Howdy, Mek here -- I run https://openlibrary.org over at the non-profit Internet Archive (the folks that bring you the Wayback Machine).
Open Library is an free, online California Library with millions of digital books to read and borrow. It's additionally an open catalog of millions of books which you may track like goodreads.
Open Library has a very strong volunteer tribe of book-loving librarians, developers, and designers working to make the project better for our community.
We meet once a week at 11:30am PT. Ask me for the link: mek+ol@archive.org
RIGOROUS BOOK RECOMMENDATIONS:
We're also cultivating a 2nd non-profit open source experiment called TheBestBookOn.com (it leverages Open Library's catalog) which allows book-lovers to ask for or make rigorous book recommendations:
It's a very early prototype (we discuss development during our weekly Open Library call). It's being organized by Lauren Milliken, Aasif Khan, and myself and we'd love more contributors to join the discussion: mek+bbo@archive.org
The philosophy behind thebestbook.com (which is currently just a looks-like demo -- you should help us implement it!) follows this essay on Less Wrong:
Recommendations are better when you have enough info to draw a line rather than consider a single point.
There are many websites online which give you detailed reviews. But how does their review compare to other similar books? And has this reader also read those books?
Thebestbookon.com requires a reviewer to have read 3 books and justify choosing one of them as a winner (like a college basketball bracket).
I think there's a very important distinction to be made between the notion of an alternative and a replacement.
I think many of us (in the Open Library community) want to see many players in the space and as much creativity as possible. We love books and we love giving book lovers options. Inventaire, LibraryThing, Goodreads, and fivebooks are all amazing services which have pioneered different, amazing experiences. And each brings something unique to the table. Goodreads is loved by 90M people.
It's also true that it's challenging for any single application (e.g. Goodreads) to serve 90M diverse users in all the ways they may prefer.
We see opportunities for the community to be better included in helping cultivate a more inclusive and diverse book ecosystem. Will it be better than service X? I don't care as much, so long as it allows researchers, students, enthusiasts, parents, and other patrons (who may be left behind by current solutions) to achieve their goals.
In order to support this mission, our primary objective at Open Library is making sure useful book and author metadata is as accessible as possible via APIs, data dumps so the community can build the right solution for their use cases. And of course, doing the best we can to build basic, useful tools to help connect readers to their reading experience and to each other.
We're a very small team (staff of 2 eng -> 4M patrons) so without the community being excited and driving, what we're able to accomplish is very limited, beyond keeping the lights on and the book data happily flowing.
So I want to echo your sentiments olah_1. As a non-profit, our goal isn't to out-complete or replace any existing service. But it very much is to provide an "alternative".
An alternative where every netizen and book lover has the opportunity to shape the future of their library. Where there are books which represent everyone. Where everyone's use cases may be served (if people are willing to put in the work). And where the limit of what's possible is bound only by one's creativity and a shared vested interest in being a good neighbor.
And for good reason.. many of those requests involve a much higher level of maintenance (moderating content like reviews, building a social network, etc)
I like that OpenLibrary is trying to be a source of open book data. But I am a bit concerned about the quality of the data (just a feeling as I am still digging into it but there are open questions on how to handle things like multiple authors, etc)
We have a small network of librarians on a slack channel who have merged more than 10,000 duplicate works.
Librarians (which is a permission role) also have the ability to merge Authors, so data improves over time.
Furthermore, members of our community coordinate with Wikidata, VIAF, etc -- we match up identifiers and sync data (and in this way, bots are able to make meaningful changes in bulk on a rolling basis).
Holy cow, thank you for pointing this out and, well, building it in the first place. For me at least this is immediately looking better than Goodreads, which I've always kind of used begrudgingly.
Many in our community are intimately familiar with and sensitive to the challenges of FRBR (e.g. serials, etc) as it pertains to maintaining a system of Works and Editions.
This is an active conversation on our slack. Send me an email and I'll be happy to invite you to join the discussion.
Honestly, a lot of our Librarian bottlenecks aren't engineering, they're documentation and organization related.
Since Aaron Swartz started Open Library ~2007, there have been scores of trailblazing Librarians (like Karen Coyle and Jessamyn West) who have nudged us in the right direction and documented their opinions.
We could really use help here if you or anyone else is interested in doing some digital archaeology and helping us better organize (so we can make good schema decisions to move forward).
Thanks for bringing this up.
Fun fact, by the way: Nearly every page on Open Library is a json API of itself if you add `.json` to the end.
My problem with Goodreads isn't the recommendations engine. The recommendations are fine. I'm actually skeptical that any ML recommendation engine is going to recommend better books than my friends will. The only thing I expect from a recommendation engine is discovery, by which I mean finding new things that neither I nor my friends have read that look interesting. Whether they're good is another story altogether...
My problem with Goodreads is actually the site performance. It's one of the most abysmally slow sites I've ever had the displeasure of using. Loading from page to page takes so long, I wonder if they're running on 486 machines in the background. It's unfathomable to me how something that has such a monopoly on the domain can have such terrible performance.
It's so bad, that I often don't even bother going to Goodreads except when my TBR pile empties out, and then I spend as little time as I can (which is a long time because it's OH MY GOD SO SLOW) so I won't have to visit it again for a while.
I wish it was open source, or something, so I could contribute to improve response times...
Really? GoodReads reviews have devolved to complete garbage. The first few pages of reviews are always from the same people trying to be GoodReads famous.
Everyone gives everything 5-star reviews because they get more likes for positive reviews. The user reviews mean nothing. I'd like to have a way to see what people actually think about books. But on GoodReads that's almost impossible to find.
I'm surprised there isn't a rotten tomatoes for books. I've considered starting it myself, but I don't see much money to be made from it. Maybe that's why...
