Information Technology Reference
In-Depth Information
tained in the profiles of the peers. To save time
during the computation of the ranking, instead
of measuring the actual similarity (through the
DTW measure), the
LB_Keogh
value is computed.
The resulting algorithm is denoted as ISMS (ISM
with sampling).
ISMS is expected to have slightly larger traf-
fic than
>
RES, since it propagates the sample of
an answered query to all nodes involved in the
search (in order to update their profiles). However,
by testing the content of the queries, it tries to
reduce the number of missed answers.
The work proposed by Karydis et al. (2005)
focuses mainly on developing a framework for
similarity searching algorithms using a sampling
method, while musical pieces are treated as se-
quences and their similarity based on DTW. Ac-
cordingly, three algorithms (BFS, >RES and ISM)
used for searching similar text documents in P2P
networks are examined and adapted for CBMIR.
The use of DTW proves too expensive and thus a
lower bound is used therein. Though, the cheap
lower bound calculation advantage is alleviated
whenever a real match is available, by requiring
the full DTW measurement. Nevertheless, DTW's
characteristic being able to withstand distortion
of the comparing series in the time axis has had
great appeal in the MIR research community.
ization feature and a common method to obtain
such features. The coordinator consists of a data
structure to store the network identifiers of all
peers, a data structure to store the PC features of
all peers and an accelerating structure which can
utilize the peer features to locate proper peers for
faster CBMIR. The query process steps are the
following: Each peer shares musical content and
registers at the coordinator information including
its network identifier and PC feature. Upon a query
the peer feature of the querier, that is, the peer
feature of the music set which includes only the
music request, can be extracted from the music
request by the PC feature extraction method and
sent to the coordinator. Based on the PC feature of
the music request and the accelerating algorithm,
the coordinator selects a high probability subset
of the peers which may have similar music to
the query and the querier directly addresses this
subset. In this case, these peers act as some short
of super-peers.
In this approach, the balance of stability and
cost can be obtained by selecting the proper num-
ber of high probability peers. Moreover, Wang et
al. (2002) report that PsPsCM has the best load
and time attitude in comparison to the family of
presented models, thus supporting the further
development of the accelerating algorithm solely
on this model.
Hybrid
cBmIr In WIreless p2p
netWorks
The fourth model proposed by Wang et al. (2002),
PsPsC, is a hybrid approach that combines char-
acteristics from the centralized models PsC &
PsC+ (Section “Centralized”) and the decentral-
ized unstructured PsPs (Section “PsPs Model”),
as previously described.
As already discussed in the introduction of this
chapter, the sovereign of the traditional music
distribution has undergone a significant alteration
under the auspices of new technologies like MP3
and the penetration of the World Wide Web. Brand
new opportunities for music delivery are addition-
ally introduced by the widespread penetration of
the wireless networks (wireless LANs, GPRS,
UMTS as described by DeVriendt, Lerouge, &
Xu, 2002) such as the pioneering applications
PsPsC Model
In the PsPsC model there is one coordinator peers
registers at, which collects and manages their
statistical data. The architecture of the system
is similar to the PsPs model (see Section “PsPs
Model”) with the addition of a peer character-
Search WWH ::
Custom Search