This is not a criticism of the work done, but I think the top 20-40 mentions are extremely obvious and a regular reader might be able to guess a good portion of

This is not a criticism of the work done, but I think the top 20-40 mentions are extremely obvious and a regular reader might be able to guess a good portion of these recommendations. What is really interesting - and started at with the “categories” — is tying the recommendations to explicit context. I didn’t dive too deep into the recommendations, but are the categories by book category, or by originating topic of conversation? It’s a narrow distinction, but a useful one. I’d love to see deep learning pull up a hierarchy of conversational topics on hacker news and match recommendations to those trees.