Can This be Done?

One of my favourite features of the site is the recomendations, but it has a problem. I get recomendations based on books that I gave very low ratings (i.e. The Redemption of Althalus, which I think I gave 1 star).

Can it be made so recomendations are only given based on books 3.5 and above, and unrecomendations for 2 and below or something?

It probably could be done, but it's very unlikely. There's a lot of disparity in how ratings are used on the site and Tim has said in the past that he doesn't want to do recommendations or anything based on ratings.

If you want to not get recommendations for books that you've assigned a low rating to, you can always use collections to separate them, so that only books with ratings of 3.5+ (or whatever rating you want) are included, and only that collection is set to be used for recommendations.

This is a much requested idea, but Tim has said that he does not want to tie recommendations to ratings, partly because comparatively few people use ratings, so the data is slim, and partly because those who do usually only rate the high-end books, so data on the low-end is slim.

That's a feature that lots of us would like, but at the moment the powers-that-be are very resistant to the idea.

As a work-around, you can set up a special collection for recommendations. It might sound like a lot of work, but it doesn't actually take very long. On the 'Your books' tab, click on 'your library', dropdown, then pick 'edit collections'. Add a new collection called something like 'recs'. Then tick 'include in recommendations' for this new collection, and untick it for 'your library'. Save those changes, then click on the 'poweredit' button with the lightning flash sign, go down the page checkboxing all the books that have 3.5 stars or over, and finally pick the 'change collections' tab and tick the 'recs' collection.

That might sound long, but will actually go very quickly. And then, as long as you remember to add any new books you rate highly to both collections, your recs whould only be based on the books you want.

#3
This is a much requested idea, but Tim has said that he does not want to tie recommendations to ratings, partly because comparatively few people use ratings, so the data is slim, and partly because those who do usually only rate the high-end books, so data on the low-end is slim.

Which of course is a self-fulfilling prophecy: people don't bother with ratings because they don't actually do anything.

I think it's more an ideological thing than a data quantity thing. The claim that there aren't enough ratings data has always been a pretty bad excuse, and as the site has grown more and more it's just gotten less and less convincing.

4> You'll also need to unclick Use For Recommendations on the Your Library collection (and all your other collections) for this to work.

#7You'll also need to unclick Use For Recommendations on the Your Library collection
I did say that!
(and all your other collections) for this to work
And I checked that the poster had only the one collection.

Sorry. I guess I was reading too fast.

Thanks a lot, that was very helpful.

There is one thing that I think EveleenM forgot, which is that it can take up to 48 hours for your recommendations to change after making those changes.

Which of course is a self-fulfilling prophecy: people don't bother with ratings because they don't actually do anything. (#5)

Exactly.

There is one thing that I think EveleenM forgot, which is that it can take up to 48 hours for your recommendations to change after making those changes. (#11)

How does a tape delay make it not a self-fulfilling prophecy? People don't use recommendations not because they would take so long to kick in, but because they don't kick in at all.

I think jjwilson was referring to her message 4 - just warning anyone who wants to do this one-collection-for-recommendations thing that they won't see instant results, so don't keep fiddling - let it alone and see what happens in 48 hours.

Right. It was supposed to be a follow-on to messages 8 and 9.

Ah. I got confused by the lack of designation. Without that context, it also makes sense applied as I thought it was.

I use ratings very haphazardly and only occasionally, but I don't care one way or another if the ratings "do" anything. I see them as my indication to someone else what I thought about something. If they did other than that, I think I'd be more confused than motivated, i.e., I doubt if I'd rate any more than I do now. Come to think of it, I think I'd feel pressured if I had to come up with a rating for everything. That would make rating just plain work. Not interested.

(Full disclosure: I rarely look at recommendations anyway, so I really don't have a pony in this horse race.)

I don't think that anyone is suggesting that people be pressured to rate everything.

Agreed: the point isn't to pressure people to rate everything (or even anything). The point is to make rating stuff less, well, pointless: people like seeing that their effort, however minimal, is actually worth something, even if that something is really modest tweak in what books get recommended.

The alternative to fiddling about with collections, and having to remember which collections have recommnedations on or off, and hence into which ones you can place a new book, is to tag them. The Tag based recommendations work well, and allow subsets with collections too. I add "use" as my tag for books I'd like to be included in recommendations calculations.

Thoroughly agree with the ratings should do stuff. Even if it's just a broad filter at my library level, rather than the interesting (but computationally heavy) across LT interactions.