Interesting point about reviewers skewing positive for popularity. My big issue with most recommendation engines is that they're biased towards false positives rather than false negatives. The other way would lead to fewer, more focused recommendations.
In particular, I'd like to see an engine take a reviewer's dislikes into account more. If someone likes (switching to film for an accessible example) Cinderella, The Little Mermaid and Mulan, their opinion on Brave isn't going to carry much information as they're pretty indiscriminate about Disney films. But if they love Mulan, dislike Little Mermaid and hate Cinderella, their opinion on Brave is more likely to line up with my own.
I would think that the challenge is to find good recommenders rather than specific recommendations.
If I can find similar tastes, maybe a little bit different, then that will result in good recommendations.
So I would want to find people who both like similar things but also dislike similar things. It will be interesting to find people my similarity and see which are best at finding books I like.
I follow my friends, and my friends tend to have good taste! This is essentially my strategy. I do have a few friends who have recommended books that I didn't like, so I just don't listen to their recs anymore. Solved problem!
That's one of the problems I have with ML-based recommenders. I can't identify any particular ML algorithm, and therefore I can't say "This one has consistently given me bad recommendations, so I'll ignore it. But I do like this one over here." Whereas with my friends, I know that Lori and Nicole gives good recs and Brad doesn't, so I'll ignore Brad's and read recs from the other two.
The "smile or die" mentality propagated by sites like FB, where you only have likes but no dislikes, is throwing half the information away. I think it stems from a misguided desire to avoid adversity.
In my experience this really depends on genre. Some sub-genres are absolutely awful in this regard, while others are pretty good. In my experience, reviews on a lot of postmodern books are good as long as you mentally filter out reviews which are basically "I didn't want to read a postmodern book" :).
I wish for better reviews and scores too. The animated reviews are simply awful. In some genres virtually everything has near 5 stars, even the most derivative badly written dross. I now ignore any positive reviews and look at the negative reviews instead. I look at those which seem the most level-headed and mention things I care about.
I think they're fine? I didn't say great. In fact, I did say that I don't think ML will ever live up to my friends' recommendations, and that's primarily what I look at on Goodreads. My news feed thing is a list of books my friends want to read, are reading, or have read and rated, and that's more than good enough for me.
As far as the recommendations goes, I mean, I guess you could say I skip past most of them and only get the interesting ones, but at least the algorithm put it in front of me.
Again, my bigger problem with Goodreads is the performance. I'd use it a lot more and interact with my friends a lot more (and thus get better recs from them!) if it was a speedier site.
My most hated part of Goodreads is that if you type something in search and try to click the middle mouse button to open the resulting, highlighted book in the new window... it just copies the page you're already on. It's a site for discovering books, why do I have to open the landing page 5 times if I want to open 5 books? It's so counterintuitive to me.
> It's unfathomable to me how something that has such a monopoly on the domain can have such terrible performance.
You’ve answered your own question there, the keyword being ‘monopoly’.
If there is no competition pushing them to improve performance or site design, it’ll inevitably fall by the wayside until they get around to it or more accurately until somebody can convince a manager/business that they should spend time working on it, which, without using ‘look how much better the competition is’ that’s gonna be a hard sell.
I’m looking at this with broad strokes though, but as a best guess I’d say that’s pretty close.
I don't think there's anything monopolistic about Goodreads .. we just convince ourselves of it. But as we can see here, people would be highly willing to try out some competition.
It's a lot harder than people give it credit it for as well, I imagine. "why don't they get a recommendation engine already" ... that's the secret sauce of companies like Amazon and Netflix who have a financial interest in every successful recommendation. Goodreads doesn't have a profit model in place afaik to support honing the engine to that degree
I'm glad the recommendations work for you, but I agree with the blog post author and echo their experience. I think Goodreads recommendations are awful. All of the other recommendation engines I use regularly (Facebook, Twitter, Amazon shopping, Amazon Prime, Google, targeted web ads, Netflix, Steam) work significantly better than Goodreads does.
Discovery engines are pretty easy if you just use clustering unless you allow transitivity (book in two clusters can be used to suggest a book in either cluster)
Because it's only actually easy of you remove the irremovable thing that makes it hard. This is one of those tropes that honestly really hurts the credibility of ML in the eyes of non-practitioners. Practitioners get really excited about some 75% solution, and talk about how this thing which is bad everywhere is really easy. The average listener doesn't realize how hard that last 25% (why would they? They were just told it's easy!) and end up writing off the whole thing as snake oil.
Spotify pretty reliably finds me one or two interesting tracks in my discover weekly. That still leaves dozens of tracks I don't particularly care for (although I generally don't actively dislike them) but I still think it's a decent result.
I guess the obvious problem is that for books the equation is different. If you get recommended 20 books and after reading them you only ended up really liking 3 of them, then you'll feel like you've wasted a lot of time reading books you don't care for. Meanwhile if you listen to 20 audio tracks in the worst case scenario you'll have wasted an hour, and you can do that in the background while you do something else.
I've found that Last.fm's recommendations are infinitely better than Spotify's. I feel like Spotify's recommendations have too much influence from their partners and sponsors.
If all recommendation engines can do is essentially try to find the right cohort for you (which is how they all work because they can just try to go "people who liked A also tended to like B"), then you'll never like them if you have narrow tastes. Recommendation engines are blunt instruments, after all. Just think of how fundamentally unreliable "people who liked A tended to like B, so you might like B too" actually is for such a personal topic.
You'd have to wait until recommendation engines can read your mind and run your simulation against books.
