Transcript The Case for Browser Provenance

The Case for Browser Provenance
Daniel W. Margo and Margo Seltzer
Harvard School of
Engineering and Applied
Sciences
Overview
• Problem: Browser Data Management
• Solution: Provenance for Web Browsers
• Use Cases
• Details and Challenges
• Implementation
The Modern Browser:
A Super-Application
• Originally a distributed document reader.
• But now most documents are distributed.
• And the definition of “document” has changed:
– Webmail
– YouTube
– Google Apps
• It is difficult for users to manage all this data.
– e.g., recall a specific web page.
Browser Data Management (I)
• A “little big data” problem…
– My history: ~25k objects in ~2 months.
– Tractable for computers, but not for users.
• Traditional solution: Bookmarks.
– Requires users to tag their data in advance…
– …and to manage the bookmarks.
• Advanced solutions:
– History Search (Google Chrome’s “New Tab” page)
– Autocompletion (form history, saved passwords)
Browser Data Management (II)
• Firefox 3’s “Smart Location Bar”
from http://support.mozilla.com/en-US/kb/Smart+Location+Bar
• Most solutions powered by history and usage statistics.
• “History and usage statistics” = provenance.
Traditional Browser History
Web Graphs (Firefox 3 Places)
Browser Provenance
Browser Provenance
Use Case:
Contextual History Search
• Most history search is textual
• Edges imply contextual relationships.
– E.g. “rosebud”  “Citizen Kane”.
• 2-phase contextual search (Shah et. al):
– Perform a textual history search.
– Then, push the weight of results to neighbors.
• Similar to modern web search…
– And good for the same reasons.
Use Case:
Personalizing Web Search
• Context is created by the user.
– So a gardener relates “rosebud”  “flower”.
– Frustrating if Google returns “Citizen Kane”.
• Browser could clarify context to search engine!
– Naïve: Just insert “flower” into “rosebud” searches.
– If engine had a better interface, we could do better.
• Personalization with privacy.
– Browser knows more about user than cookies can.
– No need to give third parties raw personal data.
Use Case:
Time-Contextual History Search
• Current histories can’t recreate prior state.
– e.g., “were these two pages open simultaneously?”
• Time relationships…
– Are natural: “rosebud, and I think I was also looking at
gardening tools around that time.”
– Narrow the search space a great deal.
• Related Work:
– Gyllstrom and Soules’ “SeeTrieve”
– Dumals et. al’s “Stuff I’ve Seen”
Use Case:
Download Lineage
• Need to know where data comes from.
– For source attribution, finding updates, etc.
• URL is not always sufficient.
– “This image came from…ImageShack!”
• This is exactly what provenance is for!
– Just query ancestors!
Conclusion
• Browsers record many statistics.
• These statistics are provenance records.
• Provenance techniques can improve:
– History search, via context.
– Web search, via personalization.
– Data management, via lineage.
• Some details in the paper.
• Excruciating details in future work.