Perhaps I have been misunderstood in post 16. I am disputing these assertions:

#5>Which of course is a self-fulfilling prophecy: people don't bother with ratings because they don't actually do anything.

#18> people like seeing that their effort, however minimal, is actually worth something, even if that something is really modest tweak in what books get recommended.

Some people (many people?) may care. Some people (many people?) really don't care. And that's why they don't do ratings. Maybe they like alpacas instead of ponies; I don't know. But I do dispute these statements as being true across the board.

I think these debates are premised on a false claim anyway. We've been told that people on the whole don't rate much, there aren't many ratings, etc. etc. I think this is based on a ridiculously optimistic idea of what it means to rate.

The Zeitgeist shows 8,259,285 ratings on 51,139,859 books, or about 16%. Since joining LT in 2006, I've rated every book that I've read, with the exception of maybe three that I didn't feel qualified to evaluate for whatever reason. And yet only about 14% of my books have ratings--lower than the overall number for the site--because there are many that I don't remember very well, many that I haven't read, and many (reference books) that I've used a bit but not enough to assign a rating. And I don't even have a wishlist, which would lower my rating percentage even more!

People do rate, and LT does have rating data. Not everyone rates, and not all books are rated, but that doesn't mean the data that we do have are insignificant or useless. Unless, of course, you have a preconceived notion about the value of ratings--which Tim, unfortunately, does.

20> I think the thing you have to keep in mind is when someone says "people" or "most people" or whatever, they're giving their opinions on majorities. They're not saying it's a rule and you'll never find an exception. I don't think most users think in terms of "everyone will always be this way, no exceptions."

21> I agree with this, but I admit my experience is limited an anecdotal. The only times I find that amazon and LT ratings are very divergent is on books with limited numbers of reviews. In those cases, I don't really pay attention to the rating anyway due to the small sample size. And yes, this doesn't answer the question of "but are LT ratings an accurate gauge of what any random person would rate a book?" Nonetheless, I find amazon ratings incredibly helpful and useful in a number of situations. LT even has the possibility to be better than amazon in this sense in that a) you can rate something on LT without reviewing it (I'd do this far more often) and b) you could bring in some coding kung fu to give more weight to ratings from people who seem to have similar tastes based on your libraries.

...you could bring in some coding kung fu to give more weight to ratings from people who seem to have similar tastes based on your libraries. (#22)

Which is how I think using-ratings-for-recs should work, and how I've thought they should work for a while now. The recommendations wouldn't ignore similarity of libraries, it would just add similarity of ratings into the mix.

23> Agreed. But I can also see the attraction in doing it the simple way, even if only at first. The coding kung fu for such a thing would not be simple.

I don't claim that it is simple, but I think LT is well past the "if only at first" stage.

This is a much requested idea, but Tim has said that he does not want to tie recommendations to ratings, partly because comparatively few people use ratings, so the data is slim,

What's that, Yossarian?

and partly because those who do usually only rate the high-end books, so data on the low-end is slim.

Obviously: people only buy books they expect to like, and most of the time they're right...and some people even get rid of books they hated...

...though some of us keep them in our LT libraries for the sole purpose of warning ourselves (and anyone else who cares) that these are awful (up to uninteresting) books. One of my primary purposes in putting my books on LT was to rate them (and now I review as well, so I can remember why I rated them that) just so I wouldn't keep buying ones that had interesting back blurbs and really bad writing!

I have a collection called Discarded where I put all the books that I no longer own - whether I liked them or not. It's useful to know what's passed through my hands, anyway.

So, there are two issues here:

1. Should LibraryThing recommend books to members based on books they gave low ratings to?

2. Should LibraryThing book-to-book recommendations take into account the ratings members gave it?

The former is easy to do. The engine can stop making recommendations on low-rated items in your catalog. This would make ratings do something. It would make them kill off recommendations.

The latter is hard to do and also in my humble and not ignorant opinion add little to the quality of recommendations overall.

If the former is easy to do, it is probably worth it. It saves members getting aggravated by recommendations based on books they hated.

Yeah, that's my feeling. I'll just make a setting so you can say "Recommend based on books below X stars."

30> I hope you mean ABOVE X stars :)

Damn. That would have been fun.

Do both - one on the Unsuggester page.

#30. PLEASE DO THIS! It's been hugely requested many times!

#28"This would make ratings do something. It would make them kill off recommendations"

Killing off recommendations that people don't want is a good thing!

