Goodreads Feedback discussion

150 views
Suggestions & Questions > Similar tastes algorithm :-)

Comments Showing 1-38 of 38 (38 new)    post a comment »
dateDown arrow    newest »

message 1: by Seth (last edited Aug 25, 2016 01:03PM) (new)

Seth | 78 comments Go to your own profile and click "compare books."

Try it.

Apparantly, my tastes are 98% similar as mine own, perhaps because some of my ratings are slightly different from my ratings, not that I know why, but I canc learly see it in the star comparison charts. :-)

I noticed that the comparison report only showed some number of books (484?) starting with 5-starred alphabetically and working down the ratings. I wonder if that's how the rating is calculated. I'd expect that books rated a 1 would be at least as good a comparator as 5s, if not better, but I'm probably overcomplicating things.

But being 98% in agreement with myself is great. Actually, it's probably a high point in my life :-)

(I'd go for full credit for a 5-5, partial for a 5-4, full for a 3-3, partial for 3-4 or 3-3, full for a 1-1, and partial for a 1-2, but that's definitely overcomplicated)


message 2: by rivka, librarian moderator (last edited Aug 25, 2016 01:03PM) (new)

rivka | 12269 comments Mod
I have 402 books in common with myself, and a 99% taste-in-common.

I could swear I used to agree with myself more, but perhaps that's just a sign of my aging memory. Which would go with increased crotchetiness, so that rather makes sense.


message 3: by Rindis (last edited Aug 25, 2016 01:03PM) (new)

Rindis | 47 comments I'm such a sheep. I agree 100% with myself.

Um, Sherri, what do you want? To give him a nose job?


message 4: by Lisa (last edited Aug 25, 2016 01:03PM) (new)

Lisa Vegan (LisaVegan) | 8939 comments I also agree 100% with myself, for 400 books. Not sure about the other 588 books though. ;-)

Suspect when it's not 100% agreement, you've rated at least one book twice, giving it different ratings. A great way to check for duplicates I guess.


message 5: by rivka, librarian moderator (last edited Aug 25, 2016 01:03PM) (new)

rivka | 12269 comments Mod
There are 72 editions combined (authored by Rostand -- I fixed a few alternate spellings of his name). That doesn't prevent you from having given two editions separate/different ratings, though.

Lisa, I'm pretty sure I have no duplicates at the moment. And a few weeks ago, I agreed with myself 100%. I'm wondering if some of the recent fixes of other issues might have had inadvertent effects on the compare books function.


message 6: by rivka, librarian moderator (last edited Aug 25, 2016 01:03PM) (new)

rivka | 12269 comments Mod
Ditto, to be sure.


message 7: by Lisa (last edited Aug 25, 2016 01:03PM) (new)

Lisa Vegan (LisaVegan) | 8939 comments Rivka, that is weird. Rivka and Sherri and Seth, now I’m curious; I’m going to keep comparing books with myself to see if I ever don’t agree with myself 100%.


message 8: by rivka, librarian moderator (last edited Aug 25, 2016 01:03PM) (new)

rivka | 12269 comments Mod
Everyone has a disagreeable day occasionally. ;)


message 9: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:04PM) (new)

Otis Chandler | 4184 comments Mod
Hehe. When you don't agree with yourself 100% I think it means you've got duplicate reviews of combined editions in your shelf.

We're actually going to optimize that page very soon. Any (other) suggestions for improvement?


message 10: by Femmy (last edited Aug 25, 2016 01:04PM) (new)

Femmy | 118 comments Hi, Otis. Do you plan to incorporate the suggestion mentioned here (http://www.goodreads.com/topic/show_g... )?


message 11: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:05PM) (new)

Otis Chandler | 4184 comments Mod
Hi Femmy,

That suggestion is definitely a good one, but is really a new feature not an add-on to the compare page. In order to do what he suggests, we'd have to normalize each person's library against every other persons library, so we could know who is the most similar to you and then give you percentages for people in between. However a way to approximate that might be to just assume that someone with all your books is the max, and to give percentages off that. So we're going to add that - let us know what you think.

