Presentation of Sung Pil Moon, Yikun Liu, Steven Entezari, Afarin Pirzadeh, Andrew Pappas, and Mark Pfaff on the topic "Top Health Trends: An information visualization tool for awareness of local health trends" at ISCRAM2013
Top Health Trends: An information visualization tool for awareness of local health trends
1. An information visualization tool for awareness of local health trends
Top Health Trends
Sung Pil Moon, Yikun Liu, Steven Entezari, Afarin Pirzadeh, Andrew Pappas, and Mark Pfaff
May 15, 2013
2. May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Table of contents
• Introduction
• Background
• Top Health Trends Application
• Evaluation with Domain Experts
• Limitations and Future Work
3. 1. Introduction
• Detecting and monitoring emerging
crisis events is vital to prevent
negative impacts on public health.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
• Researchers have started
detecting and predicting
emerging events from social
network services data due to:
o explosive growth rate
o frequency
o message volume
o public availability
4. 1. Introduction
• Many information visualization tools with visual analytic techniques
have been developed to effectively and efficiently present SNS (Social
Network Services) data
• Primary foci of prior work:
o Novel methods of capturing and rendering data on a map overlay
o Limited attention to usability
o Limited attention on improving situation awareness
(MacEachren et al, 2011; Hodgson, 2012a).
• Situation awareness (SA) is crucial to make effective decisions for
appropriately controlling a broad range of complex, dynamic, and high-
risk situations (Endsley, 1988).
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
5. 1. Introduction
• Our Top Health Trends application
o An information visualization tool to support awareness of local
health-related trends based on Twitter data.
o Our application displays the most frequently tweeted health-
related topics in both a ranking chart and a map.
o It also provides trend graphs of tweets and tag clouds of related
keywords mentioned with the tweets so that users of the tool can
better aware of emerging public health issues.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
6. 2. Background
2.1. Trend detection with twitter data
• Explosive growth of social media activity with hundreds of millions of
active users (Nielson Inc., 2012)
o Social media and blogs reach 80% of all active Internet users in the
US (245 million).
o In 2012, Twitter had 108 million accounts in the US, 465 million
worldwide, and 175 million tweets daily (Hodgson, 2012b)
• Twitter in particular has received much attention as a new research
data source due to message volume, frequency, and public availability
(Culotta, 2010).
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
7. 2. Background
2.1. Trend detection with twitter data
• Examples of tools using SNS data
o Social Network Enabled Flu Trends (SNEFT) framework.
(Achrekar, Gandhe, Lazarus, Yu, & Liu, 2011)
- Monitored Twitter messages to track and to predict the
emergence and spread of an influenza epidemic in a population
- Showed that Twitter messages could be used for almost real-
time prediction of influenza activity trends two weeks earlier
than current Influenza-Like Illness activity (ILI) reports by the
Center for Disease Control and Prevention (CDC)
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
8. 2. Background
2.1. Trend detection with twitter data
•Examples of tools using SNS data
o Earthquake detection system using Twitter users as social
sensors. (Sakaki, Okazaki, & Matsuo, 2010)
- Used Twitter users as social sensors and monitored Twitter data
to detect earthquakes,
- Detected earthquakes faster than the Japan Meteorological
Agency (JMA) with high probability (96% of earthquakes)
•If you know what you’re looking for, you can probably find it
somewhere in Twitter
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
9. 2. Background
2.2. Information visualization tools
•Definition
o Defined as “the process of presenting information in ways that
are naturally compatible with human visual capabilities”
(Gershon, Eick, & Card, 1998)
•Benefits
o improved efficiency
o more interactivity
o increased user satisfaction
o reduced cognitive loads
o broadened insights about data and information
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
10. 2. Background
2.2. Information visualization tools for Twitter
• TwitVis (D. Ferreira, Freitas, Rodrigues, and V. Ferreira,
2009)
o An information visualization tool allowing Twitter users to identify
and to discover social networks of both known and unknown users
with related interests.
