Database Reference
In-Depth Information
Web usage mining aims to mine auxiliary data generated by Web dialogues
or behaviors. Web content mining and Web structure mining use the master
Web data. Web usage data includes access logs at Web servers, logs at proxy
servers, browsers' history records, user profiles, registration data, user sessions
or trades, cache, user queries, bookmark data, mouse click and scroll, and any
other kind of data generated through interaction with the Web. As Web services
and the Web2.0 are becoming mature and popular, Web usage data will have
increasingly high variety. Web usage mining plays key roles in personalized space,
e-commerce, network privacy/security, and other emerging fields. For example,
collaborative recommender systems can personalize e-commerce by utilizing the
different preferences of users.
6.2.4
Multimedia Data Analysis
Multimedia data (mainly including images, audios, and videos) have been growing
at an amazing speed. Multimedia content sharing is to extract related knowledge
and understand semantemes contained in multimedia data. Because multimedia
data is heterogeneous and most of such data contains richer information than
simple structured data and text data, extracting information is confronted with
the huge challenge of the semantic differences of multimedia data. Research
on multimedia analysis covers many disciplines. Some recent research priorities
include multimedia summarization, multimedia annotation, multimedia index and
retrieval, multimedia suggestion, and multimedia event detection, etc.
Audio summarization can be accomplished by simply extracting the prominent
words or phrases from metadata or synthesizing a new representation. Video
summarization is to interpret the most important or representative video content
sequence, and it can be static or dynamic. Static video summarization methods
utilize a key frame sequence or context-sensitive key frames to represent a video.
Such methods are very simple and have been applied to many business appli-
cations (e.g., Yahoo!, Alta Visa, and Google), but the playback performance is
poor. Dynamic summarization methods use a series of video clips to represent
a video, configure low-level video functions, and take other smooth measures to
make the final summarization look more natural. In [
25
], the authors proposed a
topic-oriented multimedia summarization system (TOMS) that can automatically
summarize the important information in a video belonging to a certain topic area,
based on a given set of extracted features from the video.
Multimedia annotation inserts labels to describe contents of images and videos
in both syntax and semantic levels. With the assistance of such labels, the manage-
ment, summarization, and retrieval of multimedia data can be easily implemented.
Since manual annotation is both time and labor intensive, multimedia automatic
annotation without any human interventions becomes highly appealing. The main
challenge for multimedia automatic annotation is semantic difference, i.e. the
difference between low-level features and annotations. Although much progress has
Search WWH ::
Custom Search