Contexts as metadata for data management

点击翻译按钮获取中文摘要

Meaning, usability and quality of data, and also data cleaning depend on the context. Information systems should represent and use contexts for data management tasks. However, the notion of context has not been fully formalized by the database community. Contexts appear in data management with a few obvious connotations, like time and geographic location. A full-fledged, general, and abstract formalization is missing and necessary. Metadata is data about data. It has to be represented and managed in computational terms, stored in databases, and made accessible for data sources. Classical examples of metadata are relational schemas and integrity constraints. More recent examples are schema mappings, view definitions, trust, provenance, quality constraints, ontologies. We see a context as a form of metadata. In consequence, it has to be properly represented, managed and used in information systems. Our long-term goal is to provide a general, logical formalization of context as required for data management, to be used for: (a) Making sense of data by providing additional meaning, term disambiguation, etc. (b) Explaining data. (c) Specifying relevance of data. (d) Capturing commonsense assumptions about data. (e) Providing appropriate dimensions for data understanding and usage. Etc. Our shorter-term goals address problems around the combination of contextual information with the data it describes: 1. Use of contexts, including quality constraints, to specify the intended, quality contents of a data source. 2. Querying data sources through a context, making query answers context-sensitive, in particular, quality-aware. 3. General formalization of data dimension for data analysis and understanding, with applications to management of dimensional data. 4. Integration of data exchange with data cleaning via matching dependencies.