About

I read a lot and love talking about book recommendations with others, and my favorite type of those discussions to have are the ones that are like "if you liked <book> then you'll probably like <other book>"… so I decided to build a tool to help make it easier to have those types of conversations about a broader variety of books.

Feedback welcome

This site is a passion project for me, and it's a work in progress. Feedback about whether the recommendations hit the spot or completely missed the mark are VERY helpful - just select the thumbs up/down on any recommendation.

There's such a huge volume of data about the books in the catalog (see below for details on how it works if you're curious) and the data is really chaotic and messy, so having concrete examples of "I tried <book> as input, but the recommendations made no sense whatsoever" is incredibly helpful. And of course it's also great to know when the recommendations are spot on, so I know what's working well.

Privacy

This site is a passion project I built for myself and anyone who wants to use it. It does not use tracking cookies, third-party trackers, or ad tech of any kind. The only cookie used is a functional one for remembering UI preferences.

This site does collect very high level analytics such as which books were chosen as inputs and which tuning knobs were used, so that I can improve the catalog and user experience.

How it works (summary)

There's an underlying book catalog with tens of thousands of books that have been sourced from a variety of different services that each provide different metadata on those books such as genre, subject tags, star ratings, etc. The metadata feeds a ranking algorithm which uses a bunch of different factors to provide options for "if you liked X, try Y" discussions.

A key part of the user experience for this site is that you can see why the ranking algorithm had a particular recommendation, and you can customize the engine to emphasize what you want to see more or less of.

How it works (slightly nerdy details)

The data sources used to build the book catalog include:

• Goodreads and StoryGraph exports from me and some friends
• Open Library
• Hardcover.app, which is also where reader counts and star ratings come from
• New York Times bestseller lists
• Google Books
• Book award winners and nominees (Hugo, Nebula, Pulitzer, Lambda, etc.) from Wikidata

A book ranking algorithm based on metadata is only as good as the data is, of course — and while some of this data is authoritative and reliable in certain contexts (like NYT bestseller lists which come from the NYT directly), other parts of it are messy and cluttered, since the data's been crowdsourced and grown organically over time.

To help make sense of the chaos of the data, there's a "tag insights layer" which sits on top of the book catalog and does its best to organize and make meaning out of the masses of metadata from all these different sources so that what's fed in to the ranking algorithm is as clean as possible.

Some examples of what this tag insights layer does:

• Filters out the clutter. A lot of tags describe the catalog record rather than the book ("in literature", "adaptation", etc). Other tags are just too ambiguous to mean anything.
• Merges tags that generally mean the same thing ("space travel" vs. "space flight").
• Clusters tags that mean different things but are closely associated with each other ("dystopian" vs. "apocalyptic" vs. "end of the world")
• Creates hierarchical relationships between different tags. "Space opera" already implies "space" and "science fiction"; "police procedural" already implies "crime" and "mystery".

When you choose one or more books as your input "seed", the ranking algorithm compares the metadata on the input book(s) to the entire catalog (~28,000 books and ~20,000 unique tags as of when this page was written) and runs the ranking algorithm with different weighting to evaluate things including but not limited to:

• Tag overlap: how much the tags on your input books overlap with other books in the catalog. Narrow, specific agreement counts for much more than broad agreement — for example, two books sharing "heist", "found family" and "unreliable narrator" says far more than two books both being tagged "fantasy".
• Tags you emphasized (the tag chips under your input books): Select one or more tag chips if you want to shift the recommendations to focus more on that aspect of your input book(s).
• Similarity to one book: how similar a book is to the single input book it most resembles, rather than to the average of everything you picked. This is what lets a three-book seed still surface something that is a dead ringer for just one of them.
• Genre & sub-genre: fantasy, sci-fi, romance, mystery, thriller… and romantasy, cozy mystery, hard science fiction, etc. These work as a discount rather than a bonus — a book from a different genre than the input books is scaled down but not taken out of consideration entirely.
• Book DNA match: the defining elements of a book's identity — for example, HHGTTG is a sci-fi book about the end of the world, but the essence of it is more about the way it's written and the satirical and absurdist humor.
• Average rating: reader count and average star rating from Hardcover and Open Library, for books that have a decent number of ratings.
• Awards: whether the book won or was shortlisted for a prize, from a sweep of Wikidata — Hugo, Nebula, Pulitzer, Lambda, Coretta Scott King and various others.
• Bestseller history: did the book ever show up on the New York Times' bestseller lists, at what ranking and for how many weeks.