Frameworks for dealing with conflicting information and applications

We are often confronted with multiple and conflicting sources of information in many applications such as cooperative information systems, organizational computing, groupware systems, and multidatabases. For example, an organization often maintains a multitude of information systems due to autonomous departmental operations and these systems may hold overlapping and inconsistent data. In a multidatabase, users are granted access to more than one databases where information from different databases may contradict each other. In this thesis, we study this problem at an abstract level and present several frameworks for dealing with conflicting information in knowledge bases/databases. Starting with a single knowledge base that contains conflicting information, we build a system to reason about what can be consistently concluded from the knowledge base. The system includes an intuitive formalization of consistent conclusions and a simple and sound (and, in the propositional case, complete) axiomatization. Our focus is then on how to merge information from multiple knowledge bases that may contradict each other. We first study the desirable properties of the merge operation, such as obedience to the majority rule in case of conflicts and independence of the syntactic forms of the knowledge bases to be merged. Then we propose a novel merge operator, which possesses all of the desirable properties. The operator is then generalized to deal with knowledge bases associated with weights, which allows the knowledge bases to have different degrees of influence on the merge operation. We then deal with merging relational databases under constraints, using the proposed operator. We present a way of merging databases with different or conflicting schemas by treating schema transformation rules as part of constraints in the merge process. We also demonstrate the application of the merge framework in genome mapping. We formulate the STS content mapping problem in genome domain based on the idea of minimizing false positives and false negatives, and use the merge operator to resolve conflicts among mapping constraints and experimental data.