I ask because fivebooks.com, which I suppose is a "competitor" of yours, actually does conduct interviews with people about the five books they choose on a subject, and it seems like a lot of work. They only do a couple a week, versus 12-13 a day.

Five Books is doing great work, as is https://hardcover.app/ and https://app.thestorygraph.com/. The more book websites we have in the world, the better the world; each one is doing things uniquely.

I am super lucky to have a small part-time team that helps make the publishing process work -> https://forauthors.shepherd.com/team

The publishing process works like this: -> We talk over email to see if they are a good fit and interested -> They fill out a Word doc and return it to us -> Bella & Joy add it to our system and start setting up the books/author profiles -> I screen each list personally for quality and fit (I learned a lot in the first year) -> Our fantastic editors, Frances and Lauren, edit and tweak everything. -> The author gets a preview, and we tweak and adjust as they see fit -> We schedule publishing once they are happy

So, ya, not a video or audio interview, but we do a lot.

https://shepherd.com/bookshelf/time-travel?topics=Q11424

Hoping to be able to combine three ideas even more serendipitously in the future. You sort of can with multiple tags + a topic, but that feels like more of a filter than an additional idea right now.

I am working to really improve the accuracy of genres & topics, and adding themes. I am hoping I can tackle that in 3 months or so...

While I value human curation, and it probably has to come first, what I wanted to do was to take the LoC classification for a book (typically a triple that has 3 levels of specificity) and build a multidimensional search "space" where you could find books close to the one you were looking at.

In 1994, alas, the LoC data was only available via an insanely slow modem-based download, which made it unusable. The main data source we were using back then (Baker & Taylor / Books in Print) sadly fucked up their record of the LoC classification to that it was impossible to infer where the gaps between the triple's parts were.

The crux of the idea was to leave book classification to librarians, who've been studying this for years and likely doing a much better job than anything us silly software types could come up with.

So to use an example from your list above, in my mental image of the interface I wanted: you'd pick "Sapiens" and see it surrounded by a cloud of other books sharing the same LoC triple (could use other library classification methods too). You could zoom out, and now see even more books sharing only the first two elements of the triple, or even further and end up in "History". Think of it as a "word cloud" made of books, rather than words. You could meander over to another book, and from that book zoom in (and out) of other aspects of the LoC's classification space.

I still wish I had been able to build this, since it mirrored the best part of my experience in large physical bookstores. But it was not to be.

Not sure if the dataset(s) you're using would permit this either, just wanted to mention it.

I do use the LOC data, but only a very small % of books have that. I also use a NLP/ML tool called Wikifier that connects books to Wikidata topics as I want to use that data to help build knowledge graphs so readers can browse a timeline of WW2 and also see books on those subjects.

What is the problem?

With the topic system, it just isn't quite accurate enough; once I switch over to the newer AI systems later this year, it should work better.

With the genre system, the big problem is that the BISAC publishing genre system is being abused by publishers. So, the data is not reliable because of abuse and ignorance.

For example, I have thousands of books marked nonfiction and fiction because the publisher doesn't know how to use the system and picks nonfiction topics for a fiction book. Or, a publisher picks a book for a genre it doesn't belong in because they think it will get them more sales.

It is really frustrating and I am spending a huge chunk of my time this year to look at how I can fix it. It probably means ignoring this system and using AI to classify it with other data we have.

There have been some tough moments in the last year, especially with funding, but we are on a good path and I am having fun -> https://build.shepherd.com/p/shepherd-stress-path-forward

Super excited to ship reader features soon!

Sci-Fi -> https://shepherd.com/search/shelf/9008

History -> https://shepherd.com/search/shelf/12669

Fantasy -> https://shepherd.com/search/shelf/12664

Behavioral economics -> https://shepherd.com/search/shelf/3105

One HUGE challenge we have right now is that the publishers abuse their genre system, and the NLP/ML system I use for topic identification isn't great. I am working on a new system to bypass both of those and hope to roll that out later this year; that will drastically improve accuracy on topics/genres.

There is also a page for the best sci-fi books of 2023 that authors picked. We asked authors to share their 3 favorite reads of the year and then tabulated the results for new books and books published at any time -> https://shepherd.com/bboy/2023/science-fiction

I walked into the bookstore, and they had little index cards to show off staff favorites and a little half-tweet on why: "Andy loves the characters in this outside-the-box fantasy series;- you will dig it."

It was this magical way to meet the books. I thought it would be fun to try to do that while helping authors meet readers.

I ask because I just started listening to "Straight Man" by Richard Russo and looked it up on Shepherd. Shepherd mentions that "Straight Man" is being made into a TV series. That's out of date and Wikipedia has the up to date info (one season of a series based on "Straight Man" was created and called "Lucky Hank").

Cool site! Thanks for bringing it to my attention.

I am working on new bookshelf pages to expand that section. I want to add a more visual format and break it apart a bit more.

Check out the early mockup here: https://www.dropbox.com/scl/fi/mrvkev7pft5luvj10eqfc/Screens...

Here is what should show up:

Math bookshelf page - https://shepherd.com/bookshelf/math

Math search results page - https://shepherd.com/search/shelf/693

Loving the site, thankyou

Still a ton of rough spots, so feel free to send me any frustrations (ben@shepherd).

As I launch reader reviews, I also want to feed into that; I plan to prompt them with "What book do you love that is like this book?" I think that is great for other readers and it helps give authors a true comp from the eyes of a reader (ETA late this year or early next I hope).

I also own pilgrimage.com; I'd love to do something like this for music one day with the tagline, "Life is a journey; let's give it a soundtrack"!

So in this very fundamental sense, it's nothing like your local bookstore at all.

Every book on our website is chosen by a reader who loved it.

Here is an example: https://shepherd.com/best-books/mystery-and-suspensewith-a-b...

The only books within our website are those that a reader loves and picked as part of their book rec list, or fav 3 reads of the year.

But okay, I guess this this how you talk to people.

The only "kinda" system we have is for our favorite 3 reads of the year, we showed which books were picked the most.

Yes there is. Pick a genre, and you get a list of books explicitly ranked (rated, same thing) by the number of people who reviewed it (in essence, by "likes").

So not surprisingly you get authors like Stephen King, Michio Kakio, etc floating the top. The same predictable blandness you get at any large chain bookstore.

So okay, fine - only part of the site is based on straight-up favorite counting. But then I did some topical searches, and the result definitely appears to be algorithmic as well (judging by the large number of plainly off-topic matches).

I'm not saying this is a horrible site or that there's no potential for the basic idea behind it. It definitely seems to be more appealing, and potentially more useful than most online bookstores I've seen.

But whatever you're doing - it's definitely not the same as actual human curation.

Here are some explore points you can try via search:

Sci-fi explore point - https://shepherd.com/search/shelf/9008

Sci-fi bookshelf - https://shepherd.com/bookshelf/science-fiction

Book list by authors - https://shepherd.com/best-books/science-fiction-for-fantasy-...

In my post above: "Then I did some topical searches ..." For large / concrete topics, the result sets look good enough. For smaller / more nebulous topics, they start to fall apart. Which is not the end of the world, of course. But which definitely leaves behind a distinctly non-human aftertaste, shall we say.

Look, I have the feeling we're talking past each other. It's your site, go ahead and have fun with it.