mea18 from sfu.ca
Enron dataset. y axis = email addresses. x axis = time. Size of circles = size of email? Or perhaps number of emails?
Crowded areas in email set for six Enron people. Very clear high-traffic, low-traffic bands...
They wrote a tool to generate the graph from teh data (and assign/adjust parameters for view)
Centrality of person when it ocmes to emails sent, received...
Histogram of directly sent emails- peaks for one guy in Oct 01. Interesting!
Identify temportal gaps (events can be corellated to these). Can identify (from several addresses of one person) when they switched, etc.
One guy, four addresses. Can see switch patterns. Analyze size of sent emails.
Q: Advantages of visual = ?
A: See patterns in organization- easier this way.
(I think it's easier to see several data sets at once.)
Q: In the example w/ 4 email addresses for one person, was that autodetected? Or manual?
A: They're all parts of his name. Done manually.
Talk 2:
#hci
Judith Klavans and Jennifer Golbeck.
How CS can help museums with their collections.
This research is funded by the Inst. for museum and library science.
18 museums are users and test subjects.
Text written about images. Social tags form crowdsourcing.
User text is... misspelled, inaccurate, which language is it in...
"I call that an adjective noun, but parsers really struggle with this stuff."
Compare multi-word tags to Wikipedia titles to try to identify them.
Overlap of tags and terms (from professional to crowdsource) is 9.4% ...very low!!!!
"People use tags differently based on image type." - religious, ceramics, abstracts...
User looking for "red" is more likely looking for something abstract.
Research shows NO correlation between how often a tag is applied, and how high the image should rank in search results. Therefore, this implies that there should be other factors included.
Only 1 person in the room had been here at the first GHC! Yay growth!
steve.museum.org
Q: Maybe user training would give you better results? i.e. is your research disappointing?
A: Most ppl writing text about images are professionals. We're doing really well.
Q: I worked on this too, years ago! Our problem was that many objects were insufficiently described + messy description text (style and/or handwriting). Did that happen to you?
A: Yes. Database at typical museum does not allow additional data. Museums have different policies about data collection. (Why would a museum not allow information to be recorded?)
Q/A: Yes, interdisciplinary team.
"This is the profession of the reference librarians!"
"Semantic ontology, whatever you want to call it- we know it when we see it!"
Talk 3:
Icon design, consisting of rings with a color for price, distance, reputation, price(including shipping). size of ring correlates to higher quality. line new/used(black) at 0, 90, etc degrees. Map icons for competing items onto x-y. x axis = item's age (time since it's been posted) and y is relevance to the query.
Dropdown lists- sorting parameters, + box for showing details when mouseover.
Ran user tests: trained users, got them to evaluate the best item from data set according to given scenarios. There's a chart of right and wrong answers by how many- text, and icon. Some types of task did worse.
Which mode did you prefer? Mostly icon. Icon is good for quick overview.
Icon weaknesses: overlap = frustrating. color-meaning relationship wasn't intuitive.
Reorder rings? Make outer bits most important.
Q: Why rings? Confusing.
A: Prevent overlaps from obscuring all data.
Q: I think it'd be more intuitive to make them partial rings instead of changing thicknesses. What's the new idea that you mentioned briefly?
A: (Misconmmunication occurs)
Q: How did you decide which attributes of products to display?
A: Got 'em from eBay's sorting parameters.
(Black line indicates bidding rate of item.)
Q: Maybe less parameters?
A: We got the parameters from eBay. Smaller would be easier to understand visually, yes.
Comment: maybe different colors?
Chrome 6.0.472.53
Ubuntu Linux 10.04