Sorry if this made no sense - too much coffee today!


message 12: by rivka, librarian moderator (last edited Aug 25, 2016 01:08PM) (new)

rivka | 12269 comments Mod
:D

I just compared myself again. Other than the fact that there is no such thing as 101% of my library (which I realize is an artifact of setting an arbitrary 100% line), and the fact that I still only agree with myself 99%, I did have a giggle at this addition:

* note: You are comparing yourself to yourself. If you don't see 100% similar rating, its likely because you have rated duplicate editions of a book.

Except I still don't see any duplicates on my shelves . . .

[Edit: So much for that. I found two, but I'm still at 99%.]


message 13: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:08PM) (new)

Otis Chandler | 4184 comments Mod
Rivka,

You actually have 4 duplicates. Since you might want to know what they are, we'll add a new feature to the 'edit view' tomorrow to show duplicates.


message 14: by rivka, librarian moderator (last edited Aug 25, 2016 01:08PM) (new)

rivka | 12269 comments Mod
Yay! (After that first sentence, I was ready to call you a tease. But I can wait a day. ;) Especially for such a cool new feature.)


message 15: by Lisa (last edited Aug 25, 2016 01:10PM) (new)

Lisa Vegan (LisaVegan) | 8939 comments Yay! All my hard work paid off but still a great feature in the bookshelves edit view; I'll check periodically-much easier than looking at my entire book list. My find duplicates results:

my duplicates

This page shows duplicates from your library. Duplicates take into account the combining of different editions of books, which is largely performed by members with librarian status.

You have 0 duplicates in your shelves.

Thanks for this feature Otis & Goodreads.


message 16: by rivka, librarian moderator (last edited Aug 25, 2016 01:10PM) (new)

rivka | 12269 comments Mod
That is SO COOL!

Would I appear horribly ungrateful (I'm not! Really! This is a wonderful new feature and I'm thrilled!) if I asked for one addition? An edit view (or just plain checkboxes) of the results, to make removing duplicates easier.

Even without that, this is exceedingly cool. THANK YOU!

AND *drumroll* I now agree with myself 100%! :D


message 17: by Marley (last edited Aug 25, 2016 01:10PM) (new)

Marley (MarleyMagaziner) | 18 comments yep, having an easy way to delete the extraneous entries would be awesome. But I'm looking forward to agreeing with myself!


message 18: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:10PM) (new)

Otis Chandler | 4184 comments Mod
Agreeing with oneself 100% should be everyone's goal :)


message 19: by rivka, librarian moderator (last edited Aug 25, 2016 01:21PM) (new)

rivka | 12269 comments Mod
Hmm. Why no more

* note: You are comparing yourself to yourself. If you don't see 100% similar rating, its likely because you have rated duplicate editions of a book.


?

:(


message 20: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:22PM) (new)

Otis Chandler | 4184 comments Mod
It only shows up if you aren't 100% in agreementment with yourself.


message 21: by rivka, librarian moderator (last edited Aug 25, 2016 01:23PM) (new)

rivka | 12269 comments Mod
Logical. :)


message 22: by Tiffany (last edited Aug 25, 2016 01:23PM) (new)

Tiffany | 244 comments The only thing that bothers me about the duplicates/not agreeing with yourself 100% is that different editions and different volumes are counted as duplicates. I have 3 versions of Plato's Republic on my shelves (lots of philosophy classes in college), and since they're different books (with different pros and cons to each one), I've rated each, and rated them differently, at that. However, it then comes up that I don't agree with myself since it's comparing the different versions as if they were the same (i.e. since I gave Republic 1 4 stars, but Republic 2 5 stars, it's saying there's a difference in how I rated the one Republic book.).

Also, when I Find Duplicates, it says that those 3 Republics are duplicates, but also a book that I have that has Volume 1 and Volume 2. While this may seem like the same book, or two parts to the same book, they're not the same book, physically. If I want to read Vol 1 and Vol 2, I have to buy two different books, so they're not really duplicates.

Not that I'm complaining, though (I love Goodreads too much to complain about anything!). Just pointing out an oddity.


message 23: by Evan (last edited Aug 25, 2016 01:25PM) (new)

Evan | 128 comments This sounds more like a problem with editions that shouldn't have been combined. Are these different translations or do they have different commentary? Personally I feel that both cases should be separated, but I'd have to reread the librarian guidelines to be sure...


message 24: by rivka, librarian moderator (last edited Aug 25, 2016 01:25PM) (new)

rivka | 12269 comments Mod
Actually, the first ones should be combined. Different editions/translations of the same book.