Personally - I rate for myself, so I can remember what I thought of a book (and I have to list how I decided to rate on my profile so I can correct myself when I drift). I don't look at recommendations because they're full of books I already know about and either have read and don't want to read again or have looked at and decided not to read. Member recommendations are a trifle better but not much. If you make it only notice my high-rated books, I'm likely to try to use recommendations a bit more - if only to play with it - and also to rate more books (about a third of mine are rated, 1,500 of 4000 (or so)).

I agree with you about the second. Just for the processing power, it doesn't seem worth it - and I don't think it would make useful changes to the book-to-book recommendations. But a recommend-by-rating that's similar to recommend-by-tag (except it's X number and above, not only exactly X number) should be simple to set up and at least mildly interesting.

>30 timspalding:

Oohhh. I'd like that. It bothers me a bit to think that books that have a low rating (only one or two) are even in a minor way contributing to my recommendations. I haven't been too worried about it though given my relative lack of low scores.

Of course, I have to admit like the others I almost never use the recommendations. I've been thinking of splitting my graphic novel/manga reading into another category and removing it from recommendations.

Which is too bad, because that's the one collection that I have occasionally seen a book that I've seen a review on here and track it down. But it assive drowns out some of my other area of interests. (They're quick to read and a lot of people read them. Books on beer brewing, not so much.)

Now that I'm thinking about it, I should probably just find some uncommon tags that seem to get applied with high frequency to books in those categories and watch the feed.

It would also be nice if higher ranked books of my own somehow influenced the recommendation algorithm, but I'll take what I can get ;). (So if I score a bunch of collections of horror comics really high and a the same number of superhero comics at an average rating it is more likely to recommend a new horror comic.

>30 timspalding: I think this is worth doing because it will make a lot of people happy. It would be really nice if we could switch quickly from one recommendation list to the other, like we can for tags, rather than waiting 24-48 hours for it to recalculate.

In a similar vein, I'd really like to see recommendations based only on date: here are recommendations based on books you've read in the last three years, last six months, etc.

LT is now running four flavors of recommendations: LT Recommendations, Read-Alikes, Member Recommendations, and the Unsuggestions. A Rating recommender wouldn't hurt anything by being added to the mix. Seven or eight flavors of recs would be even better. Of course I'm on record saying recommendations are only going to become more important as time goes on and media production increases. For every hour a person lives thousands of hours of media are being/will be produced. Finding 'your' media in the media haystack is a complicated but needed service.

Book-to-book recommendations deserve some attention too. For one thing, it would be nice to get some meaningful Special Sauce recommendations for prolific authors. Recommending 24 other books by the same author is just laughably bad.

And then there is List recommendations. Wait, we need Lists first.

The beautiful thing about Lists is that they can be recommendations on their own, with no further work required. Except the initial implementation, of course.

>28 timspalding:
Not doing option 2 would also avoid the dangers of "rating spam" where author/publisher accounts give themselves a high rating! (not that they don't have a right to think the world of their work *grin*)
But option 1 would be a definite plus I think.

Ah, but you could take your list and run it through a recommendation engine and see what pops up. Or poll the collective lists, say Vietnam/US War, and find out which books appear on the most lists or (horror!) which ones have the highest ratings. I think there is an active thread right now on Vietnam fiction recommendations. Somebody is always looking for something.

41> I think that's not as much of a point, considering how many mechanisms Tim has made to nuke spammers. Plus, if the recommendations took into account how similar your libraries and ratings of shared books were (yes, I know this was talked about as an advanced topic), it'd be even more likely to make rating spam not be worthwhile.

35: Tag your books on beer brewing, and then look at the recommendations for that tag.

Yeah, that's my feeling. I'll just make a setting so you can say "Recommend based on books below X stars." (#30)

30> I hope you mean ABOVE X stars :) (#31)

Yes, as long as it's above X stars. But more importantly, as long as it's filterable.

I should change my list:

1. Should LibraryThing recommend books to members based on books they gave low ratings to?

2. Should LibraryThing book-to-book recommendations take into account the ratings members gave it?

3. Should recommendations downplay (ie., move lower or remove) recommending books that have low ratings.

#3 is also doable and, to a limited extent, done.

Here's the long explanation.

It makes intuitive sense that ratings would matter. After all, you don't want more books that are "like" a bad book, right?

