Will you like it?

See blog post:

http://www.librarything.com/blog/2009/01/will-you-like-it.php

Sweet! I like it.

"Will you like it?"

I love it! (certainty: very high)

What a neat little tool! I've tried it out on a few of my books and so far the predictions are spot on. I can't wait to use it to find new books.

Thank you so much!

I like the theory. I've tried this with a few works, but I'm only getting a 'Null' error when click the "click to find out" link.

Fun!

I tried it on a few of the books in the recommendations in my home page, and it was pretty accurate (especially since I had already had one of them out of the library) :-)

>4 veracity:
Same here. I was just assuming it was because I'm using my antique desktop at the moment.

I'm getting a negative response on a few books I added to my wislist today. If I don't spend money on those books and the economy doesn't recover it's all on you, Mr. Spalding.

Oh, I know where the NULL comes from. What browser are you using?

>7 DaynaRT:

Ha. I would trust your gut, not it. Also, don't spend money on books. Spend it on crappy American cars.

Same here (the NULL) issue (Firefox 3.0.5) - it tries to redirect you to javascript:loadScore(8400); as a page, instead of running the JS - I shifted it into an "onClick" for that "a" tag, changed the href to "javascript:void(0);" using Firebug and it worked fine. Odd!

Changed the link technology.

Does the javascript:void(0) trick work across browsers?

I'm getting NULL as well. I'm using IE7, fully updated.

Update - its working now for some reason.
I tried it on a few authors I *absolutely hate*, but that people recommend to me all the time. It said I would *love* the books. I think it needs a little tweaking. The only thing it says I might not like is the cookbook I have on my wishlist, and it wasn't sure about that.

I think you missed the update by seconds.

12> I think so Tim, just avoids an anchor - but your fix is working fine now.

I do the return false trick, as you saw.

16> I just read that trick on a blog, I'll have to make a note of that, handy!

One thing needs certainly a fix - labels on the bar are not translatable.

This is nifty! It certainly passes muster with me because my one real and true litmus test was that it MUST bring back a "don't think you will like (very high certainty)" on that piece of schlock which grates on my nerves in particular, Paulo Coelho's The Alchemist. . .and it did so I am good to go! (((high fives Tim)))

I do like it! I was a bit worried at first when it kept telling me I would probably like everything I entered -- I know I read across a lot of styles, but still... Then it correctly predicted my loathing for Lord of the Flies (even though, yes, it IS in my library) and I was relieved. It's a fun new toy.

I love this feature! Thanks a bunch, Tim! Some of my results are turning up a little odd but others are correct to the 'T.' :)

Wow, I like this! Just when you think LT has enough features, the staff manages to surprise ^_^ !

If I had only checked that before borrowing The Ice Child from the library. Lovely, Tim.

On the other hand it thought I wouldn't like In a Sunburned Country, either. But it thinks I will like both the ones I have started now, so I'm ahead of the game, I think.

All I want now is a "what do you really think" feedback... I went to a friend's library, clicked on one of her most recently added books, clicked on "will you like it?" and got a 97% chance I'd hate it... and I wanted to click through with a "85% chance I'd totally love this!!" .... because it's a book she has in a realm I *don't* have in my library, but would love!

Of course, now we're talking a completely new database level for "all the books I would love to love, if only I could!"

Does it use ratings? Why not? Just because I catalog a book doesn't mean I like it.

Seems good though on the couple I tried.

Looks like that should be pretty entertaining.

25: It can't be using ratings, I just checked it against some of my four-to-five-starred books...

Erm, I do like the feature. No grumbling.

>7 DaynaRT::

Seems to be working now. Thanks. Nifty. :)

But I was (and am) using Firefox on OSX. And I had tried it both with books that are and aren't in my library.

I notice that the book page is now wider in 2 of the 3 books I tried it on, and the page extends past the right edge of the screen. FF 2.XX

It was right on one of the books, but it said I would hate a book, I loved, and said I would love a book that is only meh.

We already had Suggester and Unsuggester, this one doesn't look really useful or new. Anyway, seems to work.

Hah - I tried it on books I'd rated and on one it did say I probably wouldn't like it. I gave it 4 stars! :P And I went out and got the Oscar-Nominated film after I read it!

:)

Just picking a little - it seems to be breaking rather nicely so far - I know that there will always be outliers.

It's funny - on the books in the TBR pile, there are two I have that I'm reading for a "co-worker says we should both read these!" sake and I've been putting them off...it's predicting I won't like them. It's almost like it's been hanging out in the living room with me.

Does it assume I would love books that I own? Because I don't, necessarily.

It just told me I'd love Fraud, which I liked, but didn't "love", and that I'd love Thrones, Dominations which made me want to strangle someone. And it said I'd merely "probably like" one of my favorite books ever.

Query: why is the number of copies on LT relevant to whether I will like a book?

I like it. It is interesting. It tells me I won't like Lord of the Rings yet I have it in my library, which I did read just never wrote a review for it. Just fun to see what it thinks I will and will not like.

This is really neat.

After it corrected identified one of my favorite books as one I would love (certainty: very high), I thought, it would be really cool to have a list of books the tool thinks I would love. Checking about specific books is fun, but that way I could discover books I'm not even aware of!

I'm sure this would take a massive about of processing power, but I thought I'd throw that out there.

