WU et al.: SURVEY ON VISUAL ANALYTICS OF SOCIAL MEDIA DATA 2137
Fig. 1. Classification of visual analysis approaches for gathering information on social media.
Fig. 2. Visual Backchannel [2] introduced to visualize online conversations
regarding a large-scale event on Twitter. The system integrates Topic Streams to
represent topical development, People Spiral to indicate the activity of par-
ticipants, Post List to show the recent posts, and Image Cloud to display
shared photos.
community depict this issue as information overload and pro-
pose various methods to address this issue.
Visual Backchannel (Fig. 2) was developed to follow and
explore online conversations regarding a large-scale event on
social media for general users [2]. Notably, the event to fol-
low is defined by a manually-specified keyword or term. Tweets
regarding an event can be continuously retrieved via Twitters
open API. The collected tweets are subsequently processed to
remove stop words and merge similar words, which results in a
number of word stems that are regarded as topics. Apart from
the traditional post listing, Visual Backchannel includes three
novel interactive visualizations, namely, topic streams demon-
strating topic evolution, people spiral indicating participants and
their activity, and image cloud displaying shared photos. Topic
streams are a primary view of Visual Backchannel that uses
stacked graphs to visualize the dynamic changes in the frequen-
cies of word stems over time from a live-changing social stream,
such that both current and previous changes in the backchannel
conversations can be clearly displayed. With such coordinated
views, Visual Backchannel provides a visual summary of the
backchannel conversations from temporal, topical, social, and
pictorial facets.
Apart from helping general users seek information, social me-
dia have become valuable sources of newsworthy information
for domain experts. Several interactive visualization systems
have been developed to assist journalists in their search for in-
formation from social streams. Vox Civitas was designed and
developed to help journalists and media professionals extract
valuable news based on large-scale visual aggregations of social
media contents [3]. Apart from the simple keywords specified by
users, four types of automatic content analysis methods, namely,
relevance, uniqueness, sentiment, and keyword extraction, are
utilized to assist users in searching and filtering tweets related
a large-scale event. Vox Civitas offers an easy-to-use user inter-
face, which aligns a video of an event to several simple views,
such as keyword and message volume graph views. However,
Vox Civitas still has several limitations, such as information
overload, lack of trustworthiness, and no support for situational
awareness.
To address the issues of information overload and lack of
context, Zubiaga et al. [4] introduced a user interface, Tweet-
Gathering, which Twitter users could easily adapt. It integrates
flexible filters, ranks trending keywords according to their news-
worthiness, displays representative tweets to hasten information
access, and add necessary context to short tweets.
Although social media offer abundant and valuable informa-
tion to journalists, finding reliable and trustworthy information
is difficult. Diakopoulos et al. [5] presented a visual analy-
sis system called “Seriously Rapid Source Review” (SRSR)
that would enable journalists to find and assess information
sources on social media. SRSR uses an eyewitness detector
obtained from a dictionary-based technique to extract first-
hand, on-the-ground tweets regarding breaking news. These
authors also used a k-nearest neighbor method to classify in-
formation sources (users) into three categories, namely, organi-
zations, journalists/bloggers, and ordinary users. SRSR offers
an intuitive visual interface to help journalists find and assess
information sources.
Maintaining geographically-grounded situational awareness,
which is critical for crisis management, has received much at-
tention. SensePlace2 [6] was developed to support situational
awareness by displaying spatiotemporal information of social
streams. It adapts a heatmap visualization technique to show the
frequencies of the retrieved tweets with respect to a particular
topic. Thus, professionals can flexibly use temporal and spatial
filters to obtain their desired information from huge volumes of
information on social media.