Good lord what I wouldn’t give for a better goodreads. Never has so much useful and actionable data been squandered. For example, I could write pages about how bad the “top books” lists are. The Goodreads Choice Awards are purely and literally popularity contests, for example. Why would this be the case - you have millions of user ratings, you should be using those to surface exciting and unknown books rather than throwing people back at the same few authors they already know. The lists work exactly the same way - they discard ratings entirely and only take into account the quantity of books. It would be as if you said that the top 40 is the best music because everyone listens to it.
So you don’t think I’m all talk, a while back I got so fed up with this that I wrote my own script to scrape goodreads and find the actual best books, not just the most popular, and I found a wealth of really good and unknown books, including two books that are now my favorite books of all time. It was a side project that took me an hour. Why goodreads can’t do this is utterly beyond me.
I'm confused. What you describe is really just another form of popularity contest, no? Unless you have a set of reviewers that you value over others. One might call them the critical reviewers. (No, I'm not trying to be subtle)
So, what books did you find? Why not make a post on your method?
But that is still a popularity contest, of sorts. In general, any ranking system is a popularity system. Which is logical, as there is no objective ranking of books. That is, it isn't an ordered set.
That said, I can see value in providing a score not just in the items, but on the ranking system used. Could even be adaptive based on genre. (Indeed, I would love a system to see if any books have factual errors and such. But, that is just trying to reinvent the citation systems of academia.)
Easier to think of this from the other side. I have a few friends that, if they tell me to avoid something, it can typically override however popular something may be.
How did you define the actual "best books" ? I usually pick a group (Goodreads users in country x), scrape the books of all its users, and sort them by frequency: https://github.com/harjoc/goodreads-group-books
Thinking on johnfn's response, and yours, these are all simple SQL queries. You could easily delineate to others what each query does, and let people choose from a list.
EG "best average" "most reviewed" "users in country + frequency" and such.
This doesn't even need to hit performance, as these queries could all run on a secondary SQL server, be cached in redis/whatever once an hour, or even day.
After all, this sort of list output doesn't need to be second by second changing.
So simple to implement, non-performance impacting, which makes it even more curious as to 'why not'. Likely, a UI/UX person thinks it will 'confuse' people.
(I know that may not be the case, but it seems to kill more sites, and more flexible functionality, than anything...)
Infinite jest by David foster wallace and only forwards - forget the author name off the top of my head. Infinite jest is pretty popular these days. The other I have yet to hear anyone say they have read.
Ok, but was Infinite Jest ever an "unknown" book? I'm pretty sure it was one of the most hyped books in the world the year it was written. And I'm pretty sure it's always been in the 4.x's on Goodreads.
I’ve read Only Forward by Michael Marshall Smith. It stuck in my mind the same way that Vurt did. I still think about the man with no head sometimes, and the idea of a road being different when you’re going the other way on it.
Wow, turns out I still have it on GitHub, including the results I ground out 8 years ago. The script probably doesn't work, but the results are still good:
Particularly, load bigdata.js into nodejs and then run a command sort of like this to parse out the results, filtering out young adult/romance/religion stuff/comics:
I get some pretty interesting stuff. First result is the Constitution... OK, fair enough... but the next 10 or so are:
'A Song of Ice and Fire',
'Collected Fictions', (by Borges)
'The Name of the Wind',
'Infinite Jest',
'The Complete Works',
'The Way of Kings',
'The Wise Man\'s Fear',
'Ficciones',
'A Storm of Swords: Blood and Gold',
'The Complete Stories',
'Labyrinths',
'Don\'t Let the Pigeon Drive the Bus!',
'The Hiding Place',
Is it an amazing list? Eh. Is Name of the Wind better than Infinite Jest? Probably not. They're both fantastic books, though! And still, it's way better than Listopia. Also, this data is 8 years old. I bet it would be way better if I were to clean it up and run it in 2020.
Happy to read that you are working on improvements! GoodReads is my most-used and most important database. I love it, I also actually like that it feels more like an “older” webpage before all this craziness with stealing the attention and time of users came up (Goodread in its current state is already addictive enough for me). I hope you fix the bigger issues (like slow loading times, recommendations, convoluted UI) but keep the aim-at-making-the-page-actually-useful (and not a manipulative time sink) approach.
Feature wishes: Extensions on book statistics, a function for discovering like-minded users, marking books as „finished“ / „decided not to finish“ and showing this information (this could provide some very useful additional information on books), and maybe a „listened to the audiobook“ marker.
Can we have half stars or quarter stars? I read a lot of books, and choosing full-star 1 to 5 makes the ratings pretty useless for people who read a lot. My internal rankings are:
1 hated it/was offended/didn't finish
2 disliked it/likely didn't finish
3 finished but it was a struggle
4 finished/enjoyed/had some issues
5 finished fast/really enjoyed/may or may not have still had issues.
If I look back at the 30 to 50 books I read in a year, it's really difficult to get any sort of ranking out of them. I've only got two choices for books I liked (4 or 5) and two choices for books I disliked (1 or 2). Having half stars would help my ratings relative to one another.
Your post only goes to point out how arbitrary rating systems are since everyone brings their own ideas into the system of what ratings mean.
What you really want is some personal bookkeeping which doesn't make sense for the 5-star system. You'd want tags or the private note textarea to scribble down your arbitrary personal rating system.
Frankly I think more ratings are just worse. You're pitching 10 or 20 different options instead of 5.
Yep, I could do with a simple thumbs up/down rating system as I usually only rate books either 1 star if I don't like them or 5 stars if I do like them. The multiple stars rating system is pretty pointless to me.
It's the problem with star systems, which is that we'll always have different definitions. Your 3s are living next to the other poster's 3s and mean very different things.
That said, I think the world has also suffered from ratings inflation. I tend to assume anything under a 4 means "bad" or "meh" myself.
This is exactly the way the current Goodreads rating is supposed to work (and I'm personally OK with it). But my guesstimate is that for 95% Goodreads users everything below 4 stars means that the book sucks.
