An open source client-side algorithmically-driven RSS reader built at the Suffolk LIT Lab. Use 👍 & 👎 + Seen or Open to feed the algo. Refresh this page to update your card order. Searches, votes, & settings don't leave your device. Customize your algo and timeline behavior in Settings. Curate sources in Feeds—starter lists available. Add feeds with the blue plus button. For project updates, follow #SuffolkRSSalgo on Mastodon, or visit our GitHub page.
You can manage your feeds by clicking Feeds at the top of this page. See additional controls below. To determine what articles you like, we make use of insights from TF-IDF, comparing all the text from the cards you vote up or down with everything you've "seen." I hope to write a more complete explanation sometime. When I do, I'll put a link here. Until then, you can always look at the code.
Note: when figuring out what you "like" we look only at the card's domain, title, description, and the RSS Feed's name. We don't look at the content of the article. We convert all feed names into hypenated words. That's why you might see things like nyt-top-stories (NYT > Top Stories) in the word lists below. Also, we do something similar for an article's domain (e.g. nytimes-com).
You don't need an account, nor do you need to provide any personally identifying information, to use My RSS Algo. Any searches or votes you make are stored only in your browser as part of localstorage. The same is true for any other persistent data (e.g., your filters and the choices you make below). Presumably GitHub (which hosts this site) keeps some log files. When fetching your feeds we make use of a couple of proxy servers, one run by the LIT Lab and corsproxy.io. Corsproxy claims not to keep log files, and the Lab only keeps recent ones. So in principle we could tie your IP address to the feeds you download. The publishers of these feeds, however, only see the IP addresses of the proxy servers when they are downloaded. We do, however, embed preview images and a site's favicon as part of most article cards. So, unless you have the "Exclude preview images and favicons..." option checked below, simply loading your timeline will download those files, likely producing log files on the servers they came from. And of course, if you follow a link to an article, you'll access that article and set in motion all that comes with that. Also, if you check the "Show local weather" option below, your browser will share your location with Open-Meteo's weather API.
There's no need to download every feed every time you hit refresh. So this site waits a defined period before downloading content. If you refresh the page before the end of this cooldown period, it will recalculate ratings based on recent votes and filter what it displays based on the other options here, but it won't try to download your feeds. Welcome to the finite scoll. Currently the cooldown period is set to 0.25 hours. You can change this below, but you must refresh the page to see results. If you set a cooldown period that is shorter than the current value, you will have to wait for the old period to pass before the new one takes effect. So, once you set it and refresh, there's no quickly going back.
The options you choose here will effect this site's performance. Generally speaking, the more inclusive your choices the larger your timeline will be and the slower it will load, reorder, and the like. Of course, the number of feeds you follow will also influence this. The trick is finding the right balance for you and your device.
You can hide/group articles that are "too" similar to each other. To measure similarity we use vectorizations of cards based on TF-IDF. Leave the cosine similarity cutoff at 1 to keep all articles. Setting the cutoff to something lower will hide all but one of the acticles with similarity greater than the cutoff. This will display a list of "hidden" articles at the bottom of the top-most rated article card for that group. Note: this will slow down sorting of your articles, esp. for larger timelines. Also, all items in a group other than the top-most are marked as "read" to hide them from your timeline. This cannot be undone! Cosine Similarity for cutoff: 1. Refresh page to see results.
After the recommendations kick in, each of your cards is given a rating which is used to order the above list. We also truncate the list based on the following cutoff value. We use the z-score of the rating. The z-score of a value is just the number of standard deviations it is away from the mean (average). So the average score of your cards is always 0. Negative numbers are below average, and positive numbers are above average. Z-Score for cutoff: -2. Refresh page to see results.
When calculating ratings, how much influence should down votes have compared to up votes? If you choose 0, they don't count at all. If you choose 1, they count equally. If you choose 2, they'll count twice as much as up votes. Why not just set this to 1? Well, people don't seem to respond equally to positive and negative experience. See negativity bias. Current value: 1.25. Refresh page to see results.
Here are your most-positive words and their approximate TF-IDF, treating all up-voted cards as a single document: not enough data
Here are your most-negative words and their approximate TF-IDF, treating all down-voted cards as a single document: not enough data
Like Barbie and Openhimer news? Try:,
Dislike Barbie and Openhