- From: <bass@mit.edu>
- Date: Fri, 30 May 2003 13:59:08 -0400
- To: www-rdf-dspace@w3.org
We discussed three topics on the call, all in review of Jason's current history system design document: 1. How to extend data model (best practice): By Subclass? By Subproperty? Both? Outcome: Need further discussion and clarification on list. 2. Capability/Utility of queries incorporating negative assertions Outcome: JK: Useful/Required for certain history system query use cases. AS: There are alternative ways of modelling that don't require it. JK: Still unclear how to implement? 3. Producing Metadata Disseminations from Metadata in the History System Store Summary: JK: History System gathers metadata about state in past situations, that can be queried. Makes no assumption about what subgraphs could/should be disseminated for particular URIs in the data model. AS: Should be able to http:get a name and receive metadata about it MB: What's missing is the mapping: what's the implied query that determines the subgraph to return for the name. Also (future) could return disseminations that are not metadata, but content. Outcome: JK to schedule additional telecon. Timing logistics TBD on list. Bass to supply phone bridge. FIN [12:01] <mickBass> Attending: [12:01] <mickBass> Kevin Smathers [12:01] <mickBass> Mark Butler [12:01] <mickBass> Jason Kinner [12:02] <mickBass> Andy Seaborne [12:03] <mickBass> Scribe: Kevin Smathers [12:03] <mickBass> Backup Scribe: Mick Bass [12:06] <kevins2> jk: design doc feedback request. new update available, includes marks comments and new diagrams [12:06] <kevins2> jk, dated may 28 [12:06] <kevins2> jk, will accept comments via email. [12:07] <mickBass> http://web.mit.edu/simile/www/resources/history- harmony/history-design.pdf [12:08] * marbut has joined #simile === On extension of Data Model using Subclass and/or Subproperty a.k.a. "Do/Should Subproperties vary meaning based on context?" === [12:09] <kevins2> jk, clarify date and event [12:09] <kevins2> jk, defined explicitly in schema [12:10] <kevins2> jk, i like subEventOf, but there is no use case for that in current simile [12:11] <kevins2> mb, need refinements of date to refine example, or use subproperties from Harmony as an example [12:12] <kevins2> mb, might be subclassed from dc:date, but dc:date is too general to use directly. [12:13] <kevins2> jk, i have seen date used as property directly. no idea what they were talking about in that case, has to be inferred. [12:13] <kevins2> mb, semantic meaning of date changes according to context; ie where it is referenced. [12:14] <kevins2> jk, classes and properties with the same name? [12:14] <kevins2> jk, created event, and created property. [12:15] <kevins2> jk, need to be equated. [12:15] <kevins2> jk, meaning is the same. [12:17] <kevins2> jk, creates is subproperty of hasResult, destroys is subProperty of hasPatient, subproperty of involves [12:17] <kevins2> range is an actuality [12:18] <kevins2> discuss figure 5. [12:18] <kevins2> jk, will it be clear how to extend schema [12:19] <kevins2> jk, event and property are defined by Harmony -- given that, how to extend schema to other possible actions and events? [12:19] <kevins2> jk, event and action are themselves too generic. [12:19] <kevins2> jk, need new classes that are subclasses of event and action. [12:20] <kevins2> jk, but these aren't acceptable to community. need to move in the direction of specific classes with specific properties. [12:20] <kevins2> jk, eg.: event modify should have subproperties that identify the record modified. [12:21] <kevins2> jk, currently only using basic Harmony/ABC in the diagrams (eg: Fig 6) [12:22] <kevins2> jk, two issues -- can't use generic property that changes its semantics in the context of different classes. [12:24] <kevins2> jk, secondly -- extending the schema by creating new subproperties for their implementation e.g transforms class representing action and event, + subproperties of hasresult that represent 'yieldsTransformation'. Can't cleanly extend via classes, nor from properties. Easier to use type of class as part of the query. [12:24] <kevins2> jk, subproperty is harder to understand [12:25] <kevins2> jk, subclasses of actions and events are needed, but need subproperties as well to really help. [12:28] <kevins2> bass, suppose i create a subclass of modified that is transforms... [12:28] <kevins2> range is ...? bass: what's wrong with using hasResult with *domain* = transforms, without defining a subproperty? (yes, I did say range on the call...) [12:29] <kevins2> jk, does hasResult sufficiently model the semantics of modified in the sense of transform? [12:30] <kevins2> bass, is it broken, or is it a user concern that the users won't model it correctly? [12:33] <kevins2> mb, users need guidance. [12:34] <kevins2> mb, key problem is modelling, and there is no set of known good practices. [12:34] <kevins2> mb, digraphs are an unknown. [12:34] <kevins2> jk, this model has some responsibility in detemining best practices. === On RDF/RDQL Query Capabilities wrt negative assertions? === [12:36] <kevins2> jk, andy -- about rdql -- is there a way to make a negative assertion in the graph pattern? [12:36] <kevins2> as, no closed world assumptions. [12:37] <kevins2> jk, most recent version of the object; can't ask what is the object that doesn't have a successor. [12:37] <kevins2> as, other ways to accomplish that -- a closed list that lists objects in creation order for example. [12:39] <kevins2> jk, subqueries for example. [12:40] <kevins2> as, does a subgraph pattern not exist is really what you are looking for [12:43] <kevins2> jk, bass: discuss query results as linked list. [12:43] <kevins2> as, there are more ways to examine a graph than by query. [12:44] <kevins2> as, fine grained tests include exact tests for a relationship. [12:44] <kevins2> jk, for remote clients that won't work. [12:44] <kevins2> jk, need to pose useful queries for haystack. === On producing Disseminations from Metadata in History System Store === [12:46] <kevins2> as, abstraction -- url per resource [12:46] <kevins2> jk, unclear what URL producing metadata dissemination should be linked to. history system can't predict [12:47] <kevins2> as, community, collection, ... [12:47] <kevins2> jk, can't guarantee that handle resolves, or resolves to metadata. [12:51] <kevins2> discussion re handles [12:52] <kevins2> as, history store can't be asked what it knows about object x. [12:53] <kevins2> bass, could be an implied query that returns data from the handles. [12:55] <kevins2> as, need to be able to ask history system what it knows about a particular object -- full data of the historical object. [12:56] <kevins2> bass, issue is what data would be needed from an HTTP get. One case is what metadata we want, the other is that the history system may need to roll back. [12:56] <kevins2> jk, right now history system has no access to that information. [12:57] <kevins2> jk, can't produce disseminations based on these handles.
Received on Friday, 30 May 2003 13:59:10 UTC