This is so vague that if you're not faking it, you might as well be. Sorry, but don't bother posting, even under a throwaway, if you're not going to engage.
Do you recommend books based on ML or do you hand pick books you want to promote because you think they should be promoted or have been paid to promote them?
Thanks. I checked out recommendations just now and indeed they make sense. I swear at some point in the past when I last tried it had very suspicious results, hence my question.
The way this is phrased is odd and makes me question the company culture. There is nothing in the content of your post that could be deemed as fireable by any reasonable person so I am left to conclude the work environment is a hostile one.
At many large companies, you need to have permission to post about the company. When the number of employees gets into the tens of thousands, odds get very good someone will say something that reflects badly on the company.
Generally, it's not that you can't post, you just can't post an an employee.
The companies generally don't want something like:
Yikes, that's rather dramatic! Especially since all you've said is "we're working to fix it. changes soon." That's a pretty harmless statement, I would think.
What gives you the idea that such statements put your employment at risk?
This implies support for Goodreads at Amazon is very thin. It could be read as a critical statement (even though it's pretty clear to everyone that the site very clearly stagnated since the acquisition, focusing almost exclusively on Kindle integration).
Amazon management is famously militaristic in attitude and would likely take action against a low-level employee criticizing the company in public
Although I know that Amazon is hiring intelligence operatives[0] to bust unions, this aspect (militaristic middle management) is news to me. Where can I learn more?
I'm pretty sure it was also reported in the past that Bezos traditionally favored ex-military for manager positions when Amazon was starting, but I cannot find a source atm (seems like yesterday to me but more than 20 years have since passed...).
I downvoted you but I feel bad for not explaining why.
One of my memories: Being pulled into a side office and grilled for a half hour for posting a similarly harmless statement to HN.
So your comment triggered mild PTSD.
It's not your place as an individual employee to post about your employer. You work in a team setting. If the team decides it's appropriate to post to HN, then that's fine. But you alone do not get to decide that.
What? You are suggesting that a person is unable to share his personal experiences and opinions unless their employer agrees? Obviously this may damage the company, but personally I have no sympathy for a company worth billions of dollars who is not known for their fair treatment of workers at all. I see nothing immoral in the original commenter releasing this information.
Strange how your response to abuse isn't to warn us about abuse, but to moralize about the actions through which your abuser justified their emotional violence.
Everyone's talking about recommendations. I don't use GoodReads recommendations at all. It's a social network with a focus on books where people will actually read and discuss books and book reviews. It's also a good way of chronicling what books you read. I wish they'd focus more on that part.
I use Goodreads to post my reviews of books I've read.
But I'm looking for a way to post reviews for individual short stories found online or in fiction magazines, anthologies, etc. The reviews should automagically pop up whenever the story appears in any other place. But I haven't found any sites that does this.
For example, I read and review Story X in anthology Y. If the story appears or appeared online at Z, printed in magazine F, etc, my review for it also appears if I look it up there.
Looking for recommendations for sites that do this.
Yeah, it's hard for me to take the complaint too seriously. The problem with books isn't that it's hard to find the good ones. It's that you get this tremendous backlog because you're short on time.
> They recommend different versions of books I’ve read. They recommend two different versions of Lord of the Rings (one of my favorite books), but I guess they don’t know these are the same book.
This is failure mode of recommendations is so common and so catastrophically useless that I do not understand how it has not been solved.
I'm into photography, so every now and then I buy a lens from Amazon. Immediately after I do, every Amazon ad banner on every website in the world switches to advertising that specific lens, for like the next month (I guess until I buy something else).
Lenses are very specific, expensive, and singular. There is almost no reason to ever have two of the same lens. The day after I buy a lens, it is practically the least likely product in the world that I will buy.
Show me anything but that. I mean, actually, don't. I like this failure mode because it makes it easy to tune out the ads. If they showed me related gear (perhaps filters that fit the lens), I might get suckered into spending more.
I think this is because the ad networks don’t have insight to your purchase data. So they don’t know if you’ve bought and converted or not, but know that people who searched for something are more likely to buy if advertised the dickens out of for the next few weeks.
That’s my theory at least, but even Amazon will show ads for stuff I bought on amazon. Not sure if that’s just them being stupid.
I’ve had this same thing happen with so many large purchases. If I’ve just bought a washing machine the odds of me wanting another one the next day are astronomically small, unless they think I’m in the process of starting a small laundromat.
The status quo regarding Goodreads and it's position in the industry is really interesting.
There is not a single competent alternative to Goodreads, yet Goodreads just sucks! As simple as that, the UI is horrible, the UX is disgusting, the perfomance is that of toaster and still people are using it.
I personally am using it for the last 5 years while continuously searching unsuccessfully for alternatives.
If you think the website is bad, you should check out the Android app, never in my life have I seen a worse app by a popular company.
I got so frustrated by Goodreads that on multiple occasions I thought of starting a competing product just to challenge the status quo, but as soon as I started dwelling deeper into it, I realized that there is just no real viable business model for it to be worth it.
I guess that explains the current status of Goodreads and why there are so few competitors.
How do you monetize a social network about reading books?
I have spent way too much time thinking about it, yet failing to achieve a result.
1)You can't really sell books, since Amazon is already so well entrenched and honestly books aren't a really hot commodity.
2)Subscription for audiobooks? Again there is Audible, and Scribd is also doing a phenomenal business doing it, you pay $9 and listen to as many as you want audiobooks.
3)Affiliate marketing? Goodreads already does it, and it isn't really a business model with a good foundation.
4)Customized ads? This might be the "least" worst solution, although what could you advertise to people who read books? Their LTV (lifetime value of customers) isn't really high.