It needs to take ratings into account, when they exist (esp. for books I own!)...and why does it say "certainty: very low" when it clearly means "very high", and vice versa?

Interesting. I tried it on Tolkien as well - it said I would like The Silmarillion (which I never managed to finish reading) but it said I would only probably like The Lord of the Rings which I have read and enjoyed many times. Looking at the 3 volumes that make up LOTR, I only got Will Like, Probably Will Like and Probably Will Like (despite having two copies of each). Now I have entered all my books perhaps I should start rating them...

I like the concept, but it doesn't really seem to work--it seems to be wrong as often as right.

What are the grounds for its high certainty that I probably won't like The Hunger Games? My library has lots of dystopia and YA.

pretty cool but if someone has it in their library you should also have it check the stars rating. The highest rated book I've tried was my most hated Agatha Christie, now in general I really like her and the synopsis of the book would suggest this to be really great, but it was very poor. But so far for the most part it has been fairly accurate. Fun stuff, now I can't wait until next week! ;)

37> in the blog Tim said this was mostly a lark and for fun, so don't take it too seriously...

Its really a neat app. I have one question, which is a better recommendation - probably will like: certainty - very high or will love: certainty - medium? It sounds like its only medium sure I will love it. Or does it mean I will definitely like it just maybe not love it?

I think this is an awesome tool, if one day it can be made to work better - especially taking ratings into consideration.

I came to LT from other book-tracking sites because of the social aspect and, largely, the recommendation system. This seems like something that could really set LT another step apart in that bracket!

That said, it's been pretty far off on more than half the books I've tried (I had the same problem as 37, with the same book, in fact - lots of dystopia and YA, but I'm probably not gonna like Hunger Games - which I read recently and rated highly in my library). So I do hope that one day there's time to shore it up a little!

I love the idea!

This is great! It would be really cool if we could readjust the ratings for books that we know and like/dislike, and have it keep those in mind for adjustments. Ah, complexity. :)

Edit: Not necessarily 5-level rating, just thumbs up, thumbs down to counteract some of the badness.

So far every book I've tried that I love and Actually own... It tells me I probably won't like it?

Does this thing even look at what I've rated?

Books I've already given 5 stars to are reported as books I probably won't like?

I think it's safe to assume it doesn't look at ratings at all. I would expect that the question it answers is more or less , "Do people who list this book in their catalogs have many or few other books that you list in your catalog?" And that there is weight given to uncommon books over common books. I would expect that niche books will fall on the ends of the spectrum based on whether you have other books from that niche and common books will fall in the middle because they don't correlate too strongly with any subset of libraries.

The confidence rating seems to be directly proportional to the number of copies of the work in LT. This may be misleading as I suspect that books with a few hundred owners will return more meaningful results than books with a few thousand.

Overall, it strikes me as a bit of fun and perhaps a useful check on questionable suggestions by the Recommendations tool.

Also, I'm getting an error occasionally:
Warning: Division by zero in /var/www/html/inc_willilikeit.php on line 97

This error accompanied all but one of the books I checked that was in the "won't like" category, and the marker was all the way to the left in these cases.

edited to add:
I just checked the top 25 books from the Zeitgeist and got 1 "will like," 10 "will probably like," 13 "will probably not like," and 1 "won't like." The "will like" was at it's lower border. There was no particular distinction between the pop books I would never read and the classics that I have read and love (The Da Vinci Code and To Kill a Mockingbird received essentially the same score).

So it looks like the common books are not falling in the middle, but the low middle. This may be just my catalog, or it may be wider spread and help explain some of the frustrated comments above.

oooh, this is fun. It gives me funny results sometimes (The Shadow of the Wind is a book that I will "probably not like" but it was one of my favorite reads this year), but I like being able to look up a particular book and see what it says.

I really do wish there was some sort of recommendation feature that took ratings into account. My Library, for instance, also includes all of my wife's fantasy books, but our fantasy tastes diverge wildly: books of The Belgariad all show up as "Will Love (certainty: very high)" even though I've got them all rated 1s and 2s. Fact is, if I'm looking for fantasy recommendations, all of the Eddings, Brooks, Feist, McCaffrey, etc. in My Library (belonging to my wife) are going to make any recommendations suspect.

Other than that, it's a pretty cool feature.

It seems to think I won't like anything, even books I absolutely love. The feature is a lot of fun, it just doesn't seem to be very accurate for me.

Now I kind of feel like I have to find a book it says I will enjoy. Maybe all this time I haven't liked anything and I just didn't know it!

47: Add a tag for your books vs her books. Or just the books that you want to use for recommendations.

49> I do, actually, but that won't help with this new feature. Will it?

Some comments:

How it works

1. What the algorithm is fundamentally telling you is "Do the 1,000 users most like you have a statistically lower or higher number of copies of the work?" That's it.

Simple as it may sound, that's a tough piece of data to calculate. Not in theory, but in practice. It requires calculating the 1,000 users, which is the hardest calculation on LibraryThing (it hits memory once for every work, and then once for every owner of that book). Then it does some rather complex statistical sampling and analysis, involving a few thousand more hits. I'm not going to spell that part out more, insofar as it's the key to moving it from "terrible" to "meh" and, well, I gotta have secrets, mom.

