Community Detection in Collaborative Tagging Systems - Community-Built Databases: Research and Development - page 128

using the HCD tag communities results in far better tag recommendations than by
using the CNM ones across all three datasets. For instance, in FLCKR-1M dataset,
the HCD-based recommendation achieves more than six times higher precision
than the CNM-based one (27.42% compared with 4.46%). A large part of the failure
for the CNM-based tag recommendation can be attributed to the few gigantic
communities that dominate its community structure. Also, in the case of HCD-
based tag recommendation, there were numerous cases in which it was not possible
to provide any recommendations due to the fact that the input tags did not belong to
any community (HCD leaves several tags unassigned to communities).
Obviously, one could employ a fall-back tag recommendation scheme (e.g.,
based on simple tag co-occurrence) to address cases where no recommendation
can be made based on HCD communities. Alternatively, it is also possible to
directly exploit the hub and outlier information of the HCD structure in order to
improve the performance of the respective tag recommendation method. In this
study, we wanted to focus on the comparison between the community structures
of two methods. In the future, we plan to develop more sophisticated approaches
for tag recommendation based on tag communities and compare them with existing
tag recommendation schemes (e.g., FolkRank [ 2 ]).
Apart from tag recommendation, HCD can be also used as a means of dis-
covering the different usage contexts/senses of tags. Since HCD identifies hub tags
that are adjacent to multiple communities, we can assume that at least some of
them correspond to different usage contexts for these tags. For instance, Table 5.8

Table 5.8 Examples of different usage contexts for three sample hub tags (one for each dataset).
Context Tag community
Networks (BIBSONOMY-200K)
Network analysis

socialanalysis, socialdynamics, sociophysics, autocorrelation, swarm,
percolation, selforg, fitnesslandscapes, socialstructure, ...

Epidemic spread large, spreading, enidtl, kcore, dtl, pysics, epidemic, gda, vespignani,

networks, delivery, pastorsatorras

Bibliometrics citation, reputation, publication, bibliometrics, academics, impact, ...

VoIP networks voip, phone, skype, telephony, jajah, vonage, internettelephony, ...

Security proxy, encryption, wlan, password, openssl, cracker, https, wep, ...

Bridge (FLCIKR-1M)
Bac de Roda bacderoda, puente de calatrava, bru, puente luminoso, structural, ...

Bridge of Sighs
(Bisbe)

carrer del bisbe, bridge of sighs, footbridge, cobblestone, bascule, ...

Port bridge port, puerto, sea, port vell, ship, boat
Pont del Centenari transporte, pont, montserrat, new
Bean (DELICIOUS-7M)
Plant seeds lentils, blackbean, greenbean, kidneybeans, hummus, chickpeas, ...

Coffee beans grown, arabica, robusta, pricey, coffea, wholesaler

Bean bag toffet, velveteen, crushed, swirl, tuffet, herbagreeen, swirled

Java beans spring, hibernate, jsf, ejb

guitars, fender, musicstuff, giger, telecaster, strat, stratocaster, ... Travis Bean

courtneylove, cobain, francesbean, lindacarroll, thevoid, girlgerms, ... Frances Bean

Cobain