5)A paid social network? Good luck trying to grow a social network which asks it's users to pay $1 monthly.
6)The latest and craziest idea I had regarding it, was to make it a hybrid of Goodreads and LinkedIn. Where employers can see what types of books are their applicants reading. I.e. if I am looking to hire a backend intern, I am sooner going to hire the intern who read in his free time Effective Java, Clean Code, etc. since you can pretty much gain a good overall picture about a person by looking at the books he reads.
> I realized that there is just no real viable business model for it to be worth it
I think this is key. It is kind of amazing that on a site all about startups and finding "product/market fit" every single thread about goodreads completely ignores any aspect of viability and expects someone else to just spend money for purely aesthetic reasons like it is a charity project.
Is goodreads perfect? Good lord no. If I owned it, would I spend money trying to improve the search or the mobile app? Hell no. Hire 5 developers at $1 million/year to...what exactly? Is a better mobile app going to translate to tens of millions of dollars of revenue somehow?
It isn't like LibraryThing or StoryGraph or BookSloth or Riffle are actually any good and they have 5+ years of development to solve all the alleged issues. Yet everyone in this thread complaining about Goodreads is still using Goodreads instead of them.
Money is where ideas thrive. I would build something based on recommendation based on users who you like. I like someone and I want to know the books they like and more and more of this. so essentially Facebook for Book recommendation. I know someone and so I know what books they read. User can choose to publicize the recommendation or to certain group of people or private.
After reading and heavily agreeing with this post, I came to the conclusion that either goodreads is not really trying, or -more likely- the data is just not good enough to make decent recommendations. There are so many biases in the review data that are impossible to fix in any kind of sparse matrix recommendation algorithm. For those who want to try anyway, it might be worth downloading an existing dataset (1) (104 million reviews) and try, before worrying about scraping and api limits.
The only solution (in my experience) is to get some other way of quantifying content, like Spotify does by manually labelling tracks. After some ddg I found storygraph (2), which does this. Its search engine is quite impressive, might be worth trying.
the problem is that when you step out of genre fiction, you cant just recommend a book with a similar plot or setting, it needs to have a similar style of writing and sensibility and thats very hard to determine.
the best way is friending/following people and learning their tastes compared to yours
> the best way is friending/following people and learning their tastes compared to yours
If only goodreads could somehow find others who like the same things you do and use it for recommendations!
Joking aside, they might already be doing that, but if so, they suck at it. They have enormous datasets of people and their tastes, yet their recommendations always seem to be simply matching genres that you happened to like a book from. You liked Watership Down? Here are 10 other books with talking animals!
If you and I can successfully identify people we share tastes with, then goodreads should be able to, too, they have even more data to base this on.
Book rating and recommendations are an impossible problem to begin with. If you browse through Goodreads, you will see that everything is rated 4 stars. People only finish and rate books they like, and it's impossible to compare books of different genres, lengths, time periods, complexity etc. on the same objective scale like it is for movies or TV shows.
So look at which books I finished and rated highly, find people who overlap strongly with that, and recommend the set of {their books - my books}.
As I recall, this is basically what won the Netflix recommendation engine contest. There's going to be a lot of computation in crunching the numbers, but it's not hard to come up with something better than "recommend what has the highest average star rating".
Like, even when you have access to the site's entire API and can write your own client for it, there's still the fact that their recommendations are generally better.
It sounds like an extremely important value-add. There are many sites that I will only use the app for because that is where the recommendations are shown. But to me taking a user profile and an article and spitting out a list of articles seems like magic.
I also find it weird that in 2020 Goodreads is the status quo for book recommendations.
I get the sense that part of the problem with book recs vs. YouTube video recs is that you can evaluate the quality of a video recommendation super quickly (click in, find out like 10 seconds in that you don't care for it, click out).
With books, actioning a recommendation involves
1) getting a copy of the book (digital or physical, both cost money or at minimum the time required to pirate),
2) starting to read it (with the sunken cost of time and/or money + the drive to "give it a chance" both looming over your shoulder, costing you more time),
3) eventually either getting to the end or swallowing your pride and bailing.
I think you'd start to see actually useful systems here if you could even eliminate step 1; make it easy as clicking in from a recommendation directly into reading mode.
Also solves a critical issue for the engine: gauge recommendation quality by how soon people click out of the rec, as opposed to waiting for the user to go through steps 1-3, and care enough to come back and provide a rating. Way more data to work with.
Of course, this all skirts the Actual Problem: the amount of IP law you'd have to trudge through in making enough books this accessible.
I also recently googled a quote, found it in a specific page in a book hosted on https://www.pagebypagebooks.com/, and just impulsively clicked through the whole book because it happened to capture my interest, and only "demanded" that I commit to a single page at a time. Hard to explain, but it felt more natural in that every-page-is-a-URL format to trivially bail at any page, without the weird apprehension I get from the same action in an e-reader / physical book.
The website is super old/limited, but if all books were that trivial to access + click through, I think we'd see something more interesting here. Made me wonder how hard it'd be to convert the entirety of Project Gutenberg into that sort of Web 1.0 format automatically.
Poorly, its either shovel so much shit against a wall to see what sticks (youtube) or you sorta watched this for 10 minutes to see if you liked it but you didn't but here is more (netflix).
Err, I disagree with your assessment of Youtube's recommendation engine being "poor." It is probably the most powerful recommendation engine in the world.
5 billion YT videos, comprising 1 billion hours, are watched every day, a meteoric rise ever since they improved the recommendation engine.
That recommendation engine has also been accused of contributing to the rise in polarization and conspiracy theories world-wide. So it may not be a force for good. But it's damn powerful.
