Information Technology Reference
In-Depth Information
perceptual sensibility, capable of complex social relationships, and possessing a
certain moral sophistication and an intrinsic moral value.” In this regard, inference
only serves as noise, adding irrelevant terms to the language models. For example,
adding 'sophistication' to a query about 'Britney Spears' will likely not help
discover relevant documents. Inference would be useful if it produced surprising
information or reduced ambiguity. However, it appears that at least for simple RDF
Schema vocabularies, information higher in the class hierarchy is usually knowledge
that the user of the search engine already possesses (like Britney Spears is a person)
and that the reduction of ambiguity is already done by the user in their selection of
keywords. However, it is possible that more sophisticated inference techniques are
needed, and that inference may help in specialized domains rather than open-ended
Web search. Further experiments in parametrization of inference would be useful
given that our exploration in this direction showed no performance increase, only
performance decrease.
6.8
Deployed Systems
In this section we evaluate our system against 'real-world' deployed systems.
One area we have not explored is how systems based on relevance feedback
perform relative to systems that are actually deployed, as our previous work has
always been evaluated against systems and parameters we created specifically for
experimental evaluation. Our performance in Sects.
6.5.1.1
and
6.5.2.1
was only
compared to baselines that were versions of our weighting function without a
relevance feedback component. While that particular baseline is principled, the
obvious needed comparison is against actual deployed commercial or academic
systems where the precise parameters deployed may not be publicly available and
so not easily simulated experimentally.
6.8.1
Results
The obvious baseline to choose to test against is the Semantic Web search engine,
FALCON-S, from which we derived our original Semantic Web data in the
experiment. The decision to use FALCON-S as opposed to any other Semantic Web
search engine was based on the fact that FALCON-S returned more relevant results
in the top 10 than other existing semantic search engines at the time using a random
sample of 20 queries from the set of queries described in Sect.
6.2.2
. Combined with
the explosive growth of Linked Data over the last year and the changes in ranking
algorithms of various semantic search engines, it is difficult to judge whether a given
Semantic Web search engine is representative of semantic search. However, we
would find it reasonable that if our proposed hypothesis works well on FALCON-S,
it can be generalized to other Semantic Web search engines.
Search WWH ::
Custom Search