Rules and Other Enabling Technologies
The possiblities to add rules to express implicit, interlinked metadata. The addition of Rules which allow to express interlinked, implicit metadata will enable to specify metadata in the form of for instance
http://polleres.net/foaf.rdf#me is author of all publications listed at http://www.informatik.uni-trier.de/~ley/db/indices/a-tree/p/Polleres:Axel.html
All researchers in the organisation http://www.w3.org/ have expertise in SemanticWeb
See also the related use cases of the RuleML Initiative on FindXpRT and of the RIF WG on FOAF rules, The former employs RuleML FOAF to introduce metadata rules as 'conditional facts' about people, which can be normalized into the 'unconditional facts' of RDF FOAF. The latter proposes the addition of lightweight rules to express interlinked and implicit metadata based on RDF; the original version of this use case contains more details and links. Also check out the ones we envision for this initiative at ExpertFinderUseCases.
Some pre-discussion interesting in this context on reputation and trust seems to have also already been ongoing on the rdfweb-dev mailinglist, see e.g. this thread, in which Jen Golbeck's work in a Trust ontology is mentioned.
Other Enabling Technologies
Apart from rules, other technologies and tools are needed to make the ExpertFinder ontology implementable and the ExpertFinder principles designed and working.
Recommendation algorithms and strategies
Recommendation algorithms and strategies comprise functional components constituting mechanisms for expert finding.
The following issues are relevant for building expert finding mechanisms:
* association rule learning: identifying common patterns and co-occurrances and deriving user rules from the available data (e.g., rules of the kind "if X is an expert in Semantic Web => he/she is likely to attend ISWC") * reusing mechanisms for calculation of trust and reputation between agents for the purpose of recommending items(e.g., see PhD thesis of Jen Golbeck) * collaborative-based filtering and other approaches where users explicitly or implicitly decide whether to put forwards or ignore the content (think also of Amazon's recommendation service and Google's page rank) * techniques for automatic evaluation of personnel performance at a job: quite a few of them stem from the field of expert systems and apply rule-based systems, fuzzy logic, genetic algorithms, etc. A brief overview of personnel performance evaluation techniques can be found at Zhdanova, A.V. "Towards Overcoming Limitations of Community Web Portals: a Classmates’ Example]. In Proceedings of the ESWC 2005 Workshop on End User Aspects of the Semantic Web, 29 May 2005, Heraklion, Greece, CEUR Workshop Proceedings, Vol-137, pp. 111-124 (2005). See page 4 of the paper.
- TBC -
(Axel: I could imagine here that we add something on adding reputation and recommender services and underlying algorithms, which allow to evaluate the value of certain information on expertfinder, in combination with the previous section, this may even include fuzzy rules, etc.)
Text Retrieval methods
- TBD -
(Axel: suggested by enrico motta, refer to the TREC Enterprise Track and approaches proposed there, i.e. text retrieval to generate and obtain Semantic Web data.)
Adopting exisisting ways to model communities and social networks is essential for designing algorithms for expert finding.
Relevant papers on modeling of social networks and web communiites include:
* Robert D. Nolker, Lina Zhou: Social Computing and Weighting to Identify Member Roles in Online Communities. Web Intelligence 2005: 87-93. * Mika, P., 2005. Ontologies Are Us: A Unified Model of Social Networks and Semantics. In Proceedings of the 4th International Semantic Web Conference, Springer-Verlag, LNCS 3729, pp. 522-536. * Zhdanova, A.V., Predoiu, L., Pellegrini, T., Fensel, D. "A Social Networking Model of a Web Community". In Proceedings of the 10th International Symposium on Social Communication, 22-26 January 2007, Santiago de Cuba, Cuba, to appear (2007).