by at the same time (ii) providing a good average approximation at each node. Our approach can be applied to a decentralized prosumer market, in which participants act as energy consumers or producers or both, referred to as prosumers.