• LarkcProject
• WP2
• telecons
• 201107

WP2 telco: Agenda and Minutes July 2011

Expected Participants

• Danica Damljanovic (USFD)
• Jose Quesada (MPG)
• Ivan Peikov (Onto)
• Mihai Lupu (IRF)
• Yi Zeng (WICI)

Agenda/Minutes

• Actions from the previous telco

  • mostly done or postponed details: http://wiki.larkc.eu/LarkcProject/WP2/telecons/201104

• the final review (http://wiki.larkc.eu/LarkcProject/WP2/finalReview)

  • Is subsetting useful? Conclusions. (Which methods are suitable for which usecases, which are not, if not why not?, etc.)
    • MPG: selection methods based on statistical semantics+RDF not useful for reducing problem space for reasoning but useful for detecting outliers; recent work is related to generating clusters so instead of cutting off by threshold we through away clusters - this is better then using threshold;
    • WICI: it would be good to ask this question to the usecases such as wp3 and wp6 who use selection; do they think it is useful? there should be a timeslot at the review to discuss this; because if subsetting is useful why then usecases in LarKC don't use it?
      • interest based selection is useful for all scenarios where results should be ranked by relevance according to the user profile; wici has the usecase where Twitter data are used so that they can show how results selected from Semantic Dog Food (e.g. location) is different with and without selection; much more results without selection so our method help filter out irrelevant entries; we intend to connect this to Urban Computing scenario
      • possible demo for review: Web-based interface which is not yet ready but we plan to work on it

      • where is reasoning? is selection by wici useful for reducing space for reasoning? ACTIONS to be answered by WICI

    • Onto: in general using selection to reduce the problem space for reasoning is great however there are some concerns: no real world testing (maybe usecases should use our methods so that they tell us whether it's useful or not?); we miss the description of which method is suitable for which usecase; the problem with selection is (as noted in MPG's experiments) the performance as we iterate over all results to select only those with certain weights; it would work if the algorithms following selection are exponential or if we are selecting a very small set out of the huge dataset;

    • USFD: running experiment (see https://docs.google.com/document/d/1IYKUbO95VivimsaVTUjgtIh0Lag4i9HZ_0SC8qNkDXE/edit?hl=en_GB&authkey=CPnY--IP) in order to answer this question: the idea is to use dbpedia 3.6 and find a set of 10 sparql queries that return different results with and without materialisation and then see whether it makes any difference; experiment in progress; can Ontotext be involved as well so that we test Spreading Activation the same way? Yes, so ACTIONS Danica to share repository on 83 with Ivan and then to send the code/spaql queries

• D2.7.3 (http://wiki.larkc.eu/LarkcProject/WP2/D2.7.3)

  • current status and timeline;
  • request for input from other partners

  • not discussed as Hans is not present ACTIONS: Jose to share minutes with Hans and then they to decide whether they are happy with work from other partners; input from partners to be requested well in advance to avoid late minute suprises; organize a telco if needed for this deliverable (Hans to decide based on the plans for this deliverable)

• migration of old plugins to the new platform and building and documenting the workflows (http://wiki.larkc.eu/LarkcProject/WP2/workflows)

  • status: all finished apart from MPG:

    • Jose action: to chase Ralph to upload the plugins to the market place and to update the workflow page