But "books like a book" isn't actually what's going on. LibraryThing doesn't know what's like anything else; it's just a shorthand way of talking about it. Rather, except for tags, the recommendations algorithms employ statistics of coincident libraries. That makes a big difference.

Take a book like Citizen Girl, the second novel by Emma McLaughlin, the author of the Nanny Diaries. It's some sort of chick lit about a young woman making her way as a fresh-faced political intern--so, chick lit with a bent toward professional situations.

The book is one of LibrayThing's lowest-rated, especially for a book with hundreds of members. Apparently it's really, really bad. (Reviews of all sorts bear this out.)

Looking at the raw data, it's clear the people who bought Citizen Girl were huge fans of the Nanny Diaries. They also bought her later books at wildly disproportionate rates. So there's every reason to beleive that's why they got the book. The book also matches up well against popular examples of chick lit, with a tilt toward "your professional learning the ropes in an interesting industry" stuff, like The Devil Wears Prada and The Second Assistant: A Tale from the Bottom of the Hollywood Ladder. A rational observer would conclude the people who got Citizen Girl like this sub-genre of chick lit particularly.

So, what does Citizen Girl being bad do for any of this? Do people who thought it was bad like other Emma McLaughlin books less? I doubt it. They seem to like them much more, including the stuff that came after Citizen Girl. Can we conclude they shouldn't get more examples of popular and professionally-oriented chick lit? That seems pretty unlikely too. It seems much more likely they like the topic, but didn't like this example of it. Most importantly it's very possible people read Citizen Girl in spite of the bad reviews--which were immediate and devastating--because they like Emma McLaughlin and professionally-oriented chick lit. That is, the ratings say the exact opposite of what people assume they do.

This phenomenon is quite frequent. When I look at my own books, the ones I've rated poorly are often the best predictors of what I want. I got them because they looked good. That they turned out to be terrible does not moot that fact. Indeed, they are a testament to my willingness to buy and read books on the topics I'm interested in.

So, even if taking ratings into account didn't cut the statistics down radically (eg., from 693 to 97 books for Citizen Girl), there's no reason to believe the effect would be positive. You are what you wanted and have as much as what you think of it.

Now, why are movies different? Because there are a lot fewer movies and they get seen a lot more. We see them more "blind" than we buy books--we know less and make fewer predictions about quality from knowing something about the (far more) people involved in making it. And we tend to see them with other people. The other people influence whether we see the movie, both directly and by providing the social prompt to do it. My wife and I agree on some things, but we also alternate picks, so I end up seeing a fair number of films I wouldn't choose to go to if I lived on a desert island with a movie theater.

Because movie-choices are more constrained and arbitrary, the personality test is generally thought not to be what movies we see, but what we think of them. Did we love Die Hard, Napoleon Dynamite and Marlie and Me or not? Even so, having talked to people who've worked on the Netflix Prize, and ranked very highly on it, statistics of having-rated-it-at-all are much more powerful and predicative than what the ratings actually were. Even with movies, the movies we see tell more about what other movies we'd want to see than what we think about them.

I don't want to be insulting, but ultimately, none of this will ever convince the vast majority of people. The horse-sense argument--why would LibraryThing recommendations not use ratings?--is too strong. The explanations, correct as they are, can't knock it down.

46> I don't want to be insulting, but ultimately, none of this will ever convince the vast majority of people. The horse-sense argument--why would LibraryThing recommendations not use ratings?--is too strong. The explanations, correct as they are, can't knock it down.

Honestly, Tim, it's not insulting. But my argument isn't a "horse-sense argument." It's based on experience. Did you ever work with the netflix prize at all? In that, netflix gave out a large dataset consisting of userids, movieids and the rating the user gave each movie and the date on which the rating was given. You then predict what rating a given user in that list will give a movie in that pool that they have yet to rate.

Now, there's a whole other line of argument on which was produces "better" recommendations, or if it's possible to meld the two. But it's not as simple as a division between "people who know what they're talking about wouldn't use ratings" and "people who don't know what they're talking about think using ratings would help."

One thing I do know is there have been a lot of people reporting greatly improved ratings when they rig up a "use for recommendations" collection that they only stuff with their highest rated books. That's simple experience and I don't think it should be discounted.

But with that said - ultimately, recommendations is one of the least important parts of LT to me. I already have a stack of books I own (125 at present) that I have yet to read. I'd much rather see the site polished, features fully implemented (touchstones, secondary authors) or annoying loopholes fixed (e.g. unique ids for authors so they dont get combined because one is named Paul A. Smith and the other is Paula Smith), etc. than I would you slaving for weeks over a recommendation system.

