From: tcmay@got.net (Timothy C. May)
Date: Sun, 3 Sep 95 22:58:32 PDT
To: cypherpunks@toad.com
Subject: Reputations and Reading Preferences
Message-ID: <ac6fdc45050210043624@[205.199.118.202]>
MIME-Version: 1.0
Content-Type: text/plain



(This post has a second part, "Using Web Sites to Get Distributed Ratings,"
which you might want to read even if the first part looks uninteresting.}


Hal has some interesting comments, as usual, about how positive feedback
about authors can be integrated into mail readers and news readers. (Or at
least I took Hal's post as a jumping off point to think about these
issues--again.)

I've given this some thought recently, but have reached no firm
conclusions, except: it's probably not needed by most of us.

That is, I have little interest in what others think. If the herd votes
that "America's Funniest Home Videos" is their favorite program, or that
Jackie Collins is their favorite author, who cares? At least I don't. I
don't read stuff on the basis of a "beauty contest." I suspect the same is
true of many others.

But Hal was perhaps thinking in terms of the more sophisticated approach of
creating scoring systems in which a matrix R[u,v] is created, where each
element corresponds to the rating of u given by v. On the Cypherpunks list,
for example, with more than 500 subscribers, each of the 500 might have a
rating of perhaps 100 of the active posters, maybe only the 20 or 30 most
active and memorable posters. (That is, many of you 500 readers would not
have a meaningful entry, just because you have posted so little.)

And more sophisticated models take into account one's reputation given to
the views expressed by others. That is, some of those R[u,v] are
discounted, some are held in high esteem. This results in a further
weighting of the reputations. As an example, Madonna gives a good rating
(0.8) to Da Dogg Pound, but I dislike Madonna and her taste, so I weight
her rating accordingly.

And one can imagine scoring systems in which cumulative ratings are
considered. E.g., a lot of people whom I respect like the works of Thomas
Pynchon, so I'll give him a try.

(I use a simple scoring system in YA-Newswatcher, though it needs more
work, IMO.)

Things rapidly spin out of control, in terms of the communication needed,
the difficulty of getting participation (like a lot of polls, the most
thoughtful often are the least likely to respond), and so on. A few years
ago on the Extropians list this was a hot topic, and various "rating
systems' were proposed. I recall a particularly complex scheme by Alexander
Chislenko which purported to solve this problem...unfortunately, it still
looked to me to have the essential characteristics of a beauty contest.

Fortunately, I can't see the need. Why not? And what's the better alternative?

Here's a recipe which seems to work pretty well, and not just for me:

* Apply filters based on one's own likes and dislikes. Killfile some
posters, skip some threads, etc.

* Periodically get inputs from others that one respects, a la book
recommendations, movie reviews, etc.

This effectively generates what the mass rating scheme would generate,
albeit not identically (how could it?). One's favorite posters get read,
and one's favorite posters can make some recommendations of other authors
one should look at. Informal, but with many emergent advantages.


Using Web Sites to Get Distributed Ratings

Concretely, one way to do this is for people to do what Eric Blossom is
doing with his "Cypherpunks Lite" mailing list. Or with the Web archives
that Todd Masco has been running, and that Jay Campbell just started.

Imagine: Rudy the Rater sets up a mail-to-Web-site archive, using whatever
these folks above are using. Except, he screens the stuff to separate the
bad from the good, using whatever criterion he thinks is right. People
either use his site, or don't. (A Web site has some logistical advantages
over subscribing to various mailing lists which do the same filtering.)

Other services emerge with differing rating criteria, different aesthetic
standards of goodness, or even different interests. (Some sites could
filter out all of the political posts, others could filter out all the
programming posts, etc.)

Readers could pick the sites they wish to read, and simultaneously be
exposed to authors they might have otherwise overlooked. The authors they
like, and the filtering of the site operator, create a "clique" (as in math
terminology) that effectively is similar to the vector weighting scheme
present in the R[u,v] scheme.

An advantage of the distributed site model over the R[u,v] model is the
lack of any central coordination, the market anarchy of the process.

Shared kill files are another possibility. As kill files become more
commonly used, they may be traded around, weighted in the same way as
described above.

Again, I claim that an adequate and workable solution is not to solve the
more general problem of everybody rating everyone else, but just of having
access to a limited number of killfiles from people one respects.

These approaches are doable today. Especially the distributed Web competing
"best of" sites.

I may do one myself! (If Jay Campbell lets me have Web stuff on his system.)

--Tim May

---------:---------:---------:---------:---------:---------:---------:----
Timothy C. May              | Crypto Anarchy: encryption, digital money,
tcmay@got.net  408-728-0152 | anonymous networks, digital pseudonyms, zero
Corralitos, CA              | knowledge, reputations, information markets,
Higher Power: 2^756839      | black markets, collapse of governments.
"National borders are just speed bumps on the information superhighway."