Key Word Filtering VS "Training" Feature

#1

I get that Newblur can use the “train” function to help sort feeds, but I really want something more absolute. There are some feeds that may have a lot of traffic, none of which I care about. I would prefer to have Newsblur mark news articles from such a feed as read by default UNLESS the article meets certain criteria. Maybe the author is a certain person, or the title contains a key word I’m looking for. I can see how the “train” function can be used to look for articles based on the attributes of an article you have currently, but what if, say, I’m following a blog on iPhones and I only care about the rare article they post about Android. Unless there’s an article viable in Newsblur currently about Android in that feed, I can’t use the “train” feature to search for it.

3 Likes
#2

You’re basically describing how to use Focus mode. You thumbs up facets (phrases in titles, authors, tags) and then switch to focus mode to only read those stories. Thumbs down actively hides the story and doesn’t allow you to easily see it, so it’s a different beast. I should really blog next about how to best use the intelligence/training, since you can do some advanced things like thumbs up the feed but thumbs up a few facets to have it split entirely between red and green.

1 Like
#3

I think a detailed blog post on how to take advantage of these features would be welcome.

1 Like
#4

I second Adam’s request. I’m really not seeing the results I was expecting from this feature when I tried it.

#5

Sam, perhaps the green/red count boxes on the top could be clicked on to get a brief view of what is liked/unliked. That way users could verify their ‘training’ is working, or find a story they think is being inadvertently suppressed.

#6

Appreciate this is waaayy after the fact, but in case anyone else is looking, I thought I’d just document explicitly how to get this effect (that is to say, only see articles with a particular characteristic from one of your feeds). Samuel does hint at it above, but this is the full deal:

  1. When reading an article that you do like, click “Train”, identify the characteristic you’re after (e.g. “title:my favourite thing”), and thumbs-up that. This should now mean that you have a few green articles of interested highlighted in the feed, but you might still have hundreds of grey normal articles that you really don’t want to see.
  2. Now, go to your feed index, and Train on the feed. At the bottom of the list of author etc. tags it offers you, there’s a global training button for the entire feed. If you thumbs-down the whole feed this way, it gets rid of everything else in the feed, leaving you only the stuff you’re interested in.
1 Like
Filter feeds with the same title across all feeds
#7

That is true only in the website, in the android app there’s no way to hide these feeds by default and there’s no button to show the hidden feeds.

#8

I would like the same thing. -

#9

I’ll go ahead and add a ticket for this, since it’s something that might get done when I finally hire a backend dev.

#10

Ah, this partly answered some questions I had about this. None of the descriptions nor the FAQ (https://newsblur.com/faq#intelligence) ever really made it clear to me what this would do/how it works.

My biggest questions were/are, is the training per feed or across the board (sounds from this thread like it is per feed), and can you easily turn off/reset all training if you don’t like it (sounds like… maybe?).

Also was wondering if it affects unread count badges (maybe?).

Edit: trying it now… It seems like stuff I vote down immediately disappears from the feed even in “all” mode? That seems more like a form of keyword blocking (really contextual keyword blocking I guess), than “training”. I guess I expected it to work more like a Bayesian spam filter. And I definitely thought it would still show me down voted things in “all” mode…

Edit 2: Looks like “focus” mode only pulls in stuff you explicitly up vote?

hrmmm.

Ok, not what I was expecting. Aside from a Bayesian filter type of deal…

If it is just going on keyword blocking or promotion, there should be a mode that shows only upvoted things, a mode that shows upvoted and neutral things (blocking downvoted), and “all” that shows things including downvoted.

#11

Yes, the training is per feed.
Yes you can easily turn off/reset all training if you don’t like it.
Yes, it affects unread count badges. The disliked articles with disliked words are not even counted as articles, in the android app you can not see the total number but you can see the total number off hidden disliked articles in the web version of newsblur.:blush::+1:

That is exactly how is works and is exactly what I’ve just said…

You are totally right!

Yes THAT’S IT!!