The development of fusion rule technology was supported by the EPSRC Structured text analysis: Addressing inconsistency in heterogeneous information project (2002-2005).
Knowledge fusion is the process by which heterogeneous information from multiple sources is merged to create knowledge that is more complete, less uncertain, and less conflicting than the input. We can view knowledge fusion as a process that creates knowledge. Knowledge fusion can also involve annotating the output information with meta-level information about the provenance of the information used and the mode of aggregation.
Our approach to developing automated knowledge fusion is to draw on formal logic-based techniques for specifiying how knowledge fusion should be undertaken and for representing and reasoning with the background knowledge that is necessary for doing knowledge fusion intelligently. One of our research directions in knowledge fusion has been the development of fusion rule technology which has been supported by funding from EPSRC in the period 2002-2005.
Fusion rule technology is a logic-based approach to knowledge fusion. A set of fusion rules is a way of specifying how to merge structured reports. Structured reports are XML documents, where the textentries are restricted to individual words or simple phrases, such as names and domain-specific terminology, and numbers and units. Each structured report is isomorphic to a logical term: Each textentry is a constant symbol, and each tagname is a function symbol. This isomorphism means we can reason with structured reports in logic. Each fusion rule is a formula of meta-level first-order logic, though they are actually written in a form of scripting language. The antecedent of a fusion rule is a call to investigate the information in the structured reports and a knowledgebase, and its consequent is a formula specifying an action to be undertaken to form a merged report. This means each fusion rule is context-sensitive and hence the overall approach to merging is context-sensitive. A set of fusion rules is defined for a given application. They can be used as a definitive specification for merging. Alternatively, different sets of fusion rules, with different merging criteria, can be used to investigate a set of structured reports by looking at the results of merging. Some useful introductory papers on fusion rule technology are the following.
A key feature of fusion rule technology is the knowledgebase that is used in conjunction with a set of fusion rules. The knowledgebase evaluates the conditions, and contains both generic and domain-specific knowledge. We are drawing on a range of existing formal techniques in knowledge representation and reasoning for knowledgebases, as well as extending some of them. These techniques are for handling "Aggregation", "Heterogeneity", "Uncertainty", "Inconsistency", "Expectations", and "Time". Using logic-based theories, with well-understood properties and semantics, within the knowledgebase means that we undertake a deeper and more sophisticated merging of knowledge.
The development of fusion rule technology was being undertaken by Anthony Hunter and Rupert Summerton. We have also collaborated with Weiru Liu (Queen's University of Belfast) on merging uncertain information. As a result of EPSRC funding, we have developed a prototype system in Java for executing fusion rules (FusionTool). This is integrated with a Prolog system and a relational database system for handling background knowledge. We have also implemented a rule engineering workbench (RuleTool) in Java for developing fusion rules for applications. In addition, the group has undertaken a comprehensive trial with weather reports, and has initiated trials on merging structured reports on protein domains with GO hierarchy annotations (with the UCL Department of Biochemistry).
A full list of publications about fusion rules and related topics can be found at Anthony
Hunter's home page.
For more information on Fusion Rule Technology,
contact Anthony Hunter (a.hunter@cs.ucl.ac.uk
or +44 20 7679 7295).