The Use of Provenance in Information Retrieval

1
The Use of Provenance in Information Retrieval

• Simone Stumpf
• Erin Fitzhenry
• Tom Dietterich

2
Defining Provenance

• To us, provenance concerns
• The origin of content within documents
• The relationships between documents

3
Why focus on Provenance for Information Retrieval?

• People remember the relationships between
  documents!
• Episodic vs. Semantic Memory
• Studies
• Blanc-Brude Scapin (2007)
• Gonçalves Jorge (2004)
• No need to formulate keyword queries
• Other common document attributes are often
  inaccurately remembered (Blanc-Brude Scapin
  2007)
• Title (20 false recall)
• Size (53.8 false recall)
• Time (47.6 false recall)

4
Example Use Case Where did I save that again?
I got an email from Tom
I saved the attachment
And I pasted some information from the attachment
into a PowerPoint document
Where did that presentation go??
5
Requirements for Tracking and Visualizing
Provenance

• Instrument all important document provenance
  events
• Provenance events are NOT automatically captured
  by Windows
• Develop a UI enabling users to locate documents
  via the provenance relationships they remember
• Integrate the UI into the Windows Desktop

6
Capturing Provenance Events with TaskTracer

• TaskTracer is a Personal Information Management
  system
• User defines a hierarchy of Projects or
  Activities
• As the user works, TaskTracer automatically tags
  (according to task/project)
• Files
• Folders
• Email Messages
• Email Contacts
• Web pages

7
Instrumenting TT to Capture Provenance Events

• TaskTracer already instruments many desktop
  events
• Open, Save, SaveAs, Close
• EmailArrived, Email Open, Email Close
• Open URL, Close URL, Follow Hyperlink
• Idea Extend existing instrumentation to cover
  key provenance events
• CopyPaste, SaveAs, FileCopy/Rename
• AttachmentAdd, AttachmentOpen, AttachmentSave,
  EmailForward, EmailReply
• FileDownload, FileUpload
• Coming soon

8
Instrumenting TaskTracer to capture Provenance
Events (cont.)

• Database of document-to-document provenance
  relationships

9
Developing a User InterfaceTaskTrail

• A tool for visualizing provenance

Double-click to open
Users Query
Mouse over details
Click to Expand
10
Integrating TaskTrail into the Windows UI

• Launch a query by right clicking on an item
  within
• Windows Explorer, Outlook, TaskExplorer

11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
Research Questions

• Does TaskTrail help users find documents more
  quickly than other methods?
• How should the provenance graph be laid out?
• What kind of provenance events do users
  accurately recall?
• How large are the provenance graphs?
• What patterns exist (if any) in terms of the
  succession of provenance events?

15
User Studies Formative

• Observational Study (planned)
• What provenance-related actions do users perform?
  Which of those do they remember?
• Observe 12 participants in their workplaces
• Record provenance-related actions performed
• Interview participants after 1 week to see what
  they remember
• Free Recall
• Cued Recall
• How do users layout their documents according to
  what they remember?

16
User Studies Summative

• TaskTrail Study at Intel (in progress)
• 4 participants (so far) are using TaskTracer for
  at least 1 month each
• Then they will use TaskTrail to locate their own
  documents
• Measures of success
• Do users locate more documents using TaskTrail?
• Do users locate documents more quickly using
  TaskTrail?
• Do users prefer using TaskTrail?

17
Provenance-related User Studies are Hard!

• Must be done in the wild
• Involves
• Long time-scales, which increase chances that
• Participants will drop out
• Situation on site will change
• Potentially sensitive information
• Emails to/from users not participating in the
  study
• Documents regarding trade secrets
• Installation of some event-tracking software
• Software installation/maintenance can introduce
  compatibility, scheduling and other problems

18
Summary TaskTrail

• TaskTrail
• Instruments desktop provenance relationships
• Allows user to query by right-clicking objects
• User can browse visualization of provenance
  relationships to find desired documents
• Exploits human episodic memory to help users find
  documents