Subject:
|
Re: Opinions wanted: article rating harmful? (was: New feature: Article rating)
|
Newsgroups:
|
lugnet.admin.general
|
Date:
|
Wed, 26 Apr 2000 17:58:57 GMT
|
Viewed:
|
2130 times
|
| |
|
|
In lugnet.admin.general, Matthew Miller writes:
> Todd Lehman <lehman@javanet.com> wrote:
> > That's what the averaging effect is for -- to smooth that out. If the
> > system also could learn what you liked, you might find that helpful.
> > (That's a long way down the road, though.)
>
> Smoothing what out, though? How does the system distinguish between "0: I
> like posts about robots, but not in .castle" and "0: not interesting to me",
> or "60: kinda funny if you're in the right mood" and "60: contains some
> useful information but could be more complete"?
It can't (and doesn't actually need to) distinguish that so greatly -- the
bottom line (to it) would be that you disfavor posts about robots in castle
and things that are kinda funny or contain some useful info.
> Also, I'm _very_ skeptical of the "match what you like" concept. It sounds
> neat in practice, but I've never seen it implemented well.
> homr/ringo/Firefly/bignote/launch/whateverthey'recallingthemselvestoday did
> an ok job, but you'd have to do some serious language parsing/comprehension
> stuff to make it work with news posts, even in such a narrow subject as
> Lego.
I'm very skeptical about that form of collaborative filtering as well. But
there's a completely other form of it which is purely statistical correlation
based. It looks only at how your responses correlate to the responses of
others, without knowing (or having to know) anything at all about the content.
Then it makes a prediction about how you would feel about some brand new data
point based on how other people before you felt about that new data point.
Of course, it's a scheme which works better for things like record albums or
books or LEGO sets than time-sensitive things like news articles. If you're
always the first one to rate something, it couldn't help you out, but if
you're always the last one to rate something, then it could. That's the
theory, anyway. It works basically on the premise that everyone tends to
have opinions which can be approximated by a linear combination of other some
set of other people having multiple partially overlapping domains of input.
--Todd
|
|
Message has 1 Reply:
Message is in Reply To:
309 Messages in This Thread:
(Inline display suppressed due to large size. Click Dots below to view.)
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
