Upload
xiaoyu-wang
View
419
Download
7
Tags:
Embed Size (px)
DESCRIPTION
In an enterprise environment, business information, such as project proposals and product discussions, is dynamic and often embedded in documents and document activities (e.g, emails, Web pages and office documents). Because this information is essential to business processes, corporate employees need an effective means to retrieve it. Some commercial products, including Google Desktop, provide keyword searches for finding some of this information. However, this approach is not always effective as successful keyword searches can be difficult to construct, and even the best queries may fail to find some important materials. In this paper, we present Taste (Temporal Activities and Story TElling), an interactive visual analytics system that enhances the enterprise employee’s capabilities in searching and sharing diverse and dynamic business information. Taste was designed, after interviews with corporate employees, to follow their information retrieval cues and help them manage, review and share the business information embedded in their document activities. Results of our lab and field studies validate that Taste provides employees the confidence and necessary features to more efficiently and effectively retrieve business information from their documents and activities
Citation preview
Finding Business Information by Visualizing Document Activity
Charlotte Visualization Center
Xiaoyu Wang
Palo Alto Research Center
Bill Janssen and Eric Bier
Palo Alto Research Center
Problem Statement• The growth of business data exceeds our
ability to find desired information
• The task to find information is exhausting:
• 3.7 hours/week on search but not finding information
• 2.5 hours/week on creating data, which can’t be found in the future
Without efficient way of finding business information,
enterprise may suffer from less effective business operations.
--- Interactive Data Coorp.
OverviewOverview Future DirectionFuture DirectionCurrent Imp.Current Imp. EvaluationEvaluation
Where’s the Business Information?
• Embedded with employee’s daily document activities
• Presented in heterogenous data/applications
• Stored in local/online storage repositories
• Need specific knowledge to retrieve and make sense of
OverviewOverview Future DirectionFuture DirectionCurrent Imp.Current Imp. EvaluationEvaluation
They are EVERYWHERE!!
Plus, employees are always expected to remember these activities in an Enterprise environment.
And Sadly, no matter how awesome one is,one can’t avoid forgetting those activities.
OverviewOverview Future DirectionFuture DirectionCurrent Imp.Current Imp. EvaluationEvaluation
Research Question
What is the efficient abstraction to What is the efficient abstraction to
recall and manage personal recall and manage personal
document activities?document activities?
OverviewOverview Future DirectionFuture DirectionCurrent Imp.Current Imp. EvaluationEvaluation
Taste PipelineOverviewOverview Current ImplementationCurrent Implementation Future DirectionFuture Direction
Computational Pre-AnalysesComputational Pre-Analyses Visual Analytic SystemVisual Analytic System UsersUsers
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
How to properly visualize the document activities?
• Pre-design survey results: (30 participants)
• The date and time when events happened (e.g., when documents were received, read, created, or modified);
• Content keywords that users associate with activities (e.g. the title of a document or the name of a person or company);
• Document types or applications that are used to perform a particular type of activity (e.g., Microsoft Excel or Apple Mail)
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
TASTE:
• Design Principles: Provide orthogonal information
Restrain the usage of screen space to minimum
Present information as direct as possible
Provide flexible user exploration paths
Visual Analytics system to manage Temporal Activities and Story TElling.
Cross platform design. JAVA+JOGL+Python+HTML
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Temporal View
• Enables interactive temporal explorations
• Present usage patterns and trends for individual activities
• Provide detail information and summarized information for particular time period
Reflecting “The date and time when events happened”
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Facets View
• Automatically extract facets in certain time span
• Automatically aggregate documents with related information
• Provide information filtering and sorting
• Detail information through mouse interactions.
Reflecting “Document types or applications”
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Entity Tag View
• Entities are automatically parsed from documents in particular time period
• Entities are presented with Tag Cloud metaphor to indicate individual frequency and importance.
Reflecting “Content keywords that users associate with activities”
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Detail View
• Focus on depicting single document from different perspectives:
• Temporal Information
• Related Documents
• Entity Information
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
But wait, here is more: Story Creation
Allows knowledge workers to construct stories based on their discoveries and findings from the visual interface
Provide supporting evidences for proposals or reports.
