Ed Chi: Augmented social cognition: how social computing is changing elearning

chi@acm.org

Finding other people interested in pancreatic cancer on del.icio.us, and connecting with them. How much learning is shared, social, collaborative, synchronous or asynchronous.

• invisible social signals from the crowd
• using information theory to model social tagging

concepts → users

topics → documents

decoding ← tags ← encoding (memory - human fallacy)

controlled vocabularies vs folksonomies

there is noise - can use information theory from Claude Shannon (see Gleick, 2011Gleick, J. (2011). The Information: A History, a Theory, a Flood. Pantheon.)

entropy - measure of uncertainty, tied to gambling, ability to guess what will happen

as time goes on, entropy is increasing - harder and harder to gamble what the next word will be in the system → the knowledge encoded in the system gets harder and harder to retrieve

rise in average tags per bookmark, more and more specific (google math and it gives you millions of sites) - fighting back against entropy

TagSearch: spreading activation in a bi-graph, computation over very large data set

use semantic analysis to reduce noise. look at user behaviour in data.

system understands tutorial and tuRorial as the same, common misspelling (T is next to R).

Mr Taggy (not live anymore). type in “interesting science”. then you can choose “by interesting - I mean “cool”, drill down (I want to learn about physics).

2 interface x 3 task domain design

• exploratory vs baseline
• future architecture, global warming, web mashups

30 subjects

• intermediate or advanced computer search
• summarization
• keyword generation

Exploratory users: (Kammerer et al, CHI 2009)

• more queries
• took more time
• better summaries
• more relevant keywords
• higher cognitive load

Suggestive of deeper engagement and better learning

Some evidence of scaffolding for novices in the keyword generation and summarization tasks - works better for novices than experts, brings “equality” of information

(Hong et al AVI08; Nelson et al HCII 2009)

trying to find a restaurant in Beijing, looking at search history

heuristics for finding good food, classic AI problem. cooperative/social search, offer “hints”.

SparTag.us - when reading webpages, you want to leave behind traces of the places you find the most interesting. Aggregate into a shared notebook, shared with friends. Tag cloud.

Recall vs first-visit, recall. Highlighting as importance indicator.

Sensemaking task

• find and read material about “Enterprise 2.0 mashups”, in order to write two essays
• seeds: “expert” content for scaffolding
  • tags from delicious
  • URLs from Google/PageRank
  • constructed and then shared through social mechanisms
• performance measures
  • learning gain: pre/post knowledge test
• conditions
  • 1: no access
  • 2: access only for yourself
  • 3: access with simulated expert notebook (choose to use or ignore)
• results
  • 3 highest, 2 lowest
  • why did users do worse in solo situation? costs time.
  • they should have done recall tests after a few weeks - that might be where note taking is meaningful
  • all users started out with Wikipedia, once they had a “schema”, understanding of domain, they moved away from this “crutch”
  • shared notebook + Wikipedia was very powerful

Van Rostorff Isolation Effect

• anything you highlight, people tend to focus on learn
• Nist and Hogrebe 1987
• if I gave you a bad shared notebook, it would really skew your learning

(Kittur et al CHI 20087; Suh et al Wikisym 2009)

exponential growth in Wikipedia - common from library studies?

but after March 2007, fall of the exponential train

preferential attachment: edits begets edits

instead

ecological population growth model (logistic growth model)

• depends on environmental conditions
• K = carrying capacity (resource limitation - new topics)
• fits the data

Analogy with Darwin:

• limited opportunities to make novel contributions
• increased patterns of conflict and dominance

WikiDashboard.com

A challenge: a modified logistic model - carrying capacity as a function of time, from exponential to linear (paralleling growth in knowledge in the real world)

• ulitization of social signals for learning and information access
• establishment of common ground
  • implicit coordination
  • explicit coordination
  • negotiation
• “all collective actions are built on common ground and its accumulation” (Clark and Brennan 1991)

Reconciliates two parts of CSCL

• communities, informal learning
• schools, formal

BUT

in both two experiments, they learned significantly. But why did they learn?

• have to write summary - challenging and difficult, much more difficult than tagging or reading annotations
• in second, have to write essay
• (please that increased learning with increased cognitive load)
• information processing through writing - might explain more of the learning than the tagging activity