Query Scripts

Experiences during the evaluation have led to the following rules for creating patterns and queries in xOperator.

Script Setup

Query formulation

Query as fuzzy as possible

Instant Messaging is a very quick means of communication. Users usually do not capitalize words and use many abbreviations. This should be considered, when designing suitable AIML patterns. If information about the person `Sören Auer' should be retrieved, this can be achieved using the following graph pattern:
?subject foaf:name “Auer”.

However, information can be represented in multiple ways and often we have to deal with minor misrepresentations (such as trailing whitespace or wrong capitalizations), which would result for the above query to fail. Hence, less strict query clauses should be used instead. When searching for information regarding `Sören Auer' the following relaxed SPARQL clause, which matches also substrings and is case insensitive, could be used:
?subject foaf:name ?name. FILTER regex(?name, '.*Auer.*', 'i')

Use patterns instead of qualified identifiers for properties

Similar, as for the identification of objects, properties should be matched flexible. When searching for the homepage of `Sören Auer' we can add an additional property matching clause to the SPARQL query instead of directly using, for example, the property identifier foaf:homepage:
?subject ?slabel ?spattern. ?subject ?property ?value. ?property ?plabel ?ppattern. FILTER regex(?spattern, '.*Auer.*', 'i') FILTER regex(?ppattern, '.*homepage.*', 'i')

This also enables multilingual querying if the vocabulary contains the respective multilingual descriptions.
Creating fuzzy queries, of course, significantly increases the complexity of queries and will result in slower query answering by the respective SPARQL endpoint. However, since we deal with a distributed network of endpoints, where each one only stores relatively small documents this effect is often negligible.

Use subqueries for additional documents

In order to avoid situations where multiple agents retrieve the same documents (which is very probable in a small worlds scenario with a high degree of interconnectedness) it is reasonable to create query scripts, which only distribute certain tasks to the agent network (such as the retrieval of prospective information sources or document locations), but perform the actual querying just once locally.