Zeitgeist: Popular now includes "Hot this month"

I've added "Hot this month" from the home page to the Zeitgeist: Popularity area:

http://www.librarything.com/zeitgeist/popularity

For clarity, hot this month is a measure of recent popularity versus overall popularity. So, while
"Popular by month" shows Harry Potter and the Philosopher's Stone at number one—it is almost ALWAYS the most popular book entered in a month—"Hot this month" shows books that shoot up. In essence, it is a list of new books or, occasionally, books that get a special boost (eg., The Help).

Finally, a list without a single Harry Potter title!

Thanks. I'm glad you made the distinction in terminology between hot and popular.

Can we get the same monthly and yearly hot lists that we have for popularity?

I'd also still like to see popularity by publication year.

I'm really liking all these new lists :)

Oh, and I'd still like to see more information given with each book: number of members, average rating, position in series, and maybe number of reviews.

I'd like the choice to jump directly to the work page. Often I see a book on the list, wonder what it's about, and want to read reviews for it without having to transition through the popularity page.

Could we have an option to show more than 10 on the homepage module?

It would also be nice if the heading of the homepage module linked to the extended list.

Stars: Stars have the usual problem--values have no diversity because people like the books they buy. Here's one screen's worth of popular books (400 books). The low numbers are generally classics people are forced to read--so the net information here is almost zero.

3 => 2
3.5 => 33 (8%)
4 => 273 (68%)
4.5 => 92 (23%)

The numbers for hots are a little different, so I've added it there.

3 => 3
3.5 => 19
4 => 65
4.5 => 13

Number of members: The problem there is that it would be unclear if that was X members during the period or otherwise. I've added it for "hots."

Average rating: In hotness list?

Reviews: Added.

#7 by timspalding> Stars: Stars have the usual problem--values have no diversity because people like the books they buy.

That's why you should let people rate books even if they're not in their catalog. In fact, you should force them to.

;)

7>

Stars have the usual problem--values have no diversity because people like the books they buy.

That's a problem only if you consider "This was pretty good" and "This was utterly amazing" to be endorsements of equal strength.

There's certainly debate in the collaborative filtering community about whether ratings are useful in general or in specific cases, but a statement like this doesn't fill me with confidence that you've actually considered the problem and run any test cases with your data so you can speak from authority about whether they're useful here.

>8 brightcopy:

I'm interesting in what you think this would add. Yes, a tiny tiny percentage of people would use it to rate books they've read and then thrown out. But it would also draw a lot of rating that people didn't stand behind—rating Ann Coulter or Obama down because you think they're ruining the country, not to mention commercial spam. You'd have to separate the two pools of ratings, one of which would be much less trustworthy. And it would be confusing. I see a net loss, here.

That's a problem only if you consider "This was pretty good" and "This was utterly amazing" to be endorsements of equal strength.

That's what 4 versus 4.5 means? I wasn't aware of that. It seems to me they are just as likely to mean "That was great" in two different rating contexts.

Anyway, my point was display, not algorithmic use. It isn't very interesting to list 400 ratings, 91% of which are 4 or 4.5.

That's a problem only if you consider "This was pretty good" and "This was utterly amazing" to be endorsements of equal strength.

That's what 4 versus 4.5 means? I wasn't aware of that. It seems to me they are just as likely to mean "That was great" in two different rating contexts.

No, that's what 3 vs. 5 means, and both of those can be encompassed under "like".

Are 90% of ratings really 4 or 4.5? System-wide, I mean, not on one particular Hot-This-Month book?

Can I clarify for my pea brain - this doesn't mean that these were read, just that they were added, right?

Is there anyway to get a list with those showing a read or someway of knowing whether they were read?

#10 by timspalding> I'm interesting in what you think this would add. Yes, a tiny tiny percentage of people would use it to rate books they've read and then thrown out. But it would also draw a lot of rating that people didn't stand behind—rating Ann Coulter or Obama down because you think they're ruining the country, not to mention commercial spam. You'd have to separate the two pools of ratings, one of which would be much less trustworthy. And it would be confusing. I see a net loss, here.

I think we just have two different philosophies on this point. I feel the good outweighs the crap. The existence (or possible future existence) of crap does not disprove that. Just because the bathwater is gross doesn't mean there's no baby in there.

On your particular example, I have no doubt this would be true. However, I don't think it would be the same case for Good Omens. And I think you should trust that I'd have enough common sense to look at a breakdown of ratings. In the case you cited, I'd notice that the ratings were very polarized. That there's what we call information.

I think part of the fundamental problem is that maybe you think we think rating is supposed to tell the user how "good" or "bad" a book is. I don't think of it that way. It tells me how people are rating it. Those are two very different things, and I have enough intelligence to sort them out.

I don't think it's important to let us add data for books that we don't have in our catalogues; what I'd like to see is better treatment of the books in various collections ("include in statistics" checkbox, etc.), such that people won't hesitate to add and rate books that they no longer own, or whatever.

