Introduction - Multimedia Database Retrieval: Technology and Applications - page 9

Fig. 1.3 A structured P2P network system utilizing a DHT to link nodes so that a query can be
effectively and efficiently resolved. The nodes are clustered by the application of a self-organizing
method in a cluster-identification search system
multimedia objects requires a large number of query transactions to be sent from
nodes to nodes in the network. In comparison to a centralized datacenter, a retrieval
task requires an initial search of the relevant nodes owning multimedia objects
potentially relevant to the query, before it can perform similarity matching. In this
regard, the automatic clustering of multimedia objects on the network is desirable
for organizing the owner nodes in order to process the initial node search quickly.
Chapter 8 looks into an automatic clustering method for the organization of
nodes in a distributed hash table (DHT) for effective node searching and retrieval
of multimedia objects. A cluster-identification search system (CCS) is developed
to organize nodes as a structured peer-to-peer network. A structured Peer-to-Peer
(P2P) network, as shown in Fig. 1.3 , uses a DHT to link nodes so that a query
can be effectively and efficiently resolved. The automatic clustering allows the
partition of nodes on the network in the DHT and Chord layers, according to the
cluster identification. This indexing stage facilitates online search by pinpointing
the relevant nodes without traversing all the participating nodes.
While the automatic indexing of nodes is one important requirement for retrieval
systems in the cloud datacenter, increasing retrieval accuracy is another requirement.
In this regard, the pseudo-relevance feedback presented in Chap. 8 can be applied
to achieve sufficient improvement in retrieval accuracy without user supervision.
In this scenario, the relevant node performs pseudo-RF and forwards the modified
query to its neighbors. This offers continuously improving retrieval without trans-
ferring training samples over the network during adaptive searching.