Peter Pirolli Keynote: Beyond Information Foraging to Ecologies of Sense Making by Peter Pirolli, Palo Alto Research Center
Information patch foraging: How do you model when someone says “I’ve looked for useful info here enough. I should move on.”
He argue that if you increase the probability (say from 0.15 to 0.015) of choosing the wrong page as you browse a graph, the number of pages you visit goes up, diverging drastically as depth increases.
But what is the typical depth of navigation? The number of pages goes crazy at depth 12, but do people really go that deep? My guess is that people usually go for a depth of 1-4.
He then shows a model predictions and actual observed browsing on Yahoo on ParcNet and (eyeballing) it looks like 90% of sessions had a browse depth of 10 or less, 75% depth of 6 or less.
Stu Card and Pirolli had a ToCHI paper looking at eye movements and what they can tell us about information scent. They found that for crappy info scent in a hyperbolic browser, eye went all over the place, but for good info scent, eye took fairly direct path along graph.
Then they tried to model the eye movements and had some luck when they treat the eye as a rational, economic decision-maker. They came up with a scary looking equations expressing the probability of visiting any node in terms for euclidean distance, number of items in the visual group, scent (category), and inhibition of return.
Still early and their model predicts time in tasks and clicks decently well for high scent tasks but still work to do with low scent tasks. Hypothesis: People have trouble finding home again when they get lost.
Microeconomic model of highly interactive visual interface.
Model can do browsing and information seeking tasks for ~7000 nodes in human-like time.
And he has a book coming out
Other comments he makes: There are real evolutionary reasons why we enjoy and are attracted to certain kinds of scenes. We’re not exploiting any of this in the interfaces we make — no regard for aesthetic.
Adaptive info interaction for intelligence project
External Data <--> Shoebox <--> Evidence File <--> Schemas <--> Hypotheses <--> Presentation
Ext. Data –> Shoebox by searching and filtering
Shoebox –> Evidence file by readinga nd extracting
Evidence file –> Schemas by schmeatizing
Schemas –> hypothesis by building a case
Hypothesis –> presentation by building a representation
At each step, you can go backwards by searching for more data
Socially Mediated Foraging and Sensemaking is becoming increasingly important from high profile sensemaking failures such as 9/11, Columbia shuttle. But remember that we know a lot about the potential and failures of group decision making, so stop gushing about collaboration.
Mathematics comes from Hogg, Huberman & Clearwater, optimal foraging theory that deals with collaboration in groups
Factors that make social media system work are:
- quasi-independent search and knowledge contribution
- decentralization in that people are looking at different things
- Interference effects (transaction costs)
- methods for sharing and aggregating information