Thanks for adding the overall member count and average rating; that's exactly what I wanted.

Thanks for making the link on the title in the Hot This Month list go to the main work page, but I think it's confusing that the title link in the Popular By lists still goes to the popularity page.

Are 90% of ratings really 4 or 4.5? System-wide, I mean, not on one particular Hot-This-Month book?

No. The range is wider. This isn't a general argument against ratings, but against a specific proposal—to put the on "Popular books." The range is, as I said, slightly larger for "Hot this month" so I have included it. But it doesn't make much sense on "Popular books," when the data is so very repetitive.

>16 timspalding: You keep missing the point that 3.8 is very different from 4.2.

Re. #17: "3.8 is very different from 4.2"

Even if extrapolated to a 100-point scale, that would be a 76 vs an 84, which isn't "very different".

To illustrate:

.........,.........,.........,.........,.........,.........,.........,.....|...,...|.....,.........,   —   pretty much "the same place" on the scale.

Of course, I'm in the "star ratings are useless" camp.

I believe I've only star-rated 3 books, which were each in "the excitement of the moment" after finishing a particularly agreeable book (to which I gave 5 stars).

Aside from these "outlier" cases, it's very difficult for me to narrow down a particular book to a star rating, as one book might rate a zero for being "engaging", yet be a 4-5 for being "informative", and only be a 3 for "well organized", or whatever parameter/axis you'd care to define for that particular book's phase space.

I've never understood how people can narrow down all these factors to give a book 1-5 stars (unless, of course, it was TOTAL crap or remarkably brilliant).

 

You keep missing the point that 3.8 is very different from 4.2.

Visually, there's no difference at all—both round to 4 and therefore look like .

Now, we've recently been playing with a system where the stars try to show more detail, by basically sliding the color along bit-by-bit through the stars. (Imagine paper and the stars are holes cut into the paper. The ratings would then by colored paper slid alone underneath everything.) In practice, however, you don't get much resolution, and you lose some of the immediate understanding of a .

More generally, is 3.8 usefully different from 4.2? I think at first glance most people won't think so. Only on long examination and reflection—once you understanding that most rating differentials are compressed to less than half a star—will the enormous differences become manifest. Even then, the differences become problematic once you realize all the ways it backfires: omnibuses are better than their contents, series almost invariably get better, Shakespeare's Venus and Adonis is almost a full star better than either The Tempest or Romeo and Juliet, etc., not to mention Marlowe's Tamburlaine , owned by 154 people and apparently almost a half-star better than any Shakespeare play! I know you really love stars, but I think you're trying to squeeze orange juice from a potato. Stars are date you can explain through a process of amusing generalizations about the situations of their starring. They have almost no independent standing.

Anyway, I've added it to the popularity display, together with member counts and so forth. Go crazy.

Easy. I rate for my own use. I don't try to analyze persnickety shadings. I award stars according to my overall feelings concerning the book.

I try to articulate my feelings and opinions in my reviews, and figure if anybody were to look at my library for suggestions, they surely would not rely solely on stars, for heaven's sake, but look to see what I really thought.

"Even if extrapolated to a 100-point scale, that would be a 76 vs an 84, which isn't 'very different'."

My students seem to think that getting a C is very different from getting a B.

"It isn't very interesting to list 400 ratings, 91% of which are 4 or 4.5."

I admit to not reading the entire thread, but I don't understand why you're rounding so coarsely. The first 6 finishers in Sunday's men's 100m dash at the World Championships all ran 10 seconds, more or less. Should we say they all tied for the gold?

>21 cpg:

The problem is that the average difference is smaller than any individual user can themselves decide. It's like 100m dash was only measured in whole integers, but we talked about tenths of a second being significant. They're only really significant when you have many, many data points.

Even then, ratings can't really be aggregated in an intelligent, comparative way. That people tag Romeo and Juliet and Tamburlaine "Elizabethan" and "drama" adds up to something—a picture of what people think. But that Tamburlaine is rated higher than every single Shakespeare play means nothing whatsoever. They're not in a race together. They're in different races, with different runners and a different length track. The people who read them are, as aggregates, different, and their expectations are too. The same applies to a greater or lesser degree across the ratings data. Ratings are often noise that pretends it's data.

Now, clearly this CAN happen to other sorts of data. The tag "classics" is used by two different populations to mean different things. But the problem isn't as widespread in tagging. "Classics" is fairly useless, but most tags aren't.

Re. #20: "I try to articulate my feelings and opinions in my reviews"

I agree ... I write a review for every book I read, and discuss the various pluses and minuses that I see in whatever parameters are involved ... I've never been able to distill these down to a single star rating unless the book was truly fabulous or totally without merit.

 

Re. #22: "Ratings are often noise that pretends it's data."

