Train based on regex

idea
1

For example, I would like to hide stories about the gold market, not Goldman Sachs.

image

1 Like
2

I’m considering it but regex are expensive to run on every story. Would you consider upgrading to an upcoming NewsBlur Premium Pro that was $10/month to gain access to regex, site-wide saved searches + notifications, and AI site summaries?

2 Likes
3

Would upgrade to Premium Pro for $10 a month in a flash for regex.

1 Like
4

It’s not a foolproof solution, but you can :+1: “Goldman” and that will take precedence over :-1: “gold”. However, then your focus list is flooded with “Goldman” articles.

5

Another “quick and dirty” solution would be to thumb down "gold " (with an empty space behind) or “gold,”, “gold.”, etc.
AFAIK you can actually modify the text you’re training so in theory, even in the “goldman” example, you could write “gold.” and thumb down that.

6

I would absolutely pay if newsblur training implemented regex. I would pay more for functionality approaching that in gnus’s scoring. Scoring (Gnus Manual)

7

I signed up just to say yes – sort of. More like, AI site/feed/article summaries and the ability to train text from the body are what sounded good to me. Fiddling with regex is partly why I subscribed here, I was self-hosting miniflux for the last few years. The “training” mechanism here is much more user-friendly, but seems limited to title, author and tags.

For example, ZeroHedge is a good source of behind the scenes financial news. (That enormous jobs revision that landed recently? They predicted that based on the same underlying data months earlier) Outside of that, the signal to noise ratio there is kind of lousy for me personally, and unfortunately they don’t tag their articles well and the true author is typically put in the body text. Being able to filter text from the body would do a lot to clean up that feed.

There is an open github ticket to allow training text from the body from 2020 that was high priority, did that get put on the back burner?