The basic problem with the method is that it's a secondary calculation--first LibraryThing needs to compute a good 1,000-members list for you. If it's not doing that well, Will you like it won't be better, and will probably be worse. Second, because it's about who shares your books, it can fail to understand your individuality. As I wrote, most people who like Greek history--a big piece of my collection--also like Republican Roman History. I don't, but LibraryThing thinks I should.

Simillarly, my two largest sets of books--Greek and Latin language and history, and computer programming--match up nicely with people who think they are above Harry Potter. People who read Plutarch and PHP in the original are snobs of a sort, generally, and have an unusually low number of Harry Potter books in their libraries. But I love Harry Potter! Well, that's life.

Why not use ratings?

1. Ratings are not very useful. Most people like their books, and they rate them within a point or two of each other--the average is 3.80 stars and the standard deviation is 1.03. Further, only about 5 million books have ratings, out of 35 million books. When you're comparing small population samples--the people you are closest to who also have the book vs. the entire population of LibraryThing--having only 1/7 of all books rated is a problem. Think of it this way: If every one of your closest, bestest friends have seen a movie, that's good data. That one of them gave it three stars and you don't know what the others thought of it isn't good data.

Netflix uses ratings because we more often see movies we don't like. It's less of a commitment and we often see movies socially, ceding our movie choice to a loved one whose idea of a good time is a romantic commedy with a dog, not the bullets and mayhem that we like. So the ratings have a higher standard deviation--the data is more interesting. Books are different. Also, Netflix doesn't know every movie you've seen. It only knows what you've rated their stuff.

2. It's technical. For speed reasons LibraryThing stores a complete list of all books you have in memory, available at any time without going to the database. It also stores a list of every user who has a work. Both are in theory; in practice only a portion of users and works are in memory--the portion it had to think about recently; the rest it gets off disk. To find your 1,000-closest we run thousands of queries against this memory--and then store the result in memory as a simple list.

The data is simple to keep the memory size down. After all, we have almost 600,000 members and 35 million books. It doesn't include ratings because ratings aren't as useful and would swell the data size. Getting it off the disk, however, would bring the site to a crashing halt.

Criticism

1. It's cool to talk about, but keep things in perspective. Read the blog post. I said "How good is it? Meh It's okay." It's clear from the posts here that many users think it works well (far more than I thought!). If it doesn't work well for you, well, sorry, but, as I said in the blog post, it's a lark.

2. I'm amused by people who say--paraphrase--"It's broken. It tell me I won't like a book I've already given five stars to?" Well, yes, we could short circuit the algorithm for every book you've rated and parrot back your own stars, but what fun would that be?

Perhaps you can go get blue ring instead of trying on a mood ring. You already know you're happy.

I'm enjoying proving it wrong. :-)

I wasn't trying see to if I like the books I like. I was trying to see if the thing was trustworthy at all. For me, apparently, it is VERY untrustworthy. It's no big deal. I never use the recommendations feature either for this very reason (and because all it seemed to do was tell me to try books I already owned.)

I've said it before and I'll say it again, I'm here so I can track my books and do a bit of socializing. The rest is gravy.

Well I like this mood ring. It keeps telling me how unpredictable I am.

Wasn't there a suggestion at some point that, after Collections had grown into toddlerhood, we would be able to exclude a collection from all this sort of thing? That would make it more accurate, surely?

So, I guess I know what the problem is: I just don't think LT is very good at figuring out the 1000 users most like me. It might be interesting if we could provide our own list of people and have the feature work from that.

Also, I think you're vastly underestimating the power of ratings. Maybe it's not feasible to use them for this feature, but it often seems like you dismiss ratings automatically based on the not-very-good argument that the average is high.

I just started getting this error when I try this feature:

Warning: Division by zero in /var/www/html/inc_willilikeit.php on line 97

It worked well at first.

Safari on XPsp2

56> Actually, I'm only getting this error for a couple of books (so far) -- they just happened to be on consecutive tries. Here are the books:

Charles Ives and His World
Beyond Suspicion (touchstone doesn't work for this one)

I get the same error for those books.

This error seems to occur with books that only a few people have, I encountered it once with a book owned by only nine members, but everything else has worked fine.

The divide-by-zero error should be fixed. I was checking for zero. I wasn't checking for values that, when all the floating-point math was done, were so close to zero it treated them as zero. Math!

Incidentally, because it relies on the cached members-with-your-books, it won't do so well if you are a new user adding books. You might, however, try visiting your profile first as, under some circumstances, it recalculates your members-with-your-books if you haven't done so in a while, as measured by books added.

Oh, not fixed. Weird.

Don't know if this helps, but based on enrique_molinero's comment, I sorted my catalog by number owned and discovered that all books owned by 5-8 members produce this error. Nine seems to be a magic number...

Cool, its Genuis for books

Ah, this is just poetic. I will like the Purgatorio better than the Inferno, and the Paradiso better than either. Wonder how universal that is.

This is fun ... I've just been going through my 'read' tag and seeing how many times I foxed it! It's surprising how many books it says that I will probably like that I did ... especially since there is a certain element of randomness about the books I own, not always bought because I think I'll like them ...

What I particularly like is that it tells me I'll basically hate the whole lot of Twilight books, which I read voraciously but with a slight guilt complex as I felt that I shouldn't like them ... the magic algorithm has now confirmed that I really shouldn't like them. I feel vindicated! ;)

Enjoying it!

