ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001
202
More specifically, for a given term a, set B contains all terms
with unknown connection value to a. B = {b }, (i=1
i
keywords into independent categories, or called multiple
independent layers, the users can operate a query easier and
get better results close to the desire. Moreover, the query
speed will be improved.
Assume a document T is one of the query results of a, and
r contains a term set C. C = { c }, (j=1,...}. Set D is the
j
intersection of B and C. . D = {d }> (k=1
k
R represents the corresponding relevance relation of
k
d and a, m( d , a ). R is determined as
k k k
follows:
The rule to divide keywords into multiple independent layers is
described as follows:
Given A is a set of terms, B and C are subset of A, and
A = B u C . B={b } (j=1.-)- C={ c }
j k
(k=1,...). If for any j and k, J3 (b ,c ) = 0 and
j k
ß ( c , b ) = 0 . then A can be divided into two
k ’ j
independent layer B and C.
1. When a user input a query that contains only term a, a list
of documents including T are given as query results. The
decision on the degree how the document F is related to
the query is based on R with prior values .
k
2. The user examine the document T and indicates whether
it is relevant to a or not.
3. Then the value R is modified to either
P k
f
P = min[ 1, p * (1 + s )]
k k
or
11
P = max[ 0 , P * (1 - e )]
where e e [ 0 ,1 ] is the change scale. The query is
evaluated again with p ' and p
k k
4. The fuzzy value for this relation is adjusted to the value
which leads to the best answer.
The approach used to adjust the strength of a relevance
relation is simple and efficient. However, the performance is
possible to be further improved if the users’ feedback can be
classified. For example, the experts' feedback should have
more weight than those of amateur users'. It is an interesting
topic to identify the experts' feedback from the amateur users',
or more practically, to divide the feedback into groups, then
identify one group containing more feedback from experts than
other groups. The analysis of the feedback shows that the
feedback from experts has some different characteristics from
that from the amateur users on statistics. The results might be
used in No-Name in the future
MULTIPLE INDEPENDENT LAYERS STRUCTURE
Most search engines provide one textbox to users. Users input
all the keywords and organize them with some simple binary
combination like "And" and "Or". Some search engines
divides the keywords into several overlapped categories like
Yahoo. Some search engines has begun to divide the keyword
into independent categories. However, the importance for
search engines to divide keywords into independent
categories has not been fully recognized. In fact, by dividing
No-Name currently support two independent layer: Term layer
and Location layer (Figure 2). Term layer has a tree-like
structure containing about 2,000 terms. Location layer also has
a tree-like structure containing 7 continents and about 200
countries that are involved into GIS. The tree-like structure of
term layer and location layer makes them easy to expand and
be organized. Term layer and Location layer meet the
requirement to be independent. The terms in term layer has no
connection strength with the terms in location layer, and it
accords to the intuition: A document represented by term
"USA" has no "cause and effect" relation with a query
represented by "GIS", and vice versa.