100 likes | 198 Views
Different layers of expressiveness QueryLang and/or Data Model No sessions ! Architecture Verify that OurLanguage fits in architecture Experiments EUNLRE, ARIADNE/GLOBE, KnowledgeMarkets Due diligence ~existing repositories Later Context (in repository and thus OL or separate?)
E N D
Different layers of expressiveness • QueryLang and/or Data Model • No sessions ! • Architecture • Verify that OurLanguage fits in architecture • Experiments • EUNLRE, ARIADNE/GLOBE, KnowledgeMarkets • Due diligence • ~existing repositories • Later • Context (in repository and thus OL or separate?) • Simple restrictions can be dealt with in OL - check against layers • result set, …
Plan • OLv0.7: end of June • Summer: implementation/experiments • October: re-asses OLv0.7 based on experiments • OLv0.9: 1 Jan 2007 • Deliverable: 1 July 2007
Layers of expressiveness for results • Assumption: result=identifiers of resources • The resource itself is out of scope. (delegate to obtain) • Identifier of metadata?? • Needed for multiple descriptions of same resource in same repository • Indicate that you relaxed the query • Some qualitative indication on path matching,from “perfect” (perfect mapping) over “pretty close” (known close mapping) to “informed guess” (partial path match) to “just guessing” • Note: relates to user profiling and context • Ranking • Level 0: bag • Level 1: partial sequence • 1st result at least as important as 2nd at least as important as 3rd … • Level 2: set of <id,%> • Level 3: set of <id, %, method to calculate %> • Method can include metadata instance that made the resource appear in the result • Metadata • Level 0: only identifier • Level 1: specific metadata • Not always all! • Level 2: actual resource ---> out of scope • Merging • Out of scope: for middle layer or application
Layers of expressiveness • Assumption: result=identifiers of resources • Ranking? • Merging? • 0: search terms (same as CQL) • “dog” • AND if several search terms
Layer 1: properties of resources • Property is • path, but no expressions, parentheses, … • Title as well as DC.title, LOM.General.Title • ~CQL: SRW.ServerChoice… • Example: DC.title, LOM.general.title, … • “Namespace” • We reserve the following path roots • DC, LOM, MPEG • No core properties, nothing special about Types • documents or people or events or … • “jaguar” cars • Only “User defined” • LOM, DC, MPEG • Core attributes of values? • Level 1: just strings • AND of path property-search term pairs • Note: • Lucene would support this • CQL Level 1: Boolean operators or reference to index
Level 2 • Parentheses • LOM.General.Identifier.(catalog=isbn and entry=xxxxx) • no variables, no skipping operators • Namespace extensions? • To be re-discussed once level 1 is clear • UP TO LEVEL 2, WE ARE WITHIN MORE OR LESS CURRENT STATE OF THE ART FOR NOW • We need to make at least explicit that we don’t do what comes beyond in what came before
Level 3 • Other kinds of values than strings? • URI to identify value? • Type of value? • Level 4 • Joins, quantifiers, nested queries • Variables • Needs to be separated out? • Level 5 • recursion
To discuss? • Properties of values: • Type (URI): restricts values • string value [with URI]: representing value • URI: identifies value • Namespace, common data model • We don’t want • Proximity operator • Hard to implement • We may not need it in our context • Because we do have structure (I.e. schema) • Relevance feedback • Separate issue? • Needs the protocol to obtain the document?
How do we go from here • Do a short note on the wiki • Present • PoLiMi • Friday 19 morning • Post slides to the wiki • By presenters • By Monday 22 May • Document that details levels 0, 1 and 2 • Structure from these slides: PoLiMi & Stefaan • 1 June • LOMI seminar to verify that we agree on what we agreed on • 6 June, 14h30, FlashMeeting&SummerSchool
+ / - • “less hot” • What is our implementation and/or research ambition? • Remains unclear • Uptake most important? • More aspects than SQI • How does it relate to professional learning? • Enabler • We need better title? • Do we understand the community we’re trying to serve? • This is more about our visions than about existing practice & there are early initiatives • Focus on concept, not on syntax • “proximity” and semantics • Not just presenting, we did a lot of work • focused discussion • LOMI seminars helped to prepare, so we could focus fast • And we needed to have f2f to move beyond what we did before • “like you all” • Nice hot weather • Later: look at protocol for interaction with repository • All have contributed at technical level • We learn to understand one another