Database Reference
In-Depth Information
with higher RF iteration. As we may have anticipated, the multi-click RF performs
better than single-click RF, but with a higher number of clicks (i.e., a high level of
user involvement). In order to compare the performance between the two methods
at the same level of number of clicks, a plot of the precision versus clicks is done, as
illustrated in Fig.
8.9
b. Since some queries may have different numbers of feedback,
it is observed that the precision is more versatile against clicks than feedback
iterations. In general, the retrieval precision versus clicks plots for both methods
indicate the relationship resembles an increasing function. It is observed that single-
click RF outperforms multi-click RF, for the same level of user interactions.
8.4
Social Network Image Retrieval Using
Pseudo-Relevance Feedback
As an alternative to the structured P2P architecture, this section presents a method
for social network image retrieval on an unstructured P2P network. The search
system works by broadcasting the request message with non-zero time-to-live (TTL)
to all the neighboring hosts. This is to form a social network group and perform
a search within this group. Each peer in the P2P network maintains two tables
of neighbors. The first type of neighbors are called the
generic neighbors
which
typically represent the neighbors with the least physical hop counts. The other type
of neighbors are called the
community neighbors
and they have a common interest
which is shared among the community. Two stages of operations are required:
social
network discovery
and
query within the social network
.
Automation in adaptive image retrieval is employed to improve retrieval accu-
racy, and to reduce the task of transferring actual image files over a network as
required in user-controlled RF. The SOTM is employed for implementing pseudo-
RF, and new techniques are utilized to improve its performance in automatic
retrieval. This helps reduce the bandwidth requirement and subjective errors caused
by user feedback in the scenario of distributed content-based retrieval.
8.4.1
Social Network Discovery
Figure
8.10
a illustrates the process of community neighborhood discovery for the
construction of a social network. A peer node originates the query request for its
generic neighbors in the P2P network. Whenever a peer node receives a query
request, it will (1) decrement the TTL, and forward the request to the generic
neighbors when TTL
1, and (2) perform a content search within the peer's feature
descriptor database. The retrieval results of each peer are transmitted to the original
query peer directly in order to improve the efficiency.
>
Search WWH ::
Custom Search