← Back to context

Comment by costco

4 days ago

Yes I would say the handling of series is probably the biggest problem. Once my test metrics got to a point I was happy with and my quality spot checks passed (can I follow the models recommendations from one generic history book to Steven Runciman, also making sure popular books don't always dominate the results), I was ready to release because I had been working on this project for so long. The solution is probably using the transformer model to generate 100-200 candidates and then having a reranker on top.

3 comments

costco

Reply
walletdrainer  4 days ago

Not just series, but I seem to mostly get a list of other books from the same authors.

The recommendations from other authors are good, but as far as I can tell I’ve read every single one of them.

Continuing to aggressively add everything it recommends eventually does seem to result in some interesting books I wasn’t familiar with, but I also end up with more and more books that are of zero interest to me.

For what it’s worth, I started with:

  Infinite Jest David Foster Wallace
  Europe Central William T. Vollmann
  Gravity’s Rainbow Thomas Pynchon
  White Noise Don DeLillo
  One Hundred Years of Solitude Gabriel García Márquez

It is possible that there simply aren’t many books like these in existence, so the pool of relevant recommendations gets exhausted fairly quickly. I’d guess trending towards unrelated popular books is also just a feature of the source data, that largely sums up my experience with goodreads anyway.

Very cool project though. I did end up ordering a couple of new books, so thank you very much.

IanCal  4 days ago

Releasing is the right choice, well done with this it’s really cool.

I’ve only had a short play but a solution to this problem might be to show authors rather than books. Or select authors outside of the list the user has shared and then a top n (1,3,5) for each of those.

I feel like that’s how you’d recommend to someone else - type of book -> unknown author -> best matching few books from them.

After that the other side would be trying to find some diversity (if you think I’d like author X, personally you might suggest three different styles of book from them rather than three very similar books from them)