Oooh, I think I'm going to love it! Hey, it's got to be yards better than the time Amazon told me that "People who bought this book also bought XYZ socks" -- !! What an algorithm! Needless to say, I clicked immediately on the socks and ordered a dozen pair, and I have to tell you that my life was immeasurably improved thereby!

>64 235711:
This is entertaining, but I would say it is an artifact of the number of people owning each work: 7,139 for Inferno, 2,191 for Purgatorio and 1,766 for Paradiso. (Far more interest in hell than heaven. Now what does that say? But I digress.)

I've been playing with "Will you like it?" a bit and I've found that the more common books (more than a couple thousand copies) tend to fall in the middle with little deviation while the less common books are spread around the scale better.

Tim, I think you need another factor to push out the deviation for the common books. And thank you for the more detailed explanation above.

I have to admit, it's been much more wrong than right for me so far. One example I tested with was my ER request A Night Out with Robert Burns: The Greatest Poems which I didn't get. It predicted I wouldn't like it with very high certainty. I've already got 8 other Burns works in > 100 volumes of Scottish poetry!

The reliability does seem to very strongly correlate with the number of shared users - at least 20 to have any chance of a sensible result.

Yes, Tim, that explanation helped a lot! You never cease to impress me with the new features to the site.

#66 stdunstan, it cracks me up that you bought the socks. Though, as a sock-lover myself, I don't blame you a bit. :)

67: I am incredibly naive. I knew of the popular tendency to stop reading at the foot of Mt. Purgatory, to refer to "Dante's Inferno", etc. It was so wonderful to seemingly see the thing work right that I didn't even think to connect it with the numbers. *Sigh.*

One more question -- is this meant to be a one-dimensional scale or a two-dimensional one? That is, is something like "LT thinks you will love this book (certainty: very low)" a possible outcome?

71 > yep

Now if we could have this feature scan a list based on a particular Tag, then suggest in sequence what book we would like most. I could have it suggest which of my unread books I should read next.

The question-mark help links to Common Knowledge help right now .....

73> Isn't that what the Recommendations feature does already?

>75 jjwilson61:
No. The poster wants to be able to indicate a specific tag (i.e. unread) from their library and have the books from their own library with that tag sorted in order based on how they "rank" in the "Will you like?" algorithm.

73: That would be cool. :)

This is entertaining, but I would say it is an artifact of the number of people owning each work: 7,139 for Inferno, 2,191 for Purgatorio and 1,766 for Paradiso.

In theory, not, because it compares ratios, not absolute numbers. But your connections may be more or less superficial in their reading of Dante ;)

Tim, I think you need another factor to push out the deviation for the common books. And thank you for the more detailed explanation above.

You mean publish the standard deviation of ratings for books?

I've been playing with "Will you like it?" a bit and I've found that the more common books (more than a couple thousand copies) tend to fall in the middle with little deviation while the less common books are spread around the scale better.

Yes, that's true. It's not common so much as "commonly read." So I suspect that, say, the Purpose Driven Life, isn't spread around the scale for you. It's common, but it's not evenly distributed among users. But Harry Potter or 1984 are just books that people have, whoever they are.

One more question -- is this meant to be a one-dimensional scale or a two-dimensional one? That is, is something like "LT thinks you will love this book (certainty: very low)" a possible outcome?

The arrow shows where on the two dimensional scale you are. If it says "you will like it" but you're arrow is just over the line with "you may not like it," take that for what it is.

The second dimension is, basically, statistical error. If a book has few owners, it's harder for LT to make good predictions. In fact, we originally displayed the result as two arrows--the farther apart they were the fuzzier the prediction. But it wasn't as immediately satisfying, so we reduced it to the "confidence" number. You can get the actual numbers for both by hovering over the down arrow.

Now if we could have this feature scan a list based on a particular Tag, then suggest in sequence what book we would like most. I could have it suggest which of my unread books I should read next.

Yeah, it might be interesting to apply it elsewhere on the site. But processing time is a problem. As you can tell, it doesn't come back instantly. Doing it for many books at a time would be a processing problem.

Incidentally, over the last year or so I've played with a dozen algorithms to do this. This is the best one! (It's a hard, hard problem.) I am interested in doing one that uses, basically you tag-mirror against the tag cloud of a book. I think that might also work. But there's no straightforward math for comparing the similarity of two tag clouds.* So I want to get that in place first before I do variations on it.

The question-mark help links to Common Knowledge help right now .....

Thanks. I am an idiot. Will fix.

*THAT is an interesting problem, btw.

The Amazon algorithm has gotten very boring. It used to come up with wild stuff. Like the time I ordered a Mexican cookbook, and it suggested a book on how to build a dinosaur skeleton out of several chicken carcasses.

Yes, we bought the book, we had a dinosaur-loving son. No, the skeleton never got built, and he gave up on paleontology half way through college after over 1o years of 'I'm going to be a paleontologist when I grow up.'

I like it, even if it ended up with a few kinks in its accuracy. As in, it said I "will probably like" Persuasion with a "very high" certainty. If the little triangular pointer is a prediction of about what I'd rate it, it says I'd give it a 2.5 or so -- I rated it a 5.

I wonder, will there eventually be a "why?" link like on Recommendations?

The why is statistical. There is no why; there is only the number!

78>

Thanks.

The only cases I looked at had confidence correlated with strength of opinion, so I wasn't sure if they were actually independent or not.