ETA: I swear. Did you just add that paragraph about the netflix prize as I was typing this or am I that blind?

Should LibraryThing recommend books to members based on books they gave low ratings to?

LibraryThing should give users a choice if it's not too difficult. The cut-off option that you proposed above seems popular. Option is the key word.

Should LibraryThing book-to-book recommendations take into account the ratings members gave it? Should recommendations downplay (ie., move lower or remove) recommending books that have low ratings.

There are far more important improvements you can make. Briefly, you just need to show more book-to-book recommendations. You've said in the past that the lists are short because you don't have enough data to make them longer. But if downplaying low-rated books would mean that other books would take their place, why can't we see those other books right now?

As I said earlier in this thread, Special Sauce recommendations for prolific authors are especially pathetic. We need an option to exclude books by the same author.

I also think I'd be much more likely to use LT recommendations if they showed cover pictures, but that's a side issue here.

Now, why are movies different? Because there are a lot fewer movies and they get seen a lot more. We see them more "blind" than we buy books. And we tend to see them with other people. The other people influence whether we see the movie, both directly and by providing the social prompt to do it. My wife and I agree on some things, but we also alternate picks, so I end up seeing a fair number of films I wouldn't choose to go to if I lived on a desert island with a movie theater.

I don't think movies are as different from books as you claim.

First, what do you mean by "they get seen more"? If this is just an issue of data quantity, maybe you just have to exclude books that don't meet a certain rating threshold. This happens already; some books have too few copies to generate recommendations. There are certainly plenty of hugely popular books that have been read by many people.

If you mean people see more movies than they read books, I have to flat-out disagree, especially on a book site like this. I can count on the fingers of one hand the movies I've seen this year (well, if I could remember them--I can only actually think of two), while I've read 30 books. Looking at the box office top ten on IMDB, I've seen none of them, though one has been strongly recommended and I do hope to watch it eventually. Looking at LT's Popular This Month list, I've read two and have another in the TBR pile.

Other people's input is important in books too. Look at the TIOLI challenge in the 75 Book Challenge group, where the whole point is to increase interaction by reading the same books as others. And of course family influences have a lot of weight too. I'll pick up books that don't seem particularly appealing to me because people I trust have said they're great. I'll even read books that are just generally popular because I want to be able to participate in the discussion. Basically, books are social too.

statistics of having-rated-it-at-all are much more powerful and predicative than what the ratings actually were. Even with movies, the movies we see tell more about what other movies we'd want to see than what we think about them.

This is ultimately where I think your whole argument comes to pieces and you reveal that what's most important is just your personal preferences, not what creates the best recommendations.

I believe you that the movies we've seen are most telling, and I think it's likely that the same holds for books.

And I can find no way to reconcile that with your stubborn refusal to add a default Read collection and start generating these great recommendations that you keep telling us about.

48> To play devil's advocate on the topic of "they get seen more": I read on average roughly a book a week. I see on average a film every three months. However, how many books are released in a year versus how many movies? I think this has some relevance on the "they get seen more" if he's talking about how many people will review each book out of the entire pool.

ETA: Yes, Tim, that means you are the devil! >:)

#46
The horse-sense argument--why would LibraryThing recommendations not use ratings?--is too strong. The explanations, correct as they are, can't knock it down.

But your 'correct' explanations are for a collective group of people, to which individual members don't necessarily belong. In my case, two facts:
-In the past, I got consistently excellent recommendations from a ratings-based algorithm;
-When I set up a ratings-based subset here for recommendations, they went from drearily off-putting to reasonably interesting.
That's not just an intuitive feeling, it's clearcut experience.

>44 infiniteletters:

Slightly embarrassed that somehow I never have seen the functionality to filter by tag in the recommendation page before. Probably just because I saw the "999 reviews" and my eyes crossed)

That there is the single best argument I've seen for me to actually stop being lazy and start tagging my books.

Now if I just had some more free time to do so ;).

>49 brightcopy: Sure, but how many of those books really get noticed? It's only a small fraction that get prominent display space in the bookstores, show up on the bestseller lists, etc. This is why I suggested excluding the books that don't have enough data; I think there's still plenty to work with among the more popular books.

