120 likes | 216 Views
Observed Behavior and Perceived Value of Authors in Usenet Newsgroups: Bridging the Gap. Andrew T. Fiore Scott L. Tiernan Marc A. Smith. Proceedings of CHI ’02, Minneapolis, Minn. People. Andrew T. Fiore was a student at HCI Lab, Cornell University Worked at M$ Research
E N D
Observed Behavior and Perceived Value of Authors in Usenet Newsgroups: Bridging the Gap Andrew T. Fiore Scott L. Tiernan Marc A. Smith Proceedings of CHI ’02, Minneapolis, Minn.
People • Andrew T. Fiore • was a student at HCI Lab, Cornell University • Worked at M$ Research • Working on NOMAD project at Cornell (wireless computing in classrooms) • Scott Lee Tiernan • was student at Psychology Dept., U. Washington, B.S. Psych. U.Wash, B.A. Economics Claremont McKenna College • Microsoft Fellow when published • Marc A. Smith • Sociologist at Micro$oft Research, works on Netscan • Research & Design of social cyberspaces • UCLA Ph.D. (sociology), M.Phil. Cambridge, B.S. Drexel (Japan studies)
Usenet • Hopefully we’ve all used it before. • Less popular than it once was. • Extremely noisy (in s/n sense) and high traffic (10,000+ messages/month in some groups) • Article explores potential mechanism to separate wheat from chaff.
Netscan Project @ M$ Research • http://netscan.research.microsoft.com • Collecting Usenet headers since 1996. • This project analyzed data from Jan. 2000 – Jul. 31, 2001. • Data-mining this data allows measures of: • which group you post to • days you posted • total messages • number that were replies • number of replies you received • etc.
Existing solutions rely upon active voting • ebay or Slashdot • critical mass of evaluators required • quality of evaluation constant concern • abuse and noise remain nevertheless (Hot grits, Natalie Portman, Aeron chairs)
Hypothesis • Using data available via Netscan project, it should be possible to derive behavioral metrics of posters. • Need to ensure that results correlate to actual perception of value from subjective evaluators. • Study performed to measure this.
Methodology • 22 evaluators (20 Male, 2 Female) • Self-described expert/frequent Usenet readers • Half read from favorite newsgroup (see Table 1), • Second half read from the a single newsgroup • Surveyed afterwards. • Described in more detail pages 2 & 3 of article. • Questions such as: (See Table 2) • “I would read a message by this person in the future” • “This person behaves rudely or disruptively” • “I might like this person as a friend”
Participant Results • Most frequently read for technical support seeking, secondly discussing news and current events, thirdly social support, looking for music/images. • Tech support seekers more likely to buy or sell via news and spend more time reading. • Social support seekers more likely to seek out entertainment. • News event readers somewhat related to political discussions, more likely to pay attention to date.
Consistency Checking • Across evaluations of same-newsgroup readers, 10 out of 12 measures strongly correlated. • Authors you want to read again correlated to authors you don’t want to avoid in the future.
General Results • Many results in article itself omitted for brevity. • More active authors participate in larger, more active conversations. • Authors who joined in more threads, tended to interact more. • More groups posted to less likely to read more likely to be deemed rude (spam/jerks) • Dominating conversations (messages/thread) bad.
Behavior Metrics x Subjective Evaluations • Tenuous linkages overall. • Need to look at authors whom evaluators were “familiar” with (6 out of 7-point scale) to get correlations. • For lower familiarity results uncorrelated. • Familiarity linked to number of posts, although temporal component (used only January – July of 2001).
Why I think this is ominous… • I have friends who are flamers/trolls. • I can imagine a system when everything we post can characterize us in future transactions. • Through mechanisms of this sort we can invisibly transform data to suit perceived style of writing/reading. • Everything you say can be used against you in a court of public opinion. • Courts are ruling against anonymizers. • Who do you trust?