Oooooh! Sweet! (if a bit snarky)

 

They're in different races, with different runners and a different length track. The people who read them are, as aggregates, different, and their expectations are too.

And your users just may be intelligent enough to recognize this, and look at the ratings in context. Two recent fantasy novels probably have similar audiences, and it would be useful to look at their ratings when deciding which to read. Comparing Brandon Sanderson to the Aeneid would be less productive.

In the case of Tamburlaine in particular, though, 17 ratings is just a very small sample size.

Thanks for showing the data, though.

Every time people argue against ratings, they completely leave out the histogram. Can't imagine why.

My two-penneth is that I found the ratings worth something- there is a book on that list that is quite a way through a series (Dead Reckoning, it's number 11 Sookie Stackhouse). I started that series but dropped off after 4 books as I thought they were getting "a bit poo". This latest book gets 3.66 stars which I'm taking to mean that only diehards are still sticking with this and there's no value for me to catch up the series anytime soon.

For series-watchers the "directionally correct" ratings indication is perfectly well functioned and needs no sort of complex stats trickery to mean something. I bet as well for large volume books that appeal fairly mainstream to a specific genre reader, you're also going to get a fairly accurate review in the ratings. It is "popularity" therefore measuring fairly mainstream books, not neccesarily showcasing a list of massively unique books that would want/ need a specialist personalised rating. Perfectly acceptable use of wisdom of the crowds principles I would think. (disclaimer- I have not looked in any deep detail to support this view).

16>

This isn't a general argument against ratings, but against a specific proposal—to put the on "Popular books."

Gotcha. You've made general arguments against ratings, on exactly the same "people only enter books they like" basis, in the past, so I was somewhat confused.

"The problem is that the average difference is smaller than any individual user can themselves decide."

I don't see why that's a problem. We do it all the time. We don't rank everyone whose GPA exceeds 3.85 first in his/her class. Whenever we calculate a proportion, we're essentially converting a bunch of {0,1} ratings into a quantity with considerably less granularity.

"It's like 100m dash was only measured in whole integers, but we talked about tenths of a second being significant."

Or like if diving judges could individually only give one of 21 scores (0, 0.5, . . ., 10.0) but their aggregate could take on more than 21 values.

"Even then, ratings can't really be aggregated in an intelligent, comparative way."

Then why do you have so many tables comparing aggregated ratings in Zeitgeist?

The people who read them are, as aggregates, different, and their expectations are too. The same applies to a greater or lesser degree across the ratings data. Ratings are often noise that pretends it's data.

If you've got enough data, and use it intelligently, that's not always the case. Sure, a rating of 4.5 on one book and a rating of 4.3 on another book in a different genre aren't usefully comparable -- but if one person rates the same books highly that I do, and the same books poorly that I do, and another person rates my favorite books one star and loves the ones I've thrown against a wall in disgust, that's useful data. This 'evil twin' scenario is a bit of a strawman, admittedly, but I still think that in the context of user-specific data, rather than global aggregates, ratings do have the potential to be very powerful. Do I care how everyone all over LT rates a particular book? No, not really. Do I care how my most-similar libraries rate it? Sure. Do I care how people who like the same things I do -- as measured by ratings, not just by having a book in their library, which they may have not read yet or may have hated -- rate it? Absolutely.

Re. #22: "Ratings are often noise that pretends it's data."

I use the star ratings and still agree with this. Many books I read when I was 14 get 5 stars for the warm fuzzy feelings I have about them.

The scholarly histories I am reading now are rated very differently but still represent my personal opinion and therefore are not really reliable indicators for anyone else.

I've changed the links on popularity pages, so the main link goes to the main work page, and a "popularity statistics" link is added going to /popularity.

32> Thanks Tim, but it looks like all the popularity statistics links go to www.librarything.com/work/ with nothing following the work/.

>33 jjwilson61: - thanks. He's fixing now.

I use the "classics" tag for my own library, to mean what I suspect Tim would want it to mean. It is useful that way to me, even if it gets drowned out in the larger universe of LT data.

I'm very much in BTRIPP's camp when it comes to star ratings. And I think that someone who does have a coherent way of interpreting their own use of star values has no guarantee of matching with anyone else's, so they are like the "classics" tag. But also like the "classics" tag, they can be useful to a user within their library, in identifying books they want to reread, to recommend, or whatever it is that they think makes books star-worthy.

So, is there any chance of seeing hot by months, quarters, and years as well?

I'd still like a link from the homepage module to the full list.

Also, I'm not seeing covers on the full list.

This reminds me, can we get collection checkmarks on the Hot This Month homepage module?

I'd like them on the Zeitgeist page too, if that's possible. I don't know that it would work with the gray, but I'm not a fan of the gray anyway.

Eh?

Which message are you responding to?

Oh, I found the typo. Checkmarks, not checkboxes.

I'd still like covers. And the other four or five things.