City Research Online

Abstract

A typical enterprise may have large numbers of information sources such as data stores, expert systems, knowledge-based systems, or standard software systems. These may need to be integrated so that, for example, an application program or a decision maker can access information from all these sources. Such architectures are generally called 'Distributed Collaborative Environments for Enterprise Integration'.

A general problem in these enterprise integration architectures is that information from heterogeneous, pre-existing sources may be obsolete, incomplete, incorrect or, for many other reasons, contradictory. Thus, conflicting results may occur when the same information is requested from semantically related sources. A mechanism is required to detect and resolve these conflicts in a way that is rational to any potential client of the integration environment.

This thesis lays open the design of a general mechanism for conflict detection and resolution that enables intelligent information agents to reason about contradictory information from pre-existing, heterogeneous and autonomous sources. The mechanism's theoretical basis is a framework that is drawn from evidence law, which shares some fundamental commonalities with conflict detection and resolution in enterprise integration environments.

Conflict detection opens with gathering the results collected by the information retrieval process. These results may have justifications or certainty assessments attached to them. Furthermore, it identifies whether and how these results are conflicting.

The design of a conflict resolution mechanism is based on a rational scheme for judging the weight of conflicting results. First, the agents assess the reliability or credibility of an information source. Judgement based on the weight of conflicting results is first applied to any available, domain-specific, resolution strategies. Second, the agent applies any 'general scientific' resolution strategies that are not specific to one domain. When no domain-related expertise can solve the conflict then the agent can only judge on domain independent evaluation criteria such as the results' reliability. A scheme is sketched out for judgement based on the reliability of conflicting results, involving three steps: Ranking the conflicting results according to their reliability; Ways to redefine conflicting results; and Heuristic decision-making.

The evaluation includes a computational implementation of an enterprise integration environment incorporating a model of an information agent. An example is realised in this environment. The conflict detection and resolution mechanism, and interfaces to each integrated source, are implemented in Visual C++. A case study is conducted on this scenario to evaluate each conflict detection and resolution step. Furthermore, this illustrates both the advantages over existing approaches and the limitations.

Publication Type:	Thesis (Doctoral)
Subjects:	H Social Sciences > HG Finance Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	Bayes Business School Bayes Business School > Bayes Business School Doctoral Theses Doctoral Theses

Export

Downloads