52> Understandable, but I don't think this makes up for the gap. Just using the US in 2008 as an example:
610 films were released
275,232 new titles and editions were published

Even with niggling data questions about books that are just reprints of older books, films that are re-releases, books that no one read, films that no one saw, etc. etc. etc., that's a pretty gigantic order of magnitude difference.

That's not just an intuitive feeling, it's clearcut experience

Well, it's an anecdotal finding that supports phenomena #1, that you don't want to see recommendations for books you gave low ratings to. That's not quite the same as #2. Also, while it may seem the recommendations went from boring to interesting the only thing that could have happened was the removal of some you find boring--leaving the remainder on average more interesting. Except for the new beta recommendations feature, member recommendations are just aggregated book-to-book recommendations.

>53 brightcopy: So the question is what percentage of our books read were from the top 600 published that year. Any idea of what that would mean in terms of LT ownership numbers?

Taking an extremely rough estimate, where at least 1000 copies means it was in the top 600 for its year (a book with 954 copies is #4752 on the site, so we're looking at less than ten years of books): Of the last hundred books I read, almost exactly a third have more than 1000 copies. Since I consume ten times as many books as movies, this means that the number of "popular" books I read actually exceeds the total number of movies I watch. The fact that I read other, obscure books as well doesn't negate the value of the data that can be gained about popular books, nor does the existence of thousands of other books that are largely ignored.

Of course my experience may not be representative, but I don't think it's way outside of the norm either.

54> Also, while it may seem the recommendations went from boring to interesting the only thing that could have happened was the removal of some you find boring--leaving the remainder on average more interesting.

Isn't this just semantics? Isn't the value in recommendations the amount of times, on average, they are interesting to you?

55> I'm afraid you lost me a bit on some of your assumptions. Why did you pick the top 600 books? Was that to somehow match the 600-ish films? Wouldn't it be more fair to compare to the top X films?

>56 brightcopy: Yeah, I was trying to match the 600 films. The theory was that if books have a longer tail, we could just cut it off and then compare equal quantities. I really don't have the data to do that, but I think books with 1000 copies can fairly be assumed to be as popular as the least popular movie ever made.

46> Well, this iteration of your constant argument that ratings don't help the way people think they do - actually finally made sense to me. It's true, I bought a book because I liked it (or thought I would like it) - if it was then badly written and therefore I rated it poorly, that doesn't mean I wouldn't like other books in the same groups (genre, subject, author).

I'd still like the cut-off for recommendations, just to see. And I would like if recommendations based on my highly-rated books came out higher than recommendations based (entirely or strongly) on my low-rated books - I don't have any subjects where all my books are bad and low-rated but I'm still interested (I don't think). Unrated in the middle? Or unrated at the bottom, to encourage people to rate more! It would work on me, but would probably drive people who rate very little nuts. Those who don't rate at all wouldn't get this sorting and so wouldn't notice (everything would be in the Unrated bucket and sorted therein), but if someone has rated 10 of their 400 books and gets recommendations based on those 10 at the top every time, they might get annoyed and complain about bugs. "Recommendations is based on only a few of my books, all the time!" Still, maybe with some explanatory text on the Recommendations page (yeah, yeah, nobody reads that...)?

#54
Also, while it may seem the recommendations went from boring to interesting the only thing that could have happened was the removal of some you find boring--leaving the remainder on average more interesting.

Okay, the more interesting recommendations were somewhere on the list already, but a numbered list where the interesting recommendations are at #700 or #1000 is pretty useless for most purposes. Subtracting hundreds of the duller recommendations may seem trivial to you, but it made the difference between a feature I'd write off altogether and one that I'd actually use.

46> I haven't read the intervening posts so my apologies if these points have already been raised.

First, I think it's just intuitive that if you have ratings and recommendations that the ratings will be used with the recommendations. If that isn't the case you need to put a big red disclaimer on the recommendation page that the ratings don't affect them. In other words, to not use ratings with recommendations is counter-intuitive and the best UI is an intuitive one.

My second point is responding to this part of Tim's post:

This phenomenon is quite frequent. When I look at my own books, the ones I've rated poorly are often the best predictors of what I want. I got them because they looked good. That they turned out to be terrible does not moot that fact. Indeed, they are a testament to my willingness to buy and read books on the topics I'm interested in.

That may be true for you, but other people may pick up Citizen Girl and hate it and not want to be recommended anything else by that author or genre. In fact there seem to be a lot of people in the latter camp judging by the number of times the subject of ratings and recommendations comes up. So I think it makes sense to at least give people the option of not including their low-rated books in their member recommendations.

