|
en nl
How does this demo work? Description
This is a demonstration of how Semantic Web technology can help to
interactively search, navigate and annotate annotated media collections. Implementation details
The prototype is built using the SWI-Prolog SeRQL engine.
For the prototype, we first uploaded the metadata of the painting collection
(after conversion to RDF, using the VRA vocabulary). Second, to be able to
use some semantic background knowledge, we uploaded RDF versions of the
Getty thesauri (AAT, ULAN and TGN).
Finally, to find search results referring to synonyms or other closely
related words, in future versions we will also upload an RDF-version of
Wordnet. It will display the results of a (fuzzy) keyword search on the string 'monet'. The engine has indexed all literals in the RDF triple store, which allows us to quickly find all RDF literals matching the search term. For every match, we look in the neighborhood in the RDF graph for resources of a particular type. For now, we select resources of rdf:type vra:visualResource that are within a certain distance metric (default threshold=0.05, lower thresholds return more results) of the matching literal. Thumbnails of all resulting resources will be generated on the fly, along with the vra:title and vra:creator are shown on the upper half of the screen. The lower half will show the same titles in an interactive timeline view, using the dc:date property. A second timeline shows the lifespan of the creators of the results. Note that the latter timeline uses background knowledge that originates from the RDF version of Getty's ULAN (Universal List of Artist Names). In addition to conveying the temporal relations among the paintings and their creators, the timelines are also used for semantic navigation and search. In the current prototype, for example, all painting titles are clickable and can be used to navigate to a web-based annotation form where (privileged) users can add, delete and edit the RDF metadata of the selected painting. The names of the creators in the bottom timeline are also clickable and can be used to refine the search. For example, in contrast with the fuzzy keyword search on 'monet' in the initial query, clicking on the name Monet, Claude in the timeline, will result in an exact search matching all vra:visualResources with a vra:creator property that is equal to the URI that identifies Claude Monet in the Getty ULAN database. What information can I see This demo presents collections from several museums and art repositories:
What is /facet? /facet will allow you access to the same art repositories by simply browse through the categories and their hierarchies. What is Relation Search? Relation search is a new type of search application. It is still under development. Unlike other search application where you use a keyword to get information about a specific topic (e.g. "Tell me about M.C. Escher"), in relation search you can get information about what is the relationship between topics of interest. Questions you can ask in this application are for example:
What is My Art? My Art is your a personalized art shopping basket for Relation Search. My Art is used to collect the topic of your interest. What kind of language support does the system provide? This demo supports English and Dutch. What does the threshold value mean? In general, a lower threshold increases the number of search results, but decreases accuracy and performance. The search strategy uses a graph traversal algorithm to find the resources that are related to a keyword. The threshold value determines (in combination with the weights, see bellow) how deep the algorithm traverses the graph. What do the weights mean? The weights indicate the importance of the properties. A high weight increases the number of search results that are related by that property. Internally the weights of the properties have a value between 0 and 1. In the graph traversal algorithm the value of a node equals the multiplication of the weights from all the properties that were needed to reach it. If the value of a node goes bellow the threshold that branch is cut of. What do the scores mean? A higher score indicates high relevance. The score is the value of a node, see above, multiplied by 100. Can I get duplicate results? The result of the keyword search are clustered. A 'hit' of the same node can come from many different paths. In non-duplication mode, duplication of the same node is collapsed and you will only see one result which is grouped in a cluster with the highest score. In duplication mode, you may see the same node appearing in more then one clusters. What kind of time queries are available? See examples below for variations on using the time queries.
|