[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Anti-Bayesian algorithms for Montgolfiering reputation capital



-----BEGIN PGP SIGNED MESSAGE-----

Reputation capital is becoming increasingly important in the online
community.  A key problem facing new entrants to cyberspace, as well as
the people who deal with them, is how much capital these new entrants
can be said to posess.

A typical solution proposed might be described as "Bayesian", whereby
new entrants to the community are assigned a quantity of reputation
capital that reflects the expectation value of their future reputation
capital.  This is unsatisfactory, because, like all Bayesian statistical
methods, it ultimately relies on the very fallible judgment of the
entities who make such assignments (or that of the implementors of the
automated agents which make such assignments).

It is clear that an effective measure of the initial reputation capital
of new entrants to the online community is greatly to be desired.  It is
felt by the author that anti-Bayesian methods, methods that make no *a
priori* assumptions about the reputations of new entrants, show great
promise as means for achieving this result.

The pioneering work of Parry et al. in the use of test messages to
inspire rapid responses from a community's new entrants.  This method,
which Parry and his colleagues name in their typically colloquial style
as "trolling for newbies", is described in detail in a review article
by Suter (1994).

In this article the author proposes the use of Parry test messages and the
responses generated thereby as a means of rapidly establishing initial
reputation capital for an online community's new entrants.  By
evaluating new entrants' responses to the test messages, messages which
can be carefully designed to virtually guarantee that a useful response
will be elicited. These responses can be reviewed by individuals or
tabulated and evaluated by software agents to estimate the new entrants'
most likely reputation capital without resorting to arbitrary assignment
of an initial value.  Established members of a community may use
existing filtering agents to eliminate Parry test messages and the
responses thereto from their own datastreams; or they may elect to allow
these messages through and review the Parry tests and resulting
responses themselves.

There are indications in the literature that other researchers are at
work on this technology.  For example, messages that resemble Parry
test messages have appeared recently which contain coding very similar
to work of May's (1993, 1994, 1995, 1996) that have generated responses
that are ideally suited for reputation capital establishment from new
community members.

This fertile field for study holds significant promise, and the author
welcomes further developments from any who chooses to work here.

- -- 
Alan Bostick               | "Dole is so unpopular, he couldn't sell beer on
mailto:[email protected] | a troop ship." (Ohio Republican Senator William
news:alt.grelb             | Saxbe on Bob Dole's early career in the Senate)
http://www.alumni.caltech.edu/~abostick
http://www.theangle.com/  The first site with a brain.  Yours.

-----BEGIN PGP SIGNATURE-----
Version: 2.6.2

iQB1AwUBMneUvuVevBgtmhnpAQEE4wL+PnPHAIjYv77ad3xDYUM/WMFMqGubduyH
vs5veTK0BlqdFSChvponVpXGmP9fHXBnTVxlipj8To12DfFgFzYy1gqkQ0NgUrud
UHygCOU503ZIx2u5FpifLY95VR0u5HqG
=EU+x
-----END PGP SIGNATURE-----