>81 timspalding:
I'm starting to really dislike numbers.

How 'bout adjusting prediction to include people's ratings? If I rated Persuasion twice what I thought it would, it's gotta indicate that I'd probably like some other books more than it would think too, right?

You don't mean "probably like" and "probably dislike", do you? Wouldn't "mildly like" and "mildly dislike" be better descriptions? The "certainty" ratings give the likelihood that the program's estimate is correct.

How 'bout adjusting prediction to include people's ratings?

Tim addressed that question in message 51.

>84 TomVeal:

No, that's too schematic. The rating itself is a sort of statistics of probability.

I don't know how useful it will be for me, since there are plenty of books in my library that I don't like (I catalog every book I read, whether I enjoyed it or thought it was awful), but I think it's a fun feature regardless. I just wish it were possible to take ratings into account :/

"Well, yes, we could short circuit the algorithm for every book you've rated and parrot back your own stars, but what fun would that be?"

It would save processing cycles, for one ;)

Though if you want to satisfy everybody you might want it to display both the prediction and the star rating from the person's own catalog. It'd be fun IMO, though the shortcomings would be highlighted...

I really like this new feature (yay!) but I haven't tested it enough yet to be able to have a firm opinion on how well the algorithm performs.

Just a (possible) re-statement of what this measures, for clarification and my own understanding:

By looking at the 1000 libraries/users most like mine, how likely is this book to be in my library.

Is this correct?

>By looking at the 1000 libraries/users most like mine, how likely is this book to be in my library.

Yes. Or even simpler, relative to other books, how often do your 1000-closest members have the book.

Awesome, thanks!

>85 sqdancer:
Ah. I missed it because there were already close to 80 posts when I found the thread.

Here's an alternative suggestion. Why not add a new stat to one of the user statistics pages? It could be some sort of fun statistic comparing ratings we've given our books to the predicted value. As in, mean/median distance of ratings from expected value?

>92 AnnaClaire:

Good idea, but I'm trying not to do this too often. It's tough math.

Also, it's somewhat circular. You connect to people because of your books, so those people will have your books. (But the effect shouldn't be that great.)

This is a neat idea. I'm going to predict (and then go see if I'm right) that it'll be a lot more accurate for me with some kinds of book (books I own because *I* think, or thought at some point, that I'd like them) than with others (books I own because they're Important Books In My Field, whether I like them or not). And since so much of my library is nonfiction, it'll also be an interesting way to see how much tastes in, say, political theory correlate with tastes in, say, poetry....

This seems to be more accurate than the Suggester/Unsuggester. I don't really know how you did this but I LOVE it!!!

>86 timspalding: To say that the certainty is "high" that someone "will probably like" a book is nonsense if taken literally, but I suppose that users will realize what is meant.

Notwithstanding this quibble, it's a nice feature - and an example of something that Amazon can't do.

Maybe there simply aren't 1000 people on LT most like me. In fact, I'm sure of that (certainty: very high) :).

>96 TomVeal:

Let's say I learn that you like chocolate custard and chocolate cake. I am fairly certain that you will like chocolate ice cream.

Now, let's say I am inviting you over for dinner. I know that I had chocolate ice cream, but I'm not sure if my wife finished it off. I think she did not and that there is some in the freezer.

I might say that there is a good chance I have an ice cream at home that I think you'll probably like.

In theory you should be able to add the probabilities together, but it's not nonsense to separate probabilities that arise from different sources.

98>

I'm sorry, but I agree with TomVeal; if you have a "confidence" axis, then don't use probability terms to describe how much you think someone will like it; "we're almost certain that you'll find this book okay" is a very different statement from "we think it's possible that you'll adore this book", but using terms like "probably" for both sets of measurements confuse the issue.

I'm still getting the division by zero error on several books I've tried. All were books with fewer than 10 members.

I agree with #95, as I try out more and more it does seem to be more accurate/precise than the suggester and unsuggester. Hope you didn't think my above remark (46) was complaining that it didn't work. I don't expect it to be able to take into account not only my ratings of books I own but also all the books I've ever read and not put into LT - it would basically have to be a mindreader to be perfect.

*bouncebouncebounce!*

ANOTHER fun toy to play with!! Keep adding fun features, Tim, and I'll have to start *living* on LT!

There is of course a margin of error, but yeah, as long as you keep in mind that it's not serious and don't expect it to be right all the time, it's really REALLY fun!!

..... I think I'm going to go look at all the books I've rated lower then 3 stars and see if it says I'll like them.

*has already gone through much of my wishlist with this new toy!* Wheee!

I'll be playing with this for sure. Although, it did tell me I would 'probably like' my most favorite book ever. And, it also told me I would love about 3 books (that are in my library) that made me want to open a vein and run bleeding through the streets.

Fun stuff. Thanks!

Still getting that divide-by-zero error:

Warning: Division by zero in /var/www/html/inc_willilikeit.php on line 97
Loading...
won't like will probably not like will probably like will like will love 0, uncertainty 0

LibraryThing thinks you won't like Transported (certainty: very high)

Which is a bit of a worry, since I wrote Transported.

Slightly more seriously, as an author, I'm not sure I like this feature - as I understand it from the discussion above, books with only a few copies in LT will always get a low rating, so a new book is always going to get a "you won't like it" rating while only a few people have it in their libraries. Am I understanding this correctly?