o Provides two main visualizations:
o An egocentric network view showing a network of friends in a
graph form, and
o The keyword view showing the relationship between the
keywords of interests and users tweeting about them
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
11. 2. Background
2.2. Information visualization tools for Twitter
• TwitInfo (Marcus et al, 2011)
o A microblog-based event tracking interface which can aggregate,
summarize, and visualize tweets about user-specified events based
on their social streaming algorithm for automatic identification of
event tweet peaks.
o When a user clicks on a event peak in the interface, it automatically
filters and refreshes its map, links, tweet list, and sentiment graph in
the time period of that peak.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
12. 2. Background
2.2. Information visualization tools for Twitter
• SensePlace2 (MacEachren et al, 2011)
o Designed to support crisis management by providing overview and
detail maps of Tweets, filtering its place-time attribute, and
supporting an understanding of temporal and spatial patterns of
activities, events, and attitudes.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
13. 2. Background
2.3. Situation Awareness
•Definition
o “the perception of the elements in the environment within a volume
of time and space [SA Level 1], the comprehension of their meaning
[SA Level 2], and the projection of their state in the near future [SA
Level 3]” (Endsley, 1988, p. 97)
o Having higher SA is important to make effective decisions in
urgent, dynamic, and high-risk situations
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
14. 2. Background
2.3. Situation Awareness
• Maintaining high SA is challenging:
o Information overload
o Imperfect data from multiple data sources
o Complexities of systems and data mining techniques
o External factors such as high stress, time pressure, and inability
to respond to unusual and unexpected incidents
(Cameron, Power, Robinson, & Yin, 2012)
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
15. 2. Background
2.3. Situation Awareness
•Classification of the use of visualization to enhance SA (D’Amico &
Kocka, 2005)
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
16. 3. Top Health Trends
3.1. Overview
•The Top Health Trends application was designed to improve the ability of
public health officials to quickly recognize local public health crises though
social network (Twitter) data.
•Produced for a competition run by the US Department of Health and
Human Services, Office of the Assistant Secretary for Preparedness and
Response (ASPR): The Now Trending Challenge (Prize: $22,000)
•The ASPR requested an application that:
1. Mines and filters Twitter data
2. Analyzes it according to a provided lexicon of terms related to 25
health conditions
3. Produces a visualization of health-related tweets to identify the top
trending health threats within user-selected geographic regions
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
17. 3. Top Health Trends
3.2. System Design
• Harness the open-source APIs and natural language
toolkits for actionable intelligence
o Twitter API and MapQuest API to collect Tweets and present the
information in a geographically situated fashion
o Natural Language Tool Kit (NLTK; http://nltk.org) to provide
ready-to-use API entries for performing text cleansing and
analyzing tasks
o Server-side processing (MySQL, PHP, Java, and others) and
client-side rendering of data (Adobe Flex and Javascript)
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
18. 3. Top Health Trends
3.3. Main Components
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Tweet Map
• Displays all tweets of the disease on a
geographic area of user’s interest.
• Updates tweets on an interactive map
o This helps filter irrelevant information
and improve decision quality and
reduce cognitive efforts (Eick and
Wills, 1995; Lurie and Mason, 2007)
• Enables visual exploring, monitoring,
and inspecting to support the
projection and comprehension
stages of SA.
19. 3. Top Health Trends
3.3. Main Components
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Top trend ranking chart
• Displays the top trending illnesses in
the selected area, including
o The previous day’s ranking,
o The number of days on the top ten
o The peak rank during the last 7 days
• Use of the metaphor of the ‘Billboard
Top 100’ chart format allows the user to
o Perceive the current situation
o Comprehend underlying trends
o Support the users prediction skills in a
familiar and accessible setting.
