Andrew:
With regard to your wife's comment, let me take the devil's advocate
role. Let's presume that the links provided by Amazon are indeed done
solely through its technology. In other words, the system observes that
when people order or search for books about abortion, a large number of
them also order or search for books about adoption at the same time or
from the same account. The system also observes that, of all the people
who order or search for books about adoption, a significantly fewer
number also order or search for books about abortion. For these both to
be true, then it would first be true that a significantly larger number
of people order or search for books about adoption than order or search
for books about abortion. Second, it would also have to be true that
the people who order or search for both are a larger subset of the total
who order or search for books on abortion, so much larger that there is
enough activity to create the one-way link. These propositions are
credible, so far as I know.
If, as your wife proposes, a librarian's collaborative filtering or
query suggesting system were to be devised, I would think it might still
be done somewhat as Amazon has done it. The Amazon system is
essentially taking a verifiable sample of users (i.e., all its users)
and noting their activity. A library circulation system used to create
similar information would likely come out with a similar group of
suggested links. On the other hand, if the system were pre-ordained,
such as LC Subject Headings, then the chosen links might be different.
Three observations: 1. If Amazon's statement is true, i.e., the
computer did it, then there is some interesting sociological information
to be found in all this. Perhaps people who are interested only in
adoptions read more than the people who are interested only in
abortions. Perhaps they have more time to read before faced with their
life-changing event. 2. The Amazon system of linking would seem to fit
well with the pattern of thinking described by Lakoff and Johnson and
found now in cognitive linguistics and cognitive science, i.e., that of
creating logical links through use of primary metaphor, creating primary
prototypes, radiating categories from them, and in the process, creating
and amending one's idealized cognitive model (or frame) along the way.
(As opposed to traditional Aristotelian logic, and its descendants, like
Boolean logic.) Of course, not everyone's individual radiating category
would be the same, but the Amazon system would seem to be a method of
determining generally how folks with in their group of users do think,
up to a certain level. I presume that Amazon does not do too much
cross-linking, e.g., suggesting different books on abortion based on the
user's previous selection of political or religious books. And, of
course, choosing books is not the same as actually reading them or
believing what they say. 3. I would prefer a computer-generated method
of creating links, even if it were to create some politically sensitive
link patterns, than to have the links decided by others, even seemingly
fair-minded librarians. We all know that Melvil Dewey meant well, but
his xenophobia shows in the Dewey system, and we are all capable of
subconsciously altering the routes to knowledge. I say this, even
though I am pro-choice.
Charles Dyer
Andrew Larrick wrote:
>Here's a reply - forwarded with permission (from my wife).
>
>To respond myself - my initial email may have muddled the fact that I'm not
>so much interested in what kinds of responsibilities these businesses have,
>or whether we should trust them or not, than with what these systems (and
>their limited, commercial, purpose) tell us, by contrast, about how systems
>employed by libraries (or, possibly, by true database/content vendors like
>Lexis or Westlaw) work, and also with what a *librarian's*
>collaborative-filtering and/or query-suggesting system might look like.
>
>For better or worse, and regardless of the fact that it is outside of the
>strict intentions of these commercial entities, our *patrons* do indeed use
>these systems/websites (Amazon, in particular) as a means of conducting
>research and these types of systems are shaping what is expected of us.
>
>...here come the 'out of office' emails...
>
>
>Two thoughts:
>
>1) Amazon has little or responsibility to be transparent unless its
>customers want that; its primary responsibility is to make money by
>meeting its customers' needs.
>2) The Wal-Mart/Amazon problem is not a filtering problem at all; it's
>a UI problem.
>
>
>
>>This points to a situation in which commercial entities (who often
>>
>>
>actually
>
>
>>strive to mask the very mechanism of how their systems "magically"
>>
>>
>identify
>
>
>>related searches) are in a bit of a quandary when the objective
>>
>>
>application
>
>
>>of software algorithms to offer up "related" searches provides results
>>
>>
>that
>
>
>>might offend some number of customers. There is a related anecdote about
>>
>>
>a
>
>
>>Wal-Mart collaborative-filtering system ("people who watched/enjoyed
>>
>>
>movie
>
>
>>X also watched/enjoyed movie Y") that ran into similar problems.
>>
>>
>
>The "social" user interface raises user expectations of appropriate
>behavior from a machine; by toning down the "gosh shucks, I'm just
>like a person" tone of the copy (from "did you mean" to "you might be
>interested in..." or "people who liked X also liked Y") Amazon or
>Wal-Mart can avoid the negative user experience associated with nuance
>falling flat.
>
>What's the quandary? If Amazon has any responsibilty to effect
>transparency by producing consistent results, that responsibility is
>trumped by their need to respond to customer desires, and customers
>(at least some loud ones) don't want to feel like their bookstore is
>second-guessing their reproductive choices. Amazon.com is, as far as
>they can see, being extremely rude.
>
>
>
>>It seems like these systems could create delicate PR situations for
>>management whether they work "right" or whether they work "wrong" from a
>>pure information-retrieval, relevance, perspective.
>>
>>
>For all practical purposes, nobody cares how they work, as long as
>they do work (where work = produce usable results without being
>insulting.)
>
>The problem is not the offering of other options, but the phrase "did
>you mean...?" There's a big difference between "did you mean
>adoption?" and "would you like to search for adoption" or "do you want
>books on adoption?" The first sounds snarky and reproving. From a user
>behavior standpoint, it's a legitimate question, but it's socially
>unacceptable.
>
>The issue is similar with Walmart; searching for Roots and getting a
>suggestion for Planet of the Apes isn't always going to be offensive,
>but a UI that implies a subject link between the two is.
>
>What say you?
>
>
>
>.
>
>
>
-- Charles R. Dyer 808 East Maple Street Bellingham, WA 98225 360-738-6439 fax 360-738-6439 (call first) mobile 619-559-1845 charlesrdyer@clearwire.net
This archive was generated by hypermail 2b29 : Wed Nov 14 2007 - 20:46:24 PST