ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001
203
The independent layers structure can help the users to query
useful results. If a user wants to find out all documents about
GIS activities in China, he/ she can choose "China" in location
layer and perform the query. The result set will contain the all
the address of web sites that are involved into the GIS
activities in China. If a user wants to know the activities and
advancement about data capture technology in China, he/she
can choose "Data Capture" in term layer, and perform the
query in former result set. Users can also get a rough sketch
about the knowledge distribution in countries, or even cities in
the future, or in a special technology field. The information is
very useful for researchers.
The work to add a new independent layer, Application layer, is
now undergoing (Figure 3). Application layer contains about a
few hundred existing or potential applications for GIS
technology, such as Government, Education, Business, Public
Service etc. Application layer has a tree-like structure, and the
terms are organized mainly according to careers.
No-Name has limited power to support Natural Language input
and Linguistic Analysis. No-Name matches the natural
language input with the terms and some common words in
database. The common words includes "all", "some",
"somehow", "somewhat", "precise", "precisely", "accurate" ...
The common words are used in No-Name to set a threshold
for query results. A query needs all related documents will
generate more results than a query needs somehow related
documents, and a query needs somehow related documents
will generate more results than a query needs only precisely
matched documents. Conversely speaking, a query needs
precisely matched documents will generate results with higher
quality than a query needs somehow documents. No-Name
achieves the goal by assigning the words different fuzzy
values, like "all" equals to 0.1, "somewhat" equals to 0.5,
"precisely" equals to 1.0. The fuzzy values are then used to set
the threshold for query results., and a document can not be
output if its relevance degree to a query is less than the
threshold.
Figure 3 : Application Layer
Natural language support and multiple independent layer
structure enable No-Name to support complicate queries from
users. For example, when a user input "I want to know
something about remote sensing used in Canada
government". No-Name will first identify "Canada",
"Government", "Remote sensing", and "Something" from the
query, and then find the terms "Canada", "Government", and
"Remote sensing" in the independent layers, and perform a
query. "Something" will be assigned a fuzzy value as
threshold, and the final results will be achieved by filtering the
prior query results according to the threshold.
EXPERIMENTS
Introduction
Due to the limitation of free disk space available at server, the
records number in Database are limited to 20,000 with a total
size up to 300M. Some general search engine usually
generate more than thousands of results toward one query.
Generally speaking, the results after the first 200 are usually
not relevant to the query according to our experience. To avoid
wasting time and speeding our test, only the first 200 results
are accepted and analyzed.
Assessment on the search engines
While Yahoo and AlltheWeb are chosen to represent the
general-purpose search engines, Geo-Community is chosen to
represent the current GIS search engine. Their performances
are compared with that of No-Name by studying their retrieval
results according to same queries. An expert board with 3
persons is responsible to evaluate the query results.
Two term sets with 10 and 50 terms respectively are adopted
to represent the queries. Each query can only contain one
term. The results are shown in Table 2 and Table 3.
Relevance
results
Irrelevance
results
No-Name
443
172
Yahoo
391
677
AlltheWeb
227
985
Geo-Community
36
12
Table 2 : Query results of term set with 10 terms