>104 timjones: "Division by zero" results aren't valid, so it's still possible that you like your own book.

When LT users have only a few copies of a work, the certainty of the rating will be low, but the ratings themselves can fall anywhere on the scale. I have a number of "you will love it - certainty: very low" books.

It should have been refusing to rate books with fewer than ten users. It's doing that now.

Tom is right about the issue of low ratings vs. low confidence.

106>

I think you broke something now. I get "There are too few copies to analyze whether you will like it." for every book I try, including Homer's Odyssey and Shakespeare's The Tempest

I think will you like it is a misleading name, as the algorthym is using nothing about you.

A better name is along the lines of:

What other people* probably think you'll make of this book

* Assuming that other people have only cataloged books they like.


Ratings:
If the average rating is so high with so little deviation the fact that some of the libraries being used to make the decision rate it at 1-2.5* is then very significant.
I accept that technically it's a hard too include them, but it would make it much better.

#s 105, 106 - thanks for the clarifications, Tim and Tom.

>108 reading_fox:
And Collections should help counter the tendency towards high average ratings with low standard deviations, as purists like me will have less of a problem keeping everything in one place.

Isn't the meaning more like "You will probably have this book in your library" rather than "You will probably like this book".

Regardless of what you think of the wording, we all understand it. I don't see what the big deal is.

Here's something to chew on: I "probably will like" Mind the Gaffe! (certainty: very low). The pointer indicates I'd probably like it only a little less than Persuasion (which was itself earmarked for half of the five stars I gave it).

Googling Mind the Gaffe! was what brought me to LT in the first place.

I agree that the name of the feature is misleading. I find it much more satisfying and prefer thinking about it along the lines of "This book would fit in with your library overall."

I like this feature, Tim. It's become a game to me. :)

I put my cursor on the line at the spot where I would rate the book and then wait to see what the bar graph shows when I load it. I like to see how close it comes. Fun!

To quote Tim again...

"but it's fundamentally a lark and a conversation piece at present, so I don't want to waste too much time on it."

Regardless of what you think of the wording, we all understand it. I don't see what the big deal is.

I don't see why it's a problem to suggest changing the wording to something clearer. I don't think people who see the feature without reading this thread will automatically understand it; the average user probably thinks "liking" involves ratings.

Regarding using ratings:

What if, instead of trying to do the (nearly impossible and potentially meaningless) task of including everyone's ratings, just include the ratings of the individual doing the asking? They could go in as a weighting scheme of some sort (when calculating the 1000 most-similar libraries, perhaps?), and be self-calibrated (so, e.g., if a user has all ratings between 0.5 and 3 stars, a 3 star rating is REALLY GOOD whereas if a user has all ratings between 3 and 5 stars, a 3 star rating is REALLY BAD). Anything unrated could just go in with an average rating, so that if the user hasn't rated anything then the algorithm is unchanged.

If the algorithm asks "relative to other books, how often do(es) your 1000-closest members have the book?" it's likely to be unhelpful with very popular widely-spread books like Harry Potter and Shakespeare. Should it not, therefore, ask "how often does your 1000-closest members have the book, compared to 1000 randomly selected LTers" and secondarily "how highly do your 1000-closest members rate this book, compared to 1000 randomly selected LTers" (processing issues aside)?

Then there's people like me who screw your math because I own 1000s of books I haven't read yet and my catalog is of all the books I own. Therefore I may own many books I will hate. So comparing my library to others for likes/dislikes... well...

>120 inkdrinker:
Exactly. This feeds into why I think Collections will improve Will You Like It's results, at least somewhat. I'm a purist (and I know there are enough purists around that I don't have to explain what that means for my catalog). My current work-around is multiple accounts, but when Collections come, I'll be able to merge books I've borrowed or sent to a new home can be included in whether or not I read a book.

Though, on the other hand, perhaps Collections will only improve things if we make some effort to take ratings into account. Even if it is only something like mollishka's suggestion in #118.

>111 jjwilson61:, 114, 117, et al

I guess it depends on what you think "like" means. There are books that are about a topic that I "like," that are written by an author that I "like," and that I would definitely "like" to have in my library -- yet, when I actually read them, I find that I don't "like" all of them equally (if fact, there are inevitably some that I don't "like" much at all). If the books are in a category of particular interest to me, then I still "like" owning the book and I still "like" having read the book (if only to discover that I didn't "like" it).

I think Collections will certainly help. The ability to separate out books into main library and the various miscellany that I own, but aren't helpful for recommendations (and in many cases should be read or disposed of) would be very powerful.

Found another division by zero error, this time on this copy of this work. Get the error both places.

>122 SilentInAWay: well, it depends on if you like it, like-like it or like-like-like it. :)

Ah - flashbacks to the seemingly logical junior high best friend ratings! Best-best, bestest, best-closest, best-phone, best-lunch, best-school, best-best-best...

Tee!

I think the thing is that we expect something that has to deal in a certain amount of logic to handle the rather illogical thing that is a person. Sort of like me saying I love polenta but hate grits. I've had more than one person say it's impossible, I say it's not.

Message 120: Then there's people like me who screw your math because I own 1000s of books I haven't read yet and my catalog is of all the books I own. Therefore I may own many books I will hate.

