QL'98 POSITION PAPER 
 Leon Shklar (Pencom Web Works and Rutgers University)
 shklar@cs.rutgers.edu 

The World Wide Web is rapidly becoming ubiquitous, and its rapid
expansion comes increasingly in the form of computable
specifications. This trend is further strengthened by growing
acceptance of Resource Description Framework (RDF). A Web site that is
composed of such specifications may or may not be co-located with the
actual content. Either way, querying such a site becomes a problem of
combining full-text queries and metadata queries against the
specifications.

Querying an RDF graph is non-trivial. The way we currently approach
this problem is through search contexts. A search context is
determined by the current node. The result set is defined as a set of
nodes reachable from the current node that satisfies query
conditions. Valid query
conditions include:

1. Path conditions, for example, friend.dog.breed = "airdale", means
   that only nodes reachable via the <friend> relation such that there
   exist nodes reachable via the <dog> relation from them, for which
   here exists a relation with target "airdale", and intermediate
   nodes, may be included in the result set.

2. Depth condition, for example, depth = 1, in addition to the
   condition in the example above means that that the intermediate
   nodes must not be included.

3. Full text search condition, for example, content ~ "tail paw",
   means that only results of the full text search for specified
   keywords may be considered. We treat "contentIndex" as a dedicated
   relation pointing to the search node. This relation has to be
   present for the "~" to be evaluated.

              "Ben"
                |
               name
                |
                *---dog---*---breed---"spaniel"
                |
             friend
                |               
  current --->  * ---friend--- * ---dog---*---breed---"airdale"
  node          |              |          |
                |             name       name
           contentIndex        |          |
                |            "Andy"     "Bim"
                |
                *---type---"excite"
                |
             location
                |
              "..."
                
We have encountered a number of algorithmical difficulties in
implementing the search engine for our query langauge, most of them
related to efficiently combining results of evaluating full text and
path conditions.