EveleenM said: That's not just an intuitive feeling, it's clearcut experience
Tim said: Well, it's an anecdotal finding that supports phenomena #1

Can't help but point out that your post #46 is pretty riddled with anecdotes and personal opinions based on your own experiences. That's not to say you're wrong, though. It's just to say that unless you come up with actual sample data, make some predictions, and then check those against the outcome of reality, people are going to keep arguing with you about what would work best based on their own anecdotes. And now we're back to that opinion warfare thing. Lucky for you, as the LT owner you've got the biggest gun. :D

ETA: Corrected to say EveleenM not Zoe.

That may be true for you, but other people may pick up Citizen Girl and hate it and not want to be recommended anything else by that author or genre.

If that were true, one would expect to see that author and genre show up less for people who have the book. But, in fact, we know the answer to the question and it's the opposite of what you assume--the statistics show that people who have Citizen Girl are overwhelmingly likely to have other books by the author and in the sub-genre. One of the all-time most hated books on LibraryThing is correlated with more books by the author and in the genre.

62> To me, I'm more focused on recommendations for books by other authors. Something like "If you liked Starship Troopers by Robert Heinlein, you'll probably like Armor by John Steakley". To me, recommendations within the same author aren't nearly as useful, as I can already look for other books by the same author on my own.

ETA: An anecdote related to touchstones re: comments about their reliability on another thread. I put two book touchstones and two author touchstones in that previous paragraph. 100% success rate on the book touchstones, 100% failure rate on the author ones.

62> Ah. But I bet those people were fans of that genre or author first and picked up that book rather late in their collecting, whereas people who read that book first may not want to see anything by that author again. And they would be rather dismayed at finding their recommendations peppered with books they don't want.

Is this a fan vs. casual reader thing?

ETA: Actually, wouldn't the fan already have the other books so the fact that they have one low-rated book in their library of that ilk wouldn't affect their ratings very much? But the person who has that one book by that author/sub-genre and rated it poorly probably doesn't want any more.

>62 timspalding:

so people with Citizen Girl and other "similar" books, are likely to want others - fine, but in that case leaving "Citizen Girl" out might make very little difference, as the other higher-rated similar books would pull in similar recommendations anyway. While those who only list "Citizen Girl" from the author/genre and rate it low could perhaps either be recommended nothing like it at all, or possibly "You did not like Citizen Girl.... would you like to look at similar but higher-rated books"

>61 brightcopy: I feel obligated to point out that I wasn't actually the one who said that, though it doesn't really matter.

I'm still waiting for Tim to follow his argument through to its logical conclusion re. the books people have actually read.

>62 timspalding:
Here's a tangent: can we, say, determine how other users with similar libraries rate Citizen Girl, and then compare our rating to that?

I'm just throwing this out for the sake of argument, not necessarily as something to put into practice. Not even as something that should go anywhere. So don't beat me over the head with it if you disagree.

That may be true for you, but other people may pick up Citizen Girl and hate it and not want to be recommended anything else by that author or genre. In fact there seem to be a lot of people in the latter camp judging by the number of times the subject of ratings and recommendations comes up. So I think it makes sense to at least give people the option of not including their low-rated books in their member recommendations.

I think the crux of Tim's point is that what people say and what they do diverge substantially. Hence, people saying they want low-rated books excluded is not reflective of their actual behavior.

>66 _Zoe_:
Maybe that's another way to go about things: instead of deciding which books to leave out and then confining them to a specific collection (and, effectively, only that collection), why not create an setting that flips the default*?

------
* That is, so that you add books you do want to use to a special-purpose collection (in addition to any status/location/subject/whatever collections).

68> I think the crux of Tim's point is that what people say and what they do diverge substantially. Hence, people saying they want low-rated books excluded is not reflective of their actual behavior.

But I think his test is faulty. He's using what books are in a persons catalog to judge what they want. That is Citizen Girl is in their catalog and all those other books are too, therefore if you have Citizen Girl then you must want those other books. But what if the order is really that they got all those other books first and then got Citizen Girl to complete their collection. In that case CG has no predictive value at all.

That's why I mentioned it was her second book, and subsequent books were highly matched.

Acquisition order is something we consider.

Acquisition order is something we consider.

What exactly do you mean by this?

71> Publication order can be a poor indication of reading order. Order of entering into LT can be a poor indication of reading order. Acquisition date can be a poor indication of reading order.