Part of what makes YouTube’s recommendation algorithm so powerful is that it doesn’t necessarily have to be very good. I scrolled through my recommendations just now and only found one video after scrolling through fifteen that I was interested in watching — but that’s fine, since it’s easy to decide if you might like a video based on its title, thumbnail, and creator.
However, if YouTube was as good at recommending books as it was recommending videos, I would never use its book recommendation algorithm. You necessarily have to get almost every book recommendation “correct” for it to be seen as actually useful by the user.
How does that follow? I can't imagine that there are any book recommendations -- be they your friends, the Book Review section of your favorite newspaper, or whatever -- that gets things right for almost every single review. That's absurd.
When I see a book recommendation, I read the associated review (if it has one), read other people's reviews, look at the author and the blurb, and, yes, even the cover.
That tells me much more than a YouTube thumbnail, with only a little more time (but of course I'm going to put more time into choosing a book that might take me two weeks to finish). Indeed, my initial filter is probably much the same as a YouTube thumbnail: I skim past many recommendations based on the genre and author alone.
I think if the Times Book Review always wrote reviews in which 1 out of 15 of them made me thing "I need to get that book," I'd be amazed at how much they "get" me. I wouldn't bother with anything else. Every 15 recommendations lands me another "must-read?" Sign me up!
The problem listed in the article is that Goodreads is no where near 1 out of 15.
One method, collaborative filtering with latent factor analysis, popularized by its efficacy in making recommendations on Netflix, is to use matrix multiplication to solve the problem.
E.G. Let’s say you have all users (rows) x all books (columns), in a massive sparsely populated matrix, where the value is the rating that user gives a book.
To make recommendations, the goal is to “guess“ what a user would rate a book they haven’t read, and if your guess is they would give it 5 stars, and then you recommend it, and the user gives it 5 stars, it’s a good recommendation.
The “latent factor” idea is breaking the problem up, so in order to compute the rating matrix that is the final size N users by M books, you split it into:
N users x D latent factor
(cross product)
D latent factor x M books
=
N users x M books
It then becomes a machine learning problem, using a loss function plus gradient descent, to solve for that D latent factor.
Customers with a similar latent factor, will have similar taste.
Once you have the latent factors, you find the nearest neighbors (the closest other latent factor vectors measured by dot product or cosine similarity), to compute the nearest books. The vectors that multiply together to give the highest rating, will be the best recommendations for the user.
For me, I only use goodreads because there’s nothing better that I’ve found. I think of it as similar to evolution, evolution is just about survival, not necessarily the best at stuff that’s not survival.
It's not hard to build something that gives you recommendations. It extremely difficult to build something that works well. For consumer companies I know only Spotify and TikTok that do it well, and Youtube that does it OK-ish.
The recommendations basically crapped out on me after a few hundred books. It seems there are a few cliques (graph) of recommendations that are pretty independent and if you have read these cliques in a genre recommendations just suck.
Seconded. The "new releases of authors I've read books from" is one of the things that would give me the most value -- if it actually worked properly, which for me it doesn't, since it only seems to show _some_ random new books from _some_ of "my" authors. Sigh.
I don't create about recommendations as much. I get those from other sources and already have enough of a backlist to last a life time.
Goodreads for me is for storing my history and ratings. Same with IMDB for movies. Both sites are terrible (and owned by Amazon), but I don't see much value in clones that don't do anything completely new.
Has anybody ever used stack ranking to power book recommendations? Instead of rating books on a scale of 1-5 star you instead rate each book relative to every other book you’ve ever read.
Some other factors that I think could be used to create a high quality book recommendation engine on top of stack ranking: number of books read (the top 5 books from someone who has read 200+ books is a stronger signal than someone who has read 10 books) and release date (give extra weight to older books to reduce recency bias).
What do you think? Am I onto something or just spewing some sat night nonsense?
Are all books comparable linearly like this though? I could see separate stack ranks for different genres (practical books, sci-fi, relationships, whatever), attributes (readability, teachability if applicable, subject matter interest), or more.
I also can't help but think in graphs, so I could also see horizontal-ish linkages between the stacks as well. Each book, then, could have a number of attributes that get an independent stack ranking for each one, leaving you with a constellation score as well as a focused look at what the book is good at or focused on. I could see it getting unwieldy though.
Thinking through this, maybe genre stacks based on some collapse of attribute ranking scores.
I think this could work, although perhaps in combination, not replacement, with the traditional ranking system. The book on the bottom of somebody’s list isn’t necessarily a bad one.
Honestly, I feel that most tech goes this way when you have a monoply. Everything has a nature lifecycle, theres new life, interest, peak, and death/replacement. However, with some of these large corps (i.e. Amazon in this case), they can choose to hamper the growth and avoid competition.
A lot of our tech can/should be a lot better than what it is. Use the tech to help people collaborate/get together/share. That was the great thing about reddit in it's hayday.
My wish: show me all the books my friends have reviewed, and allow me to sort by number of reviews. I want to see what a plurality of my friend recommend.
I'm skeptical of this person's ability to "know" what they don't want to read. E.g., they are recommended Shoe Dog and a King novel, which they claim they don't want to read. However, if other people have read similar books and rated them similarly to this person and have read Shoe Dog and that King novel and rated those well, then this person may like those books. It seems they are assuming they won't like them, but not for particularly strong reasons. One of my favorite authors is Haruki Murakami and I would have rated Colorless Tsukuru a 2 or 3. If that were the only Murakami novel I had read and I made the same assumptions this person made, I would have missed out on some of my favorite books.
I feel the same way in that I wonder if maybe I would like it if I read it.
For the King book, my wife read it and we talked about it and I just don’t want to read it. For the Knight book, I read the first few pages and know the Nike story and don’t want to read it.
