I mostly did intelligence training when I started using Newsblur 10 (?) years ago.
Now, especially with the new features, I’d like to improve my setup, but I have no idea what training input I gave in the past. How can I check and correct this?
You can do this with the API if you are familiar working with the command line, I do it now and then with my feeds to check the training I have and if I can remove any (eg I may have trained to ignore CES2024 in title but that’s now long over so pointless being there).
I can share what I did if you are familiar with doing this sort of thing. It’s is pretty crude but it works for what I needed.
Share please! Esp if it’s something I can turn into a feature for everybody. I think a global view of training would be an amazing feature.
So this is what I have … bear in mind it was just what I needed and not that generic
DATE=$(date +'%Y%m%d')
curl --cookie-jar cookie -d username="<yourNBusername>" -d password="<yourNBpassword>" https://www.newsblur.com/api/login
curl --silent --cookie cookie https://www.newsblur.com/reader/feeds -o $DATE-feeds.json
jq -r '.feeds[] | [.id, .feed_title, .feed_address] | @csv' $DATE-feeds.json > $DATE-feeds.csv
cat $DATE-feeds.csv | awk -F',' '{print $1}' > $DATE-feed_ids.txt && echo "Number of Feeds: `wc -l $DATE-feed_ids.txt`"
while read id; do echo "----- $id -----"; curl --silent --cookie cookie https://www.newsblur.com/classifier/$id; printf "\n"; done < $DATE-feed_ids.txt | tee $DATE-training.txt
its pretty basic and you end up with a few files that show your training for that feed, I was gonna make it prettier but never got round to it, I was doing it steps as well so some could be consolidated.
20260113-training.txt I'd then have
----- 1778 -----
{"code": 0, "payload": {"feeds": {}, "authors": {}, "titles": {"Android": -1, "Beta": -1, "Long": -1}, "tags": {}, "texts": {}}, "result": "ok", "authenticated": true, "user_id": myuserid}
then I can look up in 20260113-feeds.csv
1778,"Google Chrome Releases","https://feeds.feedburner.com/GoogleChromeReleases"
maybe @samuelclay you could clean it up and it could be somewhere in your settings to show the training and could do more with it from there
You can review your past training by checking the Intelligence Trainer settings inside NewsBlur (usually under story settings or site preferences). There you can see what you’ve liked, disliked, or trained over time — and adjust tags, authors, or titles to better match your current interests.
Just like Agri-Systems optimizes performance through precise processes such as grain conditioning, refining your training inputs helps improve long-term results and overall system efficiency.
1 Like