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
 An overview of SemSearch

One major goal of this work is to hide the complexity of semantic search from
end users and to make it easy to use and effective for naive users. To achieve
this goal, we identified the following key requirements:
Low barrier to access for ordinary end users. Our semantic search
engine should overcome the problem of knowledge overhead and ensure that
ordinary end users are able to use it without having to know about the
vocabulary or structure of the ontology or having to master a special query
language.
Dealing with complex queries. In contrast with existing semantic-based
keyword search engines which only answer simple queries, our semantic
search engine should allow end users to ask complex queries and provide
comprehensive means to handle them.
Precise and self-explanatory results. Our semantic search engine should
be able to produce precise results that on the one hand satisfy user queries,
and on the other hand are self-explanatory. Thus, ordinary end users can
understand the results (e.g. what they are and why they are there) without
having to consult the back-end semantic data repositories or their underlying
ontologies.
Quick response. Our semantic search engine should provide quick response
to user queries, thus encouraging ordinary end users to harvest the benefit
of the semantic web technology. This requires that we make the mechanism
of semantic search as simple as possible.
To meet these requirements, we chose the keyword-based searching route
rather than the natural language question answering route, and deliberately
avoided linguistic processing which is a relatively expensive process in terms of
search. We overcome the limitation of current keyword-based semantic search
engines by supporting a Google-like query interface which supports complex
queries in terms of multiple keywords. Figure 1 shows a layered architecture of
our semantic search engine. It separates end users from the back-end heteroge-
neous semantic data repositories by several layers.
The Google-like User Interface Layer, which allows end users to specify
queries in terms of keywords. As will be described in Section 4, the Google-
like query interface extends traditional keyword search languages by allowing
the explicit specification of i) the queried subject and ii) the combination of
multiple keywords.
The Text Search Layer, which makes sense of user queries by finding out
the explicit semantic meanings of the user keywords. As will be described
in Section 5, central to this layer are two components: i) a semantic entitySemSearch:

A Search Engine for the Semantic Web 5



Fig. 1. An overview of the SemSearch architecture.

index engine, which indexes documents and their associated semantic entities
including classes, properties, and individuals; and ii) a semantic entity search
engine, which supports the searching of semantic entity matches for the user
keywords.
The Semantic Query Layer, which produces search results for user queries
by translating user queries into formal queries. This layer comprises three
components, including i) a formal query construction engine, which trans-
lates user queries into formal queries, ii) a query engine, which queries the
specified meta-data repository using the generated formal queries, and iii)
a ranking engine, which ranks the search results according to the degree of
their satisfactory on the user query. The mechanism of formal query gener-
ation will be described in Section 6.
The Formal Query Language Layer, which provides a specific formal
query language that can be used to retrieve semantic relations from the
underlying semantic data layer.
The Semantic Data Layer, which comprises semantic metadata that are
gathered from heterogeneous data sources and are represented in different
ontologies.
Figure 2 shows an overall diagram of the SemSearch search engine. It ac-
cepts keywords as input and produces results which are closely related to the
user keywords in terms of semantic relations. The search process of SemSearch
comprises four major steps:
Step1. Making sense of the user query, which is to find out the semantic
meanings of the keywords specified in a user query.
Step2. Translating the user query into formal queries.
Step3. Querying the back-end semantic data repositories using the gener-
ated formal queries.



Fig. 2. An overall diagram of the SemSearch search engine.

Step4. Ranking the querying results.
Step1 is carried out within the Text Search Layer. The rest of the steps are
associated with the Semantic Query Layer. In the following sections, we will first
describe the Google-like query interface (in Section 4). We will then come back
to detail the first two steps of the search process (Step 1 in Section 5 and Step2
in Section 6). The two further steps will be briefly described in Section 7 when
we describe the implementation of the search engine.