20. 3. Top Health Trends
3.3. Main Components
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Previous day’s rank: Shows if this condition is trending
upward (value < current) or downward (value > current)
Days in the top 10: Shows recency (if low) and persistence (if high)
Peak Rank in past 7 days: Shows trend trajectory and context
Current Rank
21. 3. Top Health Trends
3.3. Main Components
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Trend graph
• Shows the trending changes over the
past 24 hours in 3 hour increment.
o Tweet counts change
o Percentage change
o Ranking change
• Uses a simple line graph visualization
o Most common ways to show trends
(Robertson et al., 2008)
o It is easy to read, and enables users to
make predictions of future values.
o Support all three stages of SA
(monitoring, inspecting, and
forecasting)
22. 3. Top Health Trends
3.3. Main Components
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Related keywords
• Shows the most frequently mentioned terms that are tweeted along
with illness.
o This provides useful context to help users determine the
accuracy of ranked terms (contextual hints)
23. 3. Top Health Trends
3.3. Main Components
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
Tag cloud of the disease
• Provides a broad keyword overview
in addition to the five top related
keywords listed for each illness.
o Shows up to 20 related keywords
and their frequency of appearance
• Supports better sensemaking
o A tag cloud is popularly used
because it places the data in a
limited space and uses tag size to
draw attention to the most important
items (Hearst and Rosner, 2008)
24. 4. Evaluation with Domain
Experts
4.1. Evaluation settings
• Main goals:
o To identify how our application can aid daily work for a variety of
domain experts
o To find unexpected problems and opportunities for
improvements for the next generation of the tool
• Participants
o Total 4 participants having at least one year of working
experience in a health-related domain.
o Two EMS experts, one doctor, and one pharmacist
o 3 females and 1 male.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
25. 4.1. Evaluation settings
•Procedure
1) To interact with the application through 19 simple tasks, which gave
users opportunities to experience features of the application. For
example:
• “What is the name of the disease currently ranked number 3 in
Indianapolis?”
• “Show all tweet icons mentioning ‘Common Cold’ on the map.”
• “What was the rank of this disease 9 hours ago?”
2) To fill out a post-task questionnaire asking
- Comprehension of the information,
- Degree of helpfulness for being aware of local health trends
- Degree of helpfulness to their daily work
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
4. Evaluation with Domain
Experts
26. 4.2. Evaluation results
•[+] Positive feedback
o All participants were able to perform the 19 tasks correctly
o All found the application
- easy to understand
- informative
- visually pleasing
- easy to use
“quite easy to use, because it is self-explanatory”
“This application has potential…in many areas. As a
pharmacist, the tool will help me knowing when I need to
order drugs.”
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
4. Evaluation with Domain
Experts
27. 4.2. Evaluation results
• [-] Two main Concerns
o Invisibility of data filtering techniques behind the visualization
o Frequent appearance of unrelated co-tweeted terms reduced
trust in the application
o Credibility of intelligence generated by Twitter data
o Twitter is still not considered a reliable source of information, at
least not by itself.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
4. Evaluation with Domain
Experts
28. 4.2. Evaluation results
• [+] Suggestions by participants
o Integration with official health data
- For example: U.S. Centers for Disease Control Influenza-
- like Illness (CDC ILI) report
o Customizable ranking and trends graphs
o Notification alerts as the number of tweets for certain disease
exceeded a given threshold
o Cross-referencing the application with external sources
- Such as weather or news reports
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
4. Evaluation with Domain
Experts
29. 5. Limitations and Future works
5.1. Limitations
• Small sample size for the evaluation
• Natural language processing of Twitter data
o Use of slang, jargon, and abbreviations decreases its accuracy
• Lack of geolocation info in most tweets
o Only 1% of tweets have geolocation information
• Unrealistic list of disease and symptom terms
o Originally provided for the Now Trending Challenge by ASPR.
o Terms more likely to be used by a limited number of people (e.g.,
medical professionals)
• Accuracy of filtering algorithm behind visualization
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
30. 5. Limitations and Future works
5.2. Future Work
• Revise application based on preliminary feedback and re-
test with larger population
• Refine natural language processing of Twitter data
o Very hot area of research
• Indirect determination of geographic location (from profile
information, keywords, social networks, etc.)
• Improved list of disease and symptom terms
o Drawn from clinicians, emergency room patient complaint data,
archival sources, etc.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends
31. Thank you!
Questions?
Contact:
GRAPPA Lab (GRoup Psychology and Performance)
School of Informatics
Indiana University-Purdue University Indianapolis
http://www.iupui.edu/~grappa/
.
May 15, 2013Top Health Trends: An InfoVis Tool For Awareness of Local Health Trends