I also have a huge number of unread books in my library and I will be really pissed if I end up not liking a large percentage of them :).

#126, That's a laugh, but I also own a lot of books that I haven't read and may not like, by obessively collecting books in one particular field far faster than I could read them... and then my tastes changed...maybe I might go back there one day.

125> Love polenta, but one time some Southerners forced me to eat grits - Bleh!

(Polenta is best with Venetian Sepia Nero over it. Looks really awful, but the taste: Pure heaven.)

>128 MarthaJeanne: Ha! I'm not crazy then! I was told Polenta and grits are the same by a grit lover. I say no! :)

Braised shortribs + polenta = awesome.

I don't know about the names, either. I mean, let's take Philly Stakes, a mild series mystery about a school teacher. The system says I'll love it, probably because I've got several other books in the series and quite a few other mysteries in that style. And I did buy even though I had already read it. But I don't love it, and I think a look at the ratings should show that few really _love_ this book; an average of 3.29, with all but the outliers between 3 and 4 inclusive. You could use language that's more clear about exactly the algorithm is saying.

(Warning: I probably don't know what I'm talking about here.)

I wonder if part of the problem with the algorithm is that it doesn't take into account how users catalog their books. I haven't been a member long, but I've already noticed huge differences in cataloging styles (and just to be clear, I'm not talking about tagging).

Personally, I've found that the weighted books feature doesn't work for me because I have a disproportionately large number of books in my library from one author who I love, but I read almost nothing else of that genre. I also don't have very many books from all my years of reading because I'm not good at remembering what I've read and, frankly, I just don't care to have a list of every title I've ever finished. I'm entering books into my library to keep up with what I've been reading lately. That also means that every book that goes into my library now has been read and gets a rating whether I liked it or not. Without factoring in ratings and because of my cataloging style, the algorithm is factoring in how good I am at picking books out that I like, which I will admit I'm not that great at.

A huge problem I see with this is that a lot of LT users catalog TBR/wishlist books in with their read books. I have a separate account for my wishlist, which means that I (my only read library self) will never have any books in common with the other I (my haven't yet read library self). Yet, I'm the same person with the same tastes in books.

Perhaps if we could have a feature that allows us to choose which cataloging style we have and then the algorithm could run against others in that list, it would help. That would also allow past users who have abandoned their libraries to be excluded and allow those of us who rate every book in our library to be factored against each other and include our ratings.

Just my thoughts. Please feel free to point out the holes in my thinking.

I originally thought that using people's high rated books to guide choices would work best, but that would be a problem when people don't rate their books.

However, it should be easy to make a comparison that excludes all books that the owner has rated less than 3 stars.

I'd certainly like my low-rated books to be excluded. I record books I've got rid of on Library Thing so that I know not to buy them again.

Well, see my comments on ratings in 78. In theory we could eliminate books you rated poorly—although I'm not at all sure we should. But the algorithm works off the 1000-closest members, and that doesn't consider ratings. Making a separate ratings-based "1000-closest" isn't going to happen with current resources.

I'll look into having ratings *factor* into 1000-closest. We could probably do it on "one side" (your books). Because the algorithm compares your books to this very simple book -> work structure I spoke of, it's probably not going to take into account what *other* users thought of it.

Incidentally, a single book is NOT going to effect your love/don't love score very much. It's based on how you relate to other libraries, and those lists don't change that much based on one book. People cluster. A cozy mysteries reader is a cozy mysteries reader, irrespective of whether you take a book out here or there.

134: (secretly wonders if that's why it doesn't pick up on my love for bdsm-related books, since there are only a few in my library)

LT thinks I'll hate the book I got for my birthday - Dewey: The Small-Town Library Cat Who Touched the World.

>136 felius:

That was an Early Reviewer book. Too bad employees aren't allowed to enter.

>135 Heather19:

No, you don't understand. With BDSM books, since pain is enjoyment, the ratings are reversed!

137: LOL! *just seriously burst out laughing!*

Well the new function thinks I won't like Concept Dawg's Test Book. And it gives me a Division by Zero error.

Warning: Division by zero in /var/www/html/inc_willilikeit.php on line 97

LibraryThing thinks you won't like Test book (certainty: very high)

#137 -- See, this is why I want to be able to STAR individual posts, and not just topic threads.

>140 lquilter:
Either go here, or bump this thread (again, since you were the last to add your voice in support).

I just noticed this feature and clicked it. Because of the results, I tried several more books in a few of my favorite areas. I've concluded it doesn't work at all for people like me, as do some other features like similar libraries.

I have a huge SF collection compared to some other areas I'm equally interested in but own substatially less books in, but which are still good sized collections in their own right. For example, LT seems to think I probably wouldn't like most astronomy and poetry books, even books I own because I found them interesting.

Would collections, or my rating more books, help? As it is, I wouldn't trust it, but I like the idea a lot. I also see how it could be a very difficult analysis to get right for all people with all types of libraries.

I wish I didn't have to go public with this, but I can't find this feature. Where is the "work page" you're talking about?

Pick a book, any book. Go to its record (Main page, not Edit or Details), and look down below the Tags and Recommendations blocks; the next one is Will I Like It.

Thanks!

" Instead, "Will you like it?" works directly from the data, examining the users who have a book and how their books relate to yours. "

If collections are implemented, will this feature be able to be used on a collection? As it is now, it looks like I would need to open a second account to get any recommendations for any of my interests other than SF. I'd rather keep them in one library, but I'm getting favorite authors that LT tells me I won't like with high certainty.

