4 ms·

A way of reading HN that lets me say "That was interesting" and "That is of no interest to me" and then presents, when asked, all the posts from the last 48 hou

A way of reading HN that lets me say "That was interesting" and "That is of no interest to me" and then presents, when asked, all the posts from the last 48 hours based on how interesting a Bayesian (or other) filter thinks I'll find them. I've been intending to write my own, but I've been too busy of late to make any progress.

AlexMuir 16y ago
I'm not convinced - there's already an incredibly powerful filter on HN that finds and reads all the articles and rates them as being of interest to the audience. I suspect that the appeal of something that filtered HN for people with narrower interests (of whom I am one) would be pretty limited. I'm looking for something that is widely useful, rather than a fun coding project (which I'm sure this would be).
- RiderOfGiraffes 16y ago
  You asked for a pain point, I provided one. These days I'm finding most of the "new" page to be uninteresting, and worse, most of the front page to be uninteresting as well. Many, many good, interesting and useful items slip through the net and get insufficient upvotes to hit the front page, and yet on occasion when re-submitted get noticed quickly enough, hit the front page and stay there for hours, sometimes more than a day. So when you say: > there's already an incredibly powerful filter on HN > that finds and reads all the articles and rates them > as being of interest to the audience. ... I guess I don't know what you mean. And if you just mean the current mechanism of having a wide readership that rates submissions, it's not working for me. But I would guess that you're unconvinced and won't work on it, but that's your choice, and I fully support it. Someone might build it, it might be me, and I'll be really interested to see how much use it gets. If any.
  - AlexMuir 16y ago
    Sorry if I came across unappreciative - I do appreciate your idea. I just tend to enjoy seeing items that I wouldn't otherwise consider reading - perhaps this is still a bit of novelty from not having been here very long. I find that if I'm away for a few days the Best list helps me catch anything good I missed.
    - RiderOfGiraffes 16y ago
      No, indeed, and I wouldn't want to read only those items marked as being of interest. I just want to catch those that I should see, but missed because no one upvoted them.
      - AlexMuir 16y ago
        Something that ranked every non-dead article is well outside the scope of my skillset, even just looking at title and source. If you could get that working you'd be onto a winner.
zephyrfalcon 16y ago
Huh. I've been having the same idea. Except it doesn't have to be limited to HN. I was thinking more along the lines of a standalone system that you can give RSS feeds or whatever (a way to fetch new articles, in any case). You train it like a spam filter, except you mark articles are interesting or uninteresting. It goes out, gets more articles, presents the ones with the best scores, you train it further, etc. I've been kicking this idea around for a while, but like you, I haven't actually started, mostly because I'm not sure about the best way to do it. (E.g. web app, traditional desktop app, etc. Not to mention the implementation language...)
- sigil 16y ago
  Check out newsblur.com. The code is on github (all python and js).
  - zephyrfalcon 16y ago
    That looks almost like what I had in mind... except you have to tell it what you like and don't like about an article (limited to author, title, publisher). What I had in mind was, feed the text of the article to a bayesian filter, and let the system make decisions based on that. At first sight, telling newsblur that you like an author or publisher, doesn't seem to useful... you might as well subscribe to their feed in a regular RSS reader then. But maybe I'm misunderstanding it.
    - samd 16y ago
      I thought your idea was great and then I tried out NewsBlur and came to the same conclusion you did. It seems to be filtering based on title, author, publisher, and tag rather than doing any statistical filtering of the text. That's not very useful because I already know who produces great content consistently and I subscribe to their feeds. What I need is a way to filter out the good articles from inconsistent sources.
JoachimSchipper 16y ago
Wouldn't it be much better to base this off the upvotes of people with similar tastes instead of e.g. the article text? That might get you the best of both worlds (social/Bayesian filtering.) It's possible that you already meant this, but it wasn't clear to me, at least...
mathgladiator 16y ago
I've thought about that it a lot but also as a way of using text classification to divide up the stream based on context (international, business, hacking, software development, research).

AlexMuir 16y ago

I'm not convinced - there's already an incredibly powerful filter on HN that finds and reads all the articles and rates them as being of interest to the audience. I suspect that the appeal of something that filtered HN for people with narrower interests (of whom I am one) would be pretty limited. I'm looking for something that is widely useful, rather than a fun coding project (which I'm sure this would be).

RiderOfGiraffes 16y ago
You asked for a pain point, I provided one. These days I'm finding most of the "new" page to be uninteresting, and worse, most of the front page to be uninteresting as well. Many, many good, interesting and useful items slip through the net and get insufficient upvotes to hit the front page, and yet on occasion when re-submitted get noticed quickly enough, hit the front page and stay there for hours, sometimes more than a day. So when you say: > there's already an incredibly powerful filter on HN > that finds and reads all the articles and rates them > as being of interest to the audience. ... I guess I don't know what you mean. And if you just mean the current mechanism of having a wide readership that rates submissions, it's not working for me. But I would guess that you're unconvinced and won't work on it, but that's your choice, and I fully support it. Someone might build it, it might be me, and I'll be really interested to see how much use it gets. If any.
- AlexMuir 16y ago
  Sorry if I came across unappreciative - I do appreciate your idea. I just tend to enjoy seeing items that I wouldn't otherwise consider reading - perhaps this is still a bit of novelty from not having been here very long. I find that if I'm away for a few days the Best list helps me catch anything good I missed.
  - RiderOfGiraffes 16y ago
    No, indeed, and I wouldn't want to read only those items marked as being of interest. I just want to catch those that I should see, but missed because no one upvoted them.
    - AlexMuir 16y ago
      Something that ranked every non-dead article is well outside the scope of my skillset, even just looking at title and source. If you could get that working you'd be onto a winner.

zephyrfalcon 16y ago

Huh. I've been having the same idea. Except it doesn't have to be limited to HN. I was thinking more along the lines of a standalone system that you can give RSS feeds or whatever (a way to fetch new articles, in any case). You train it like a spam filter, except you mark articles are interesting or uninteresting. It goes out, gets more articles, presents the ones with the best scores, you train it further, etc. I've been kicking this idea around for a while, but like you, I haven't actually started, mostly because I'm not sure about the best way to do it. (E.g. web app, traditional desktop app, etc. Not to mention the implementation language...)

sigil 16y ago
Check out newsblur.com. The code is on github (all python and js).
- zephyrfalcon 16y ago
  That looks almost like what I had in mind... except you have to tell it what you like and don't like about an article (limited to author, title, publisher). What I had in mind was, feed the text of the article to a bayesian filter, and let the system make decisions based on that. At first sight, telling newsblur that you like an author or publisher, doesn't seem to useful... you might as well subscribe to their feed in a regular RSS reader then. But maybe I'm misunderstanding it.
  - samd 16y ago
    I thought your idea was great and then I tried out NewsBlur and came to the same conclusion you did. It seems to be filtering based on title, author, publisher, and tag rather than doing any statistical filtering of the text. That's not very useful because I already know who produces great content consistently and I subscribe to their feeds. What I need is a way to filter out the good articles from inconsistent sources.

JoachimSchipper 16y ago

Wouldn't it be much better to base this off the upvotes of people with similar tastes instead of e.g. the article text? That might get you the best of both worlds (social/Bayesian filtering.) It's possible that you already meant this, but it wasn't clear to me, at least...

mathgladiator 16y ago

I've thought about that it a lot but also as a way of using text classification to divide up the stream based on context (international, business, hacking, software development, research).