Decision-makers in critical gelds such as medicine and finance make use of a wide range of information available over the Internet. Mediation, a data integration technique for distributed, heterogeneous data sources, manages the complexity and diversity of the information schemas on behalf of clients. We raise here the issue of trust: is the information so obtained trustworthy? Each client can have different perspectives on the desired trustworthiness the information he or she needs. We consider here the scaling problem that arises from a very large number of users accessing information from many different sources. A mediator cannot be expected to manage the potentially quadratic scaling of trust relationships clients can have with information sources. Furthermore, the possibility of using untrustworthy data increases the risk that the resulting data will be unacceptable: a mediator might evaluate a complex query for a client, only to have the answer rejected because the client does not trust the sources of the information. To help address these issues, we introduce a general static trust-typing model, which can infer the trust ratings of query plans, based on trust meta-data about the input data to the query, even before executing the query. We also define essential properties of such a trust-typing model, namely correctness, precision and completeness. We present an example of a trust-typing model and describe some algorithmic frameworks for the use of such trust-typing models in mediator-based query evaluation.
展开▼