7.1 Example of system operation When the Quickstep recommender system is first initialised, it retrieves a list of people and their publication URLs from the N ontology. Quickstep analyses these publications and classifies them according to the research topic hierarchy in the ontology 1/ publication age(n) Paper topics are associated with their date of publication, and the new-system initial profile'algorithm used to compute a set of initial profiles for each user. profile interest(u, t)=interest of user u in topic t* CoP confidence Tables I and 2 shows an example of this for the user Nigel Shadbolt. His publications are analysed and a set of topics and new-user initial p (t, topic interest(t))* dates formulated. The 'new-system initial profile'algorithm then t=research paper topic computes the interest values for each topic. For example, Knowledge Acquisition'has one publication two year old(round y= weighting constant >=0 up) so its value is 1.0/2=0.5 Nsimilr= number of similar users pubs= number of publications belonging to class t Tablel. Publication list for Shad bolt CoP confidence= Communities of practice confidence Publication Date Topic Figure 9. New-user initial profile algorithm User Preferences: ontologies 2001 Recommender The task of populating and maintaining the ontology of user recommender systems search interests is left to the recommender system. The recommender system compiles user profiles on a daily basis, and Knowledge Technologies these profiles are asserted into the ontology when ready. Figure 10 2001Technology details the structure of these profiles. In this way up-to-date The Use of Ontologies for 2001 Ontology terests are maintained, providing a solution to the interest Knowledge Acquisition acquisition problem. The interest data is used alongside the more Certifying KBSs: Using atic information within the ontology to improve the accuracy of CommonKADS to provide the Onto CoPl system. Supporting Evidence for Fitness for2000Knowledge user profile=(topic, interest Purpose of KBSs topic =research topic Extracting Focused Knowledge from interest interest value the Semantic Web 2000Acquisition Figure 10. Daily profiles sent to the AKT ontology Knowledge Engineering and 2000 Knowledge Management Table 2. Example of new-system profile for Shadbolt Interest dge Technology\Ontology AIAgents\Recommender Systems Knowledge Technology\Knowledge Acquisition 0.5t = research paper topic u = user γ = weighting constant >= 0 Nsimilar = number of similar users Npubs t = number of publications belonging to class t CoP confidence = Communities of practice confidence topic interest(t) = ˇ n 1.. Npubs t + 1 / publication age(n) ˇ u 1.. Nsimilar _____ profile interest(u,t) Nsimilar γ profile interest(u,t) = interest of user u in topic t * CoP confidence new-user initial profile = (t, topic interest(t))* Figure 9. New-user initial profile algorithm The task of populating and maintaining the ontology of user research interests is left to the recommender system. The recommender system compiles user profiles on a daily basis, and these profiles are asserted into the ontology when ready. Figure 10 details the structure of these profiles. In this way up-to-date interests are maintained, providing a solution to the interest acquisition problem. The interest data is used alongside the more static information within the ontology to improve the accuracy of the OntoCoPI system. user profile = (topic, interest)* topic = research topic interest = interest value Figure 10. Daily profiles sent to the AKT ontology 7.1 Example of system operation When the Quickstep recommender system is first initialised, it retrieves a list of people and their publication URLs from the ontology. Quickstep analyses these publications and classifies them according to the research topic hierarchy in the ontology. Paper topics are associated with their date of publication, and the ‘new-system initial profile’ algorithm used to compute a set of initial profiles for each user. Tables 1 and 2 shows an example of this for the user Nigel Shadbolt. His publications are analysed and a set of topics and dates formulated. The ‘new-system initial profile’ algorithm then computes the interest values for each topic. For example, ‘Knowledge Acquisition’ has one publication two year old (round up) so its value is 1.0 / 2 = 0.5. Table1. Publication list for Shadbolt Publication Date Topic Capturing Knowledge of User Preferences: ontologies on recommender systems 2001 Recommender systems Knowledge Technologies 2001 Knowledge Technology The Use of Ontologies for Knowledge Acquisition 2001 Ontology Certifying KBSs: Using CommonKADS to Provide Supporting Evidence for Fitness for Purpose of KBSs 2000 Knowledge Management Extracting Focused Knowledge from the Semantic Web 2000 Knowledge Acquisition Knowledge Engineering and Management 2000 Knowledge Management … Table2. Example of new-system profile for Shadbolt Topic Interest Knowledge Technology\Knowledge Management 1.5 Knowledge Technology\Ontology 1.0 AI\Agents\Recommender Systems 1.0 Knowledge Technology\Knowledge Acquisition 0.5 …