Big Data Applications - Big Data: Related Technologies, Challenges and Future Prospects - page 64

been made, the performance of the existing automatic annotation methods still needs
to be improved. Currently, many efforts are being made to synchronously explore
both manual and automatic multimedia annotation [ 26 ].
Multimedia index and retrieval involve describing, storing, and organizing
multimedia information and assisting users to conveniently and quickly look up
multimedia resources [ 27 ]. Generally, multimedia index and retrieval include five
procedures: structural analysis, feature extraction, data mining, classification and
annotation, query and retrieval [ 28 ]. Structural analysis aims to segment a video
into several semantic structural elements, including lens boundary detection, key
frame extraction, and scene segmentation, etc. According to the result of structural
analysis, the second procedure is feature extraction, which mainly includes further
mining the features of necessary key frames, objects, texts, and movements, which
are the foundation of video index and retrieval. Data mining, classification, and
annotation are generated to utilize the extracted features to find the modes of video
contents and put videos into scheduled categories so as to generate video indexes.
Upon receiving a query, the system will use a similarity measurement method to
look up a candidate video. The retrieval result optimizes the related feedback.
Multimedia recommendation aims to recommend specific multimedia contents
according to users' preferences. It is proven to be an effective approach to
provide quality personalized services. Most existing recommendation systems can
be classified into content-based systems and collaborative-filtering-based systems.
The content-based methods identify users or general features in which the users
are interested, and recommend users for other contents with similar features. These
methods purely rely on content similarity measurement but most of them are limited
by content analysis and excessive specifications. The collaborative-filtering-based
methods identify groups with similar interests and recommend contents for group
members according to their behaviors [ 29 ]. Presently, a mixed method is introduced,
which integrates advantages of the aforementioned two types of methods to improve
the recommendation quality [ 30 ].
The U.S. NIST initiated the TREC Video Retrieval Evaluation detecting the
occurrence of an event in video-clips based on Event Kit, which contains some
text description related to concepts and video examples [ 31 ]. The research on video
event detection is still in its infancy. The existing research on event detection mainly
focuses on sports or news events, running or abnormal events in monitoring videos,
and other similar events with repetitive patterns. In [ 32 ], the author proposed a
new algorithm on special multimedia event detection using a few positive training
examples.

6.2.5 Network Data Analysis

Network analysis evolved from the initial quantitative analysis [ 33 ] and sociological
network analysis [ 34 ] into the emerging online social network analysis in the
beginning of twenty-first century. Many prevailing online social networking services