The second (different volumes) probably should not be.


message 25: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:27PM) (new)

Otis Chandler | 4184 comments Mod
Tiffany is right, that is an oddity that we will need to address at some point. Hope ya'll can live with it for now!


message 26: by rivka, librarian moderator (last edited Aug 25, 2016 01:27PM) (new)

rivka | 12269 comments Mod
I'm curious as to how you would address it. How could GoodReads differentiate between Tiffany -- who sees the different editions of Republic as different entities, each deserving its own rating, and me -- who only ends up with duplicates by accident?


message 27: by Otis, Chief Goodreader (last edited Aug 25, 2016 01:28PM) (new)

Otis Chandler | 4184 comments Mod
It would involve some serious brainstorming :)

I think the solution would be something like allowing sub-groups when combining books, to allow those who care about the different editions to specify so in a better manner. But thinking that through fully would just hurt my head right now, so I'm going to stop :)


message 28: by Alex (new)

Alex (little_Alex) | 3 comments rivka wrote: The second (different volumes) probably should not be."

I searched under "volumes" and this is the only thread that turned up any results, so I hope this is the right thread...

So the current rule is that if a book has been separated into different volumes, it shouldn't be combined with another edition that has all the text? (e.g., Cyteen I, II, III and the Cyteen book).

I'm asking this because many Japanese novels, etc. are in two volumes and right now, for most of them, both volumes are dumped into the same book. (e.g., I've just separated a bunch of Kafka at the Shore V. 1, V. 2 from the same book but in one volume; would just like to know if I'm doing it right (or wrong))


(P.S. Would it be possible to add a note to the tips section on the combine editions page? :) )


message 29: by rivka, librarian moderator (new)

rivka | 12269 comments Mod
I searched under "volumes" and this is the only thread that turned up any results, so I hope this is the right thread...

That's because these questions are more commonly discussed in the Librarians group.


So the current rule is that if a book has been separated into different volumes, it shouldn't be combined with another edition that has all the text? (e.g., Cyteen I, II, III and the Cyteen book).

That is the consensus, yes. If you are finding books where that is not the case, it is either because they were combined before a consensus had been reached, or by a librarian who was unaware of the consensus. (Or, I suppose, by one who chose to ignore it. But that seems unlikely.)


message 30: by Lisa (new)

Lisa Vegan (LisaVegan) | 8939 comments I can't remember when, but post one of the GR improvements/changes, I noticed that the compare books often has very high percentage agreement with just about all members I check. I've rarely looked at the details the way Elizabeth did, but I find the result suspect. The compare books feature is still helpful though.


message 31: by Sandra (new)

Sandra | 1103 comments Yeah, I just compared myself and I'm .... 100%.




message 32: by Lisa (new)

Lisa Vegan (LisaVegan) | 8939 comments Sandra wrote: "Yeah, I just compared myself and I'm .... 100%.

"


Sandra, Well, that's how it should be. ;-)
If we compare ourselves with ourselves and we don't match 100%, it usually means we have duplicates (rated differently) on our shelves.


message 33: by Sandra (new)

Sandra | 1103 comments I know Lisa, I was just happy I didn't goof up and have duplicates! again!

I'm running at just a smidge under 2500 books and didn't want to have to figure it out.


message 34: by Lisa (last edited Jan 04, 2010 06:28AM) (new)

Lisa Vegan (LisaVegan) | 8939 comments Sandra wrote: "I know Lisa, I was just happy I didn't goof up and have duplicates! again!

I'm running at just a smidge under 2500 books and didn't want to have to figure it out."


Ah Sandra, Goodreads does it for us!!! Go to my books, then go to edit view, then click on find duplicates. If you have any duplicates, they will be listed on the page that shows up; if you have no duplicates, it will say so.

ETA: When I find duplicates, it's always meant some librarian has combined books that shouldn't be combined, but when I see that, I can correct it.



message 35: by Sandra (new)

Sandra | 1103 comments Lisa,
When I see duplicates it's because I've been over-enthusiastic and added the wrong edition, haven't noticed, then added the right edition later. Voila! Goof-up!

