State-of-the-Art: Semantics Acquisition and Crowdsourcing - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 10

Since only a fraction of theWeb is coveredwith descriptive semantics, the applica-
tions utilizing them are also limited. Typically, a “semantic application” is dedicated
to some specific corpus (e.g., an e-shop search and navigation supported by product
descriptions). Heavyweight semantics are not common—applications rely more on
lightweight semantics (e.g., product taxonomies).
Today, the query-based search fueled by keyword indexing approaches represents
a dominant information retrieval paradigm. A majority of Web users utilize it as
their primary way to satisfy their information needs. It also has many drawbacks:
expressiveness limits [ 14 ], keyword ambiguity [ 36 ], invisibility of information space
[ 44 ], just to mention a few. To solve these issues, some researchers and practition-
ers suggest a radical change of search paradigm (e.g., exploratory search), others
argue for solutions that would not disturb the user who is unwilling to change his
keyword search habits (e.g., result re-ranking, query expansion). Though, almost all
alternatives somehow rely on semantics.
2.2.1 Query-Based Search
An example of the use of semantics for improving keyword-based search is solving
a problem of term meaning disambiguation [ 33 ]. Searchers utilizing keyword search
often encounter problem with homonyms used in the query—their result set gets
spoiled with irrelevant (from their perspective) results, because the search terms have
multiple meanings. Not all users are able to overcome such issues by themselves.
Researches employed different strategies for solving the search term ambiguity
issue. The utilization of semantics plays a vital role in them. Köhler et al. [ 33 ]used
existing ontologies to index a corpus of websites. They disambiguated terms within
the websites using term relationships from the ontologies and were therefore able to
infer related concepts, not just keywords. Using the same strategy for search queries,
they were able to match relevant search results more accurately.
Another approach to term disambiguation implements the modeling of the
searcher. The idea is to track the user's desires, long term interests or context, repre-
sent it in a formal model and measure its relatedness to the potential search results
either by query enhancements or result filtering. Comparison is used to re-rank the
results, so they satisfy the searcher. Though some researchers [ 6 , 36 ] attempt to do
it on the syntactical (keyword) level, much better results were achieved, if the user
and resources were modeled using “heavier” semantics [ 5 ].
The search query expansion approaches [ 4 , 9 ] directly involve searcher's actions
in query disambiguation. Normally, when searcher formulates an ambiguous query,
he tries to reformulate it by introducing other keywords, if he can think of any.
The point of query-expansion approaches is to aid him with this by recommending
a possible search terms to append to the query and to refine the search. Prior to this,
a domain model is required to provide relationships of terms in the original query,
to other possible search terms.