Or I could just avoid the feature, but there is something compelling about it. :-)

So, the goal is to have recommendations collection-by-collection. This won't be rolled out immediately, as there are big-time resource issues with it. But Will I Like It is very much an item-by-item thing, as it takes a while to calculate for just one...

I like the feature, but the little pointer hoochiedingus is never in the place the text says it should be. The reliability is better-than-average since it predicts I will like (certainty low) books I thought I would like and ended up not liking.

What browser, OS are you on?

#148

The pointer on the linear graph indicates if you will or won't like it. The text is how sure the program is of it's answer. At least that is how I understand it.

IE7 updated, Windows XP.

Pretty blah stuff, eh what?

I looked up my all-time favorite books on "Will You Like It?" and each time it said "you probably won't like it". I don't understand it. LOL. So I don't trust Will You Like It.

I think that a lot of people have books in their libraries that they don't like. We don't happen to love every book we read, even though we may want to keep track of what we read on the site. I think "Will You Like It" shout be determined on what you *RATE* your books.

If it was determined on ratings, it could gather the ratings you have given to similar books and thus determine if you'll like it. For example, if you give "Sense and Sensibility" 1 star, meaning you hated it, "Will You Like It?" should say that you probably won't like "Pride and Prejudice" based on the 1 star rating of a similar book, rather than whether or not you have a lot of Jane Austen books in your library.

>152 cosmiclove22:
Tim's usual response to this kind of suggestion generally includes some mention of the fact that many books cataloged on LT go unrated.

Of course, if ratings actually did anything (besides look pretty in our catalogs), maybe more people would use them.

>152 cosmiclove22:

If you want to read Tim's actual response regarding the idea of using ratings, see messages 51 and 133 in this thread.

I'm getting a "division by 0" error for Transformers Legends. It's also saying the I won't like it with high certainty (which seems odd given my catalog) but I assume that's just a side-effect of the division error?

155: I'm not getting a Division by Zero error. Try refreshing?

Yep, even after some CTRL-F5s and a browser reboot. (I'm on Vista/IE7, btw, though I've not seen this error before.) Could it have something to with my specific library?

Anyway, the exact error is this:
Warning: Division by zero in /var/www/html/inc_willilikeit.php on line 97

Division by zero for me, too (Firefox 3 & IE8).

Works just fine for me (XP & Firefox 3)

Works ok for me too (IE7 on XPsp2)

Same error last night at home (XP/IE6).

Odd. You should probably post this on the Bug Collectors group ( http://www.librarything.com/groups/bugcollectors ) - much more likely that somebody is around who can give you advise or even fix it.

LOL, I hope this algorithm isn't part of the decision making process for ER books! I have two books by an author in the March group and the conclusion is that I probably won't like them.

It works quite well & is fun, but with a few glitches.

It brings up Harry Potter Box Set 1-6 as a book I wouldn't like. The same set also appears in my unsuggester. This is strange, because I own all Harry Potter books (the first 4 as a set, the rest as single books) and have given consistently high ratings to all of them.

#164 Oh, I agree it's fun.

It's just consistently offbase, telling me I definitely wouldn't like many books I like. I find it wrong more than right. It's possible I have just the combination of varied interests that break the algorithm. I figure that's pretty unlikely that I'm that special though. :)

>164 yapete:

That's because very, very few people have both the individual books and the 6-book set. :-)

I figure that's pretty unlikely that I'm that special though. (#165)

I feel the same way -- and I grew up watching all those TV shows that made a point of saying I was.

But really, how on earth can it have a "low" certainty that I'd like The Summer of 1787? In fact, it's opinion of my opinion is all over the place with American history.

It's not an opinion, it's a statistical analysis based on probabilities. If it is really wrong about you and your tastes, it generally means your tastes are outside the expected (at least for that book).

167> There's not enough people with The Summer of 1787 for it to speak with confidence whether you'd like it or not. The certainty is not its opinion of your opinion; the certainty is its confidence in how well it knows the book.

Nice feature, but a bit of a misnomer.

Maybe it could be renamed, "How well does this book fit in my library?" or something like that.

>170 nsblumenfeld:
Agreed.

LT thinks I will like The Communist Manifesto. Gah.

#172> Well, duh. Of course Adam Smith would have "liked" The Communist Manifesto in the sense that he definitely would have wanted a copy for his library.

>173 prosfilaes:
Oh yeah? Then how come it also says I'll like The Story of O? I mean, I live with my mother for god's sake.

What, you've never hidden anything from your mother? Like I believe that ;-)

>170 nsblumenfeld:

I disagree. It's pedantry. All algorithms that infer an unknown from statistical data could be re-expressed to avoid predicting. Ditto next time your friend says "I know you'll dig this chick I know!" That would be "Based on available, if insufficient evidence, this chick I know fits with your previous relationship patterns."

#172> Well, duh. Of course Adam Smith would have "liked" The Communist Manifesto in the sense that he definitely would have wanted a copy for his library.

Exactamundo.

176> But your friend knows which relationships were lousy and which were good, and can apply intelligence to the pattern. They don't blindly extrapolate from which girls you've dated before. There is a difference between "I know you'll dig" and "she fits with previous patterns".