notes taken during the round table:
in a recent article in science, it has been shown that humans learn by example rather than by argument. consequently, your health/happiness is influenced by your friends as you tend to ”imitate” (learn from) them. sandy pentland co-athoured a paper (soon to be published) that tries to explain social outcomes based on this
idea of learning by example.
see also his article on the american scientist, which elaborates on the concept of people’s beliefs and actions being often “dominated by attitudes and actions of peers, rather than logic or argument. “
arthur spirling discussed difficulties/joys of doing text analysis in the area social science. in this area, duncan watts has been running interesting experiments (e.g., the music lab ) on the web. the web is a very good virtual
lab for two reasons: 1) researchers can do random assignments and 2) they can control the (experimental) environment. his group is planning a “public good” experiment. the experiment consists of a large number of networked individuals (MechTurk users) who interact synchronously, and the goal is to study how performance is affected by different payoffs and network topologies (how people are interconnected). norish contractor is working on a variety of hot topics: 1) how to study multi-dimensional networks (networks whose nodes are of different nature – they are docs, people, organisations). 2) the relationship
between different types of nets (e.g. for the same multi-player gamers, they study their net of partnership, istant messaging net, trade net, and email net). 3) how to make casual inferences (they apparently used exponential random graphs – don’t ask me the details of this approach ;-)) 4) how to connect people that might like each other. the great social scientist michael macy showed a tool built by two of his students (one of whom is scott golder) that studies temporal trends in twitter. they initially started with 2B tweets produced last fall & winter and, after filtering the tweets outside the US, they ended up with 540 M tweets. this tool is able to visualize the temporal occurrences of certain expressions (it uses hour-by-weekday heat map to show the temporal trends of those expressions.), which include the occurrences of:
- ‘breakfast’ and then ‘dinner’,
- ‘skipped breakfast’ and then ‘headache’ (it takes usually 6h for those two to show up)
- ‘peace’ (it’s popular on sundays)
- ‘happy hour’ (it’s popular not only on fridays)
- ‘love you’ (popular during weekends)
- ‘pregnant’ (on thursdays!!!)
- ‘exhausted’ (usually at night)
they now plan to use markov chains to model how one expression follows another.
tuesday will follow