Community-Contributed Media Collections: Knowledge at Our Fingertips - Community-Built Databases: Research and Development - page 31

2.4 Taxonomies on Research Issues
A lot of investigation has been carried out on community-contributed media
collections to improve the user searching activity and document classification of
media collections. To provide a clear overview of the research efforts, we propose
three taxonomies (see Figs. 2.2 - 2.4 ), one for each media collection, to sum up the
main issues addressed in the research.
Figure 2.2 summarizes the works addressing video-sharing communities. A first
class of approaches is aimed at studying whether and how meaningful, coherent
annotations can be derived from the collaborative tagging effort of community
users. The second group of research activities is instead focused on the exploitation
of these video collections from a data mining perspective. In the works reviewed,
video clips and their associated tags are exploited to (a) train concept-based classi-
fiers and improve query representations in a media retrieval framework or (b)
generate relevant metadata about captured events and enhance the user viewing
activity.
Figure 2.3 depicts the taxonomy proposed to classify the works focused on
photo-sharing communities. Research efforts can be grouped according to the
addressed topic, like tag recommendation to effectively support photo annotation,
and automatic extraction of semantics. Among the tag annotation works, two main
approaches have been proposed. The first one resorts to basic techniques that
consider tag frequencies in the past to suggest useful and relevant tags, while the
second one exploits collective knowledge, residing in the Flickr community, to
support photo annotations. Furthermore, numerous works have been devoted to
exploiting data mining techniques to (a) discover location and event semantics from
Fig. 2.2 YouTube:
taxonomy of discussed
approaches