Maybe I’m missing out and they are truly amazing. But I can’t read all that they recommend just in case they suddenly get great at recommending.
It’s a challenging problem and the risk of missing out is there. But I don’t know of any way to truly eliminate the risk unless I read every book recommended. And my life isn’t long enough.
If you are looking for more of magical realism Murakami has become known for, I highly recommend Hard-Boiled Wonderland, The Wind-Up Bird Chronicle. But really anything 1Q84 and earlier is good IMO.
I love goodreads, but I've wished for so long for two things way more basic than good recommendations which I don't really care about (except the finding people with similar ratings, I would kill for that):
- Proper search engine (all books with x+ ratings, rated y+ start, in genre z). From what I've heard one has to scrape the site to get any useful information.
- Forbidding rated reviews before release dates. This seriously skews ratings for some genres and is insanely annoying.
Okay maybe three... managing shelves is super slow. I want to be able to drag and drop and have things instantly re-ordered without a reload.
I’ve been building a prototype over at https://longtweetsapp.com because I’ve had the same problems. It’s hard to find books so I took a more personalized, data-driven approach, starting with Twitter networks, because I was tired of bestseller and most popular lists.
It’s still early goings, but I’d love feedback - I’m dogfooding it with interesting results. Send me a note: hareesh.ganesan+longtweets@gmail.com
Hi HN. I come from a parallel universe in which book recommendation systems are not based on genre, ratings or communities. Book recommendations are based on their content.
You should check out: https://beta.thestorygraph.com/
It's still in beta but I've tried it and like it a lot. One thing that it's missing that is holding me back from a full jump over to it is an android app. The mobile site looks good, but I think an app would substantially improve usability.
I'm trying to tackle a small part of this by building https://www.readthistwice.com which gives contextual recommendations from 1300+ leaders. What I mean by contextual in this case is every recommendation comes with a verified quote from the recommender on why they recommend the book.
There are many Goodreads competitors. To the others already mentioned, I'll add https://inventaire.io/ , which is fully free software and based on open data (Wikidata).
It's useful if you want to have full control on your own book catalogue while not having to produce all the data yourself.
If they're using collaborative filtering, there probably isn't a simple explanation. Basically, you feed in a list of "<user id>, <book id>, <rating>" and the algorithm generates a feature vector for each book and each user. So the reason some book gets recommended is because... the dot product of that book's vector with your vector was high. You can do this easily with off-the-shelf libraries, like Surprise[1] (I'm using that lib in my startup[2]).
At least this is what happens in matrix factorization methods. Recommendations from k-nearest neighbor methods can be explained more easily, but knn doesn't scale as well.
This is a minor drawback of matrix factorization--people have been shown to trust recommendations more if they understand how the recommendations were generated. Twitter recently published a super interesting paper[3] about how they generate recommendations at scale, and as a bonus their method is explainable. They create a model that describes the communities in Twitter--groups of people who follow the same set of influencers. Users and items are represented by vectors, where each element of a vector describes to what degree a user/item fits in a certain community. When generating recommendations, you can get the user's top communities and then fetch items for those communities. If you generated a text description for each community, you could include that with each recommendation.
> My idea for an 11-star experience 1 in finding new books is that Goodreads knows me even better than I know myself and constantly recommends the perfect book.
> goodreads shows me five books that I don’t want to read.
I wonder if these two ideas are at odds with each other. Imagine for a moment that recommendation engines were solved problems, and definitely worked given the above statement. They know you better than you know yourself. Would it’s recommendations likely only include books that you obviously wanted to read? Or would they include books you didn’t know you wanted or needed to read? I mean this in terms of judging the books by their cover rather knowing about their existence. Isn’t it likely or even probable that the majority of books recommended would be based on the value they contribute to something deeper than the pure enjoyment purposes?
As an aside, I remember in a college literature class I took the instructor told us that it’s up to the reader to derive value or meaning from stories. This was a class that studied short stories of early American authors. Most of which were slice of life narratives that didn’t have any apparent meaning, or commentary from the authors themselves. The exercise was to study the characters, scenery and tone and try derive what might either lie beneath the words or story themselves. Whether the ideas we deduced from the stories were accurate (and in most cases probably were not) the value Of the process was of critical thinking about the stories that made us consider and express ideas and beliefs we normally don’t.
Back on topic, would a working recommendation engine likely suggest things that on the surface seemed either boring or blatantly unappealing that would provide tremendous value if we put work in to reading and studying?
That being said, is it possible that current recommendation engines are already working? Most are at least driven by reading behaviors of the masses, so it seems like it might be feasible that that Steven king book that is unappealing to you is something you should actually read.
(This is not including recommendations for books you have already read in different languages which seems like an obvious bug, but then again reading books you’ve already read in different languages might be an excellent way to become more fluent in a new language or gain a deeper understanding of translation of ideas between languages....)
So, maybe the tech isn’t something that needs to be fixed. Maybe we just need to be open to what the tech is telling us?
I think you are right. Being recommended a book I've never heard about by some author I'v never heard of by some computer system, I'm probably gonna dismiss it. However, if I've seen it in a bookstore, read about it in a paper, heard a friend talk about it etc. then I might act on the recommendation.
Basically being exposed to something enough times. First time and a cursory look, most things don't look to exciting. This goes for everything. Movies, restaurants, gadgets..
One thing I'd add about the UX is that the search bar is buried below the fold on the homepage of the desktop version. Everywhere else, even on mobile, it's at the very top. It makes it a little more difficult to just jump to their site and start searching.
One interesting facet of all these recommendation and ratings systems online (Yelp, Goodreads, Rotten Tomatoes, etc.) is that they don't provide a key value that in person recommendations do.