Although I do have a couple of different editions of the same books, but they have the same ratings. Basically the same books with different covers.

Glad to know the tip on duplicates though, for next time I goof-up.
:-)


message 36: by This Is Not The Michael You're Looking For (last edited Jan 04, 2010 10:12AM) (new)

This Is Not The Michael You're Looking For | 947 comments Elizabeth, at one time the algorithm was flawed in that it calculated the %similarity for a book as 1 minus the difference in the number of stars divided by 5. This was a flaw because a 1 star and 5 star rating, which are maximally different under the current system, would still be considered 20% similar, inflating the results. I thought they fixed this to divide by 4, meaning a 1 star and 5 star rating would be 0% similar, but maybe it never got fixed or was accidentally reverted at some point (there is an old thread about this somewhere).

Also, the discrepancy in library sizes is why the statement at the top of the compare has been made more explicit to emphasize what proportion of each person's library overlaps. The original statement included a lot less information and was a bit more misleading.

I would like to see unrated/to-read books completely left off of the list or sorted into a secondary comparison, to make the primary comparison seem a bit more meaningful, but my impression is that the compare feature is a bit of a CPU hog and they try to keep it a bit more minimalistic to avoid overloading the system.

I don't use the compare feature very often, but the results usually do seem inflated. I'm not sure how much of this might be due to a problem with the algorithm and how much is due to biases in the way people rate books and/or choose books to read. Since most users' average ratings are closer to 4 than 3 (I have no definitive data to support this statement, but I am predicting it is correct), there will be a tendency for the majority of books to be given a much narrower window of ratings than the 1-5 stars would suggest. Also, the vast majority of people self-select books they expect to like (not many people regularly go out of the way to read books they a priori expect to dislike), meaning that if their selection process is not completely flawed, most books that someone reads will be highly rated by them. The more books you have in common with someone, the more similar your tastes are likely to be and, therefore, the more similar your ratings (not just your reading choices) are likely to be.


This Is Not The Michael You're Looking For | 947 comments Update: I just did a comparison with a friend (actually an IRL cousin) who only has 21 books shelved. GR reports “You and xxx have 3 books (or 0.12% of your library and 14.29% of her library) in common. Your tastes for those 3 ratings are 75% similar. Here are 2 books you have in common:”

Of the two books it lists, one was rated 5 by her but only 2 by me. The other book was rated 4 by her and is unrated by me because it is a to-read.

There are a number of problems with what GR is reporting.

1. She only has 21 books on her shelves so I looked for the third book. It doesn't exist. We only have 2 books in common, so the fact that it reports that we have 3 is outright incorrect. The percents of the library’s it reports are “correct” if the 3 was correct, not the 2, so its calculating the percents correctly, but not the actual count which goes into the percents.

2. For that single book which we both rated, one of us rated it a 5 while the other rated it a 2. Yet GR claims that our ratings are 75% similar. By my understanding of the logic, the similarity should be 1 - (5-2)/4 = 1 - 3/4 = 25% similar (Even under the flawed older system it would only be 1 - (5-2)/5 = 1 - 3/5 = 40%). The GR report of 75% is clearly nonsense.

There are only two ways I can see to get 75%. The first (which I do not think is the problem) would be if they forgot to subtract the value from 1.

The second (which I suspect is the problem) is if is counting two more books as 100% identical. Two books at 100% plus one book at 25% average to 75%. I suspect the second book is the phantom 3rd book of the comparison (which does not actually exist), while the third is the book which only one of us rated. If it gives both of these a 100% similar rating, we'd find the 75% the system reports.

Assuming this is a systematic error, both of these (phantom books being counted as 100% similar and books with only 1 rating (or even no ratings?) being counted as 100% similar) would cause the upward bias we are seeing in similar ratings, since both would lead to a lot of extra 100% similars being added to the average value.

Either way, the compare feature is clearly flawed at this time.


message 38: by Otis, Chief Goodreader (new)

Otis Chandler | 4184 comments Mod
It looks like there somehow was a bug with the number of ratings you had in common with the other user. This should help things look better.

Good catch!


back to top