Group events based on user’s own memories
Share with others through communicating with another instance of TASTE
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Why should one use TASTE?
• Can TASTE be more effective than regular tools in helping users reconstruct document activities?
• What are the determining factors that make TASTE better or worse at helping users organize and manage their document activities?
Questions that we asked:
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
User Studies
• Two studies conducted : Laboratory (20 participants )and Field Study (three weeks, 12 participants)
• Two sessions per each study per person, counter balanced
• Three measures taken: Accuracy, Efficiency and Confidence
• Total of 21 tasks per participant
• TASTE v.s. Regular Toolsets that participants normally use
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Significant ResultsLab Study Field Study
Document retrieval with TASTE was more accurate. As suggested by results (F (1, 78) = 45.49, p < 0.0001), TASTE has an overall 17.3% accuracy gain (as computed by comparing mean values (90.5-77.13) /77.13) during the two sessions,
TASTE provided better retrieval accuracy. The ANOVA result (F(1, 82) = 8.80, p < 0.004) indicates significant differences between TASTE and the regular tools on the accuracy of retrieving information. Even though participants had prior knowledge of the data, TASTE still delivered nearly an 8% ((84.88–78.57)/78.57) increase in retrieved information
TASTE was more efficient. The ANOVA result finds differences between TASTE and the regular tools (F(1,78) = 10.84, p < 0.001). As illustrated in Figure 5(B), timed comparisons show a 31.1% reduction in time ((166-114.3)/ 166) with TASTE.
A significant efficiency improvement. A significant ANOVA result (F(1,82) = 6.13 p < 0.015) indicates that TASTE helped participants perform the tasks efficiently. As suggested in Figure 7(B), TASTE reduced by 36.8% ((176.3-111.3)/176.3) the time spent on searching for document information.
Users felt more confident with TASTE retrieval. We expected TASTE to increase user confidence in organizing document activities, and the results from ANOVA (F(1, 78) = 17.12, p < 0.0001) support this expectation.With TASTE, users felt more confident in the accuracy of their answers in the lab study
Significant confidence increase. Initially we were expecting the confidence value to be similar between TASTE and RT, since participants had prior knowledge about the location of their own data. However, ANOVA (F(1, 82) = 16.16, p < 0.0001) suggests a very significant difference between the confidence values participants gave for both tools. As shown in the Figure 6 (right), participants trusted TASTE more than the tools they used daily
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
In summary & in quotes
• Cohesive Visual Exploration• “[TASTE] is very good at giving a quick impression of data across the board. I like how
you can mix different types of files/people and represent them cohesively.”;
• Visual Facilitation of sense-making• “ By looking at all the views [in TASTE], I can easily relate all the information together
and quickly see what is going on from different perspectives.”
• Temporal Searches• “The frequency ranking is very useful and it’s not available in other tools like Spotlight
search. TASTE really provides you the flexibility in retrieving information in time frames and helps you reconstruct the amount of time spent on the various activities.”
OverviewOverview Current Imp.Current Imp. Future WorkFuture WorkEvaluationEvaluation
Future WorkOverviewOverview Current ImplementationCurrent Implementation Future DirectionFuture Direction
Thanks
• Thanks to Eric Bier, Bill Ribarsky, Bill Janssen, Kyle Dent, Stu Card and Wenwen Dou for their supports throughout the project
• Question please
• Xiaoyu (Derek) Wang: [email protected]
OverviewOverview Current ImplementationCurrent Implementation Future DirectionFuture Direction
Intern Poster
Xiaoyu Wang [email protected]
OverviewOverview Current ImplementationCurrent Implementation Future DirectionFuture Direction
Eric
Xiaoyu
Intern Poster Instruction
EricXiaoyu
Intern Poster Instruction
PARC
We are getting here…Future Work
Story Board Disassemble
Construct Knowledge Structure
OverviewOverview Current ImplementationCurrent Implementation Future DirectionFuture Direction
EricXiaoyu
Intern Poster
Instruction
PARC
It’s Intern Poster time again.
Here are the instructions
Xiaoyu Wang [email protected]
OverviewOverview Current ImplementationCurrent Implementation Future DirectionFuture Direction