HYKE LogoSemantic Search


Home
Introduction
State of the art
An overview of SemSearch
The Google-like query interface
Making sense of the user query
Translating the user query into formal queries
Implementation and experimental evaluation
Conclusions and future work
1st Workshop on Semantic Search
2nd Workshop on Semantic Search
3rd Workshop on Semantic Search
4th Workshop on Semantic Search
 State of the art

In this section, we investigate how current semantic search approaches address
user support. We have identified four categories of semantic search engines, ac-
cording to the user interface they provide: i) form-based search engines, which
provide sophisticated web forms that allow users to specify queries, in the format
of choosing ontologies, classes, properties, and values; ii) RDF-based querying
languages fronted search engines, which provide sophisticated querying languages
1
http://semanticweb.kmi.open.ac.uk/semantic searhing.jsp/SemSearch:
A Search Engine for the Semantic Web 3
to support semantic search; iii) semantic-based keyword search engines, which en-
hance the performance of traditional keyword search techniques by making use of
available semantic data; and iv) question answering tools, which exploit available
semantic mark-up to answer questions asked in natural language format.
The SHOE search engine is one of the first form-based semantic search
engines. It provides sophisticated web forms that allow users to specify queries.
Such forms however are only suitable for those users who are fairly familiar with
the back-end ontologies and knowledge bases. Naive users have difficulties in
understanding these forms. Further, they have difficulties in formulating queries
using their own view on the information they aim to find.
The Corese search engine is an example of RDF-based querying language
fronted search engines. Other examples include the engines built in CS AKTive
Space and the SemanticWeb.org portal
2. Such search engines usually provide
a sophisticated querying language to support semantic data querying. However,
in order to be able to ask queries with these search engines, end users will have to
be fairly familiar with both the back-end ontologies and the provided querying
language.
The TAP search engine and the search engine presented in are examples
of semantic-based keyword search engines. The search process of such search
engines often comprises two major steps: i) finding an instance match for the
user keyword and ii) retrieving instances which are closely related to the instance
match of the user keyword. Such search engines often provide comprehensive
means to support the clustering of search results.
AquaLog and ORAKEL are examples of ontology-based question an-
swering engines. They make use of natural language processing technologies to
reformulate natural language queries into ontological triples (e.g., in AquaLog)
or into specific query languages (e.g., in ORAKEL). While these tools appear to
be ideal for naive users, their performance on searching is heavily influenced by
the performance of the used natural language processing techniques.
All the tools described above are able to enhance the search performance by
making use of available semantic data and their underlying ontologies. With the
partial exception of ontology-based question answering tools, state-of-art tools
are however not suitable for naive users. One problem is knowledge overhead,
which is requiring users to be equipped with extensive knowledge of the back-end
ontologies and knowledge bases (e.g. form-based search engines) or specific SQL-
like querying languages (e.g. RDF-based query language fronted search engines)
in order to be able to formulate queries or to understand the search result.
Another problem is the lack of support for answering complex queries
presented by current semantic-based keyword search engines. These search en-
gines are often only able to accept one keyword as input and give back the
semantic entities which are related to the keyword as results. Relation centered
search that finds relations between multiple keywords is not supported. This
greatly limits the scope of user queries. For example, current semantic-based
keyword search engines typically could not even handle simple queries where
two keywords are involved such as news about phd students.