Database Reference
In-Depth Information
Fig. 2.6 The similarity measure in the content and context domains. ( a ) Semantic gap exists in the
content domain. ( b ) There might not be sufficient data to extract accurate contextual information
Fig. 2.7 Block diagram of
the integration of STRF and
LTRF in an adaptive retrieval
system. The solid and dashed
directed lines indicate the
information flow and the
user-controlled components,
respectively
User
Query image
Query
A user history
model
Likelihood
evaluation
based
on content
A priori
probability
evaluation based
on past retrieval
results
LTRF
A posteriori
probability
evaluation for
ranking images
User history
accumulation
STRF
Result display
and the context-based methods into a mathematically justifiable framework. In the
beginning, there is no available retrieval history from which to learn the context
model but the system can still work using the content-based component and incre-
mentally accumulate the retrieval results. When past retrieval results are available,
the context component of the system performs LTRF by extracting information
from the data gradually, which can be considered as a knowledge accumulation
process. When a user presents a query, the content component of the system learns
the user's information needs from the query through similarity measures and STRF.
If the context component has been trained by the time a user queries the database,
 
Search WWH ::




Custom Search