I'm sure you already realize this, but it's important to point out.

I'd be surprised if you had a large enough base of data for people actually filling out reading dates, but who knows, maybe you do. But if not, it's just beginning to seem very anecdotal. Which, as I said before has it's place, but it isn't a good way to make generalizations.

Let me just say that if my wife gave me Citizen Girl to read and I hated it, but I duly entered into my library as a book I read. It would perturb me if I then started getting recommendations based on that book if I had rated it with a half star.

So I am a different case then the person who loves that author in which case having that book would be predictive. But for me it isn't. So it seems that your theory works well in the aggregate but not for everyone. So doesn't it make sense to give the individual the option of whether to include books with low ratings in their personal recommendations?

Nanny Diaries has an average LT rating of 3.25. Not so good. The rating of ND is likely making a difference in LT entered copies of Citizen Girl. 4,874 people have ND entered on LT. 609 have CG, or 12.5% of the number that have ND. LT is recommending CG to users that have ND. At most, 12.5% have taken LT's rec to heart and entered CG. 87.5% of those getting the recommendation haven't yet been convinced to add CG. Why? Ratings probably play a part. Some of those that get the rec for CG click on the work page, see the rating of 2.37, and decide, nope, not for me. The ratings of CG have influenced how many copies of CG have been added to LT. In this case, a large majority have ignored the LT rec for CG based on ownership of or reading of ND. LT is making the rec but not convincing readers it's a good rec.

If I'm a hypothetical ND reader.

I've bought the first book, read it, and enjoyed it, rated it 5* I'm now looking for recommendations of similar works to enjoy.

SHould CG appear on that list?

Overwhelmingly by Tim's statement in #46 "The book is one of LibrayThing's lowest-rated, especially for a book with hundreds of members. Apparently it's really, really bad. (Reviews of all sorts bear this out.)
" It shouldn't. Now LT probably can't parse reviews for positivity, but it does have ratings to judge this by. Currently they are ignored. This would seem to be a case where they shouldn't be.

Now if I only have CG in my catalogue, and I liek the majority hated it, should I get recommendations based on it. NO

I have lots of other chicklit also in my catalogue whic I have rated positively, then use THOSE and as the basis for other recommendations.

#46"Most importantly it's very possible people read Citizen Girl in spite of the bad reviews--which were immediate and devastating--because they like Emma McLaughlin and professionally-oriented chick lit. That is, the ratings say the exact opposite of what people assume they do.
"

But there are all sorts of confounding factors. Maybe they didn't check out the reviews. Had they done so they wouldn't have read it. I seldom read reviews before buying a book. But I do sometimes check out the ratings. Maybe they thought the reviews were from non-chicklit fans. Maybe they saw CG on their recommendations list and read it assuming that LT was using ratings and hence it must be good. Maybe they read it, expected it to be good and then hated it. Shouldn't recommendations be trying to avoid this last scenario?

It's a resolution issue: Recommendations aren't about genre. We can all find thousands of books in our preferred genre. It about specific books. Is THIS book - from an author I normally like, in a genre I normally like - worth my while to read. Ratings are a piece of information informing this decision.

76> Agreed.

Recommendations shouldn't just be about predicting the next book you will read, they should be about predicting a book that you'll likely enjoy if you read it.

> 76

Also agreed. I think a rec system which takes ratings into account should not only take /your/ ratings into account, but also the fact that there are 50 other ratings that put this book below a 2.5.

That is, if there are a substantial number of ratings on a book, and that book still has a significantly low rating (and yes, I would set these numbers very high and very low, respectively, to avoid knocking out books which are simply controversial, not bad), then keep it off the rec list!

I was thinking about a simple android application for librarything, so it would be possible to take a look at my (or someone elses) library-thing, anywhere and anytime, which could be handy in bookstores for example when you've forgotten which books you've got already or whether someone has some great recommodations.

You can already do that with the LibraryThing mobile site.

>80 jjwilson61:
Assuming you have a mobile way to access LT mobile. (The nearest I can muster at present is a laptop -- and it's a bit large to carry around to bookstores.)

81> I suppose a netbook would work out well for this.

>82 brightcopy:
I think LT mobile is for cell phones: the point was that the nearest thing I have to a cell phone, for practical mobility purposes, is a laptop computer.

83> I was responding to markiske who was talking about an Android application which I believe is also for cell phones.

84> My android cellphone seems to work with LT mobile, thanx for mentioning that option :)