Mediated Information Gain

We deal with the problem how to gain information from many heterogeneous data sources. As a solution we propose a general framework for mediated information gain: it minimizes and supports the human burden, clearly identifies the algorithmically tractable subtasks, and assures a well-understood level of quality. We identify a “mediation task” as an extended amalgamation of searching, querying and updating in traditional information systems, and we introduce the notion of a “formal information state” which unifies and extends the classical notions of a schema, an instance, a query with additional specifications and a returned answer. The quality is concerned with correctness, precision and completeness. Correctness is guaranteed by maintaining internal consistency. Precision is achieved by embeddings identifying only relevant parts of the sources. And completeness is reached by user interactions with revisions and by iteration. The whole framework is characterized by eighteen rules which capture the essentials of mediated information gain and suggest a high-level architecture for implementations.