When a person recommends a restaurant they are attaching blame for the restaurants performance to themselves. Its performance is a reflection on them and their credibility. It is very hard to replicate this on the internet, a place where people can suggest things with very little responsibility.
If I strongly suggest a movie to someone and the movie sucks, I can be blamed for that. I don't know how to replicate that comfortable eschewing of responsibility that in person recommendations provide.
I don’t care about reviews or recommendations. I have a to-read list that will last me the rest of my life easily.
The thing I hate about goodreads is there are two search fields one to search books you’ve read and one to search every book.
I’ve wanted to search books I’ve read like 5 times in my life but for some reason I always end up using the only search books I’ve read search bar and I get no results. Just make it a checkbox or return two lists or something.
It's pretty basic right at the moment and we've a lot left to build (largely a reading list only, we've got more in the works) – but here's my profile as a preview, since our marketing site & messaging needs a bit of work: https://beta.readng.co/user/joe
One of the things I always wished was built into goodreads was a good system for running a book club. They have groups, and those are close. But it just seems like they took Facebook groups and tried to make it work for a book club. It doesn't feel right to me.
Goodreads doesn't get enough love from Amazon, It always seemed really odd to me that the recommended books in the kindle store were better than goodreads even though it's all the same company. It seems like a wasted opportunity. There are also numerous bugs I've ran into on the app that I just work around.
A buddy of mine has started on a better designed alternative to Goodreads. It's in early access right now but it has potential and I'm looking forward to seeing what else he adds to it: https://readng.co
> I wish there was some way to note books that I don’t want to read.
Minor point but there is a "Not interested" button in the screenshot shown above this paragraph. While Goodreads' initial suggestions were off-the-mark, I wonder if using that would help?
Isn't it the same problem with Amazon? It shows me different versions of the 5 products I last viewed. And unlike, goodreads it seems like there is clear monetary value for amazon in fixing product recommendations but somehow it is the same for years.
For all the AI and machine learning being thrown in the pot of every dish being made, they still haven't figured out that if I just bought a flashlight, I'm not likely to get a double-pack to save $2 anytime soon.
make it like myanimelist or mydramalist, one of those sort of websites....
such websites were created by fans can cater to what fans care for:
1- Finding out what's coming next, in an easy to view format
2- Finding out what's related to the item I'm currently checking, and it's place in the overall verse's sequence
3- Damn simple tracker for the item I consume
4- Simple rating and genre system
In other words, you want to fix goodreads? create a mybooklist.
Because MAN it is SO hard to find out when my favorite author is having a new book coming out, or if I discover a new author, wtf is their flow of their book series, often ending in me accidentally reading the book out of order.
I just checked my Goodreads recommendations. It is very good at recommending books by authors I really do not care for in genres I really enjoy. I wonder if these kinds of systems fall over for that situation.
One of the best parts of Goodreads for me is the Listopia search. Regular search doesn't get me much, but this search is fantastic for going really niche on certain topics.
I'd like to propose a different thing to fix first: when I open that page, I can't scroll with the keyboard unless I first click in the main content area. I.e. the focus is initially somewhere else.
Contrast that with something like Apps or Shopping where you can use affiliate networks and monetize right from the start. And a competitor like Google doesn't have control over the Affiliate networks (which in books, Amazon has)
C) Most authors are not tech savvy and keep doing crazy things like giving exclusivity to one store and create monopolies. They then get screwed by these monopolies
Why would you step into a market where
A) The competitors are monopolists who have no compulsion against doing illegal things
B) The participants (authors and Publishers) are in the 1950s technology wise
C) The participants (authors and publishers) keep doing strange stuff like give one ebook store exclusivity
CONTRAST what happened with Apps versus what happened with Books
Apps
Amazon wanted pricing control of apps and exclusivity. Developers gave them the middle finger
Amazon tried unlimited apps (including unlimited IAP purchases if you are on subscription) and developers didn't play along. They had to close down the service
Books
There were FIVE major stores - Kindle, Sony, Apple, B&N, Kobo
Amazon offered exclusivity to authors in return for (the comically bad incentive) - allowed to make book free for 5 days every 90 days, allowed to discount book as a special Kindle Countdown Deal for 7 days every 90 days
Authors gave Amazon exclusivity. Helped it get from 30-40% market share to 60-70% market share
Now completely dependent on Amazon
Kindle Unlimited - Amazon won't give authors and publishers data on how many people borrow, etc. Only 'total number of pages read'
Authors play along and now many are completely dependent on income from this 'subscription service' where they don't even know how many readers downloaded their book, how many read X% etc. Just one vague metric of 'total pages read by all readers'
So, to do well in this market you are suited for it IF
A) Your default is to take advantage of authors and publishers
OR
treat them like naive sheep who don't understand the basics of business
B) You are willing to go up against a large monopoly that will also do illegal things without remorse
Amazon just hired Keith Alexander - head of NSA during warrantless surveillance
Do you really want to compete against that?
C) somehow bridge the gap between books and publishing and technology
while keeping in mind that most of the top quality supply is controlled by 5 large companies (the Big 5 very big publishers)
You would be hard pressed to find ANY business to go into that is worse
The saying - An intelligent enemy is better than a stupid friend
This is an ecosystem where - your friends are naive (in effect, clueless), and your enemies are stupid (doing all sorts of illegal things because they are a monopoly and know the US does @#$$-all to monopoly power abusing companies
Is this a good thread under which to have a more general discussion about data ownership and access? Goodreads is a Schelling point for book reviews, reading lists, and personal catalogs, but the API leaves something to be desired, not to mention the UI. Why shouldn't the Goodreads database be accessible to other applications? Why shouldn't the data be stored locally, or on our "own cloud"?
Why can't we fix what's wrong with it? It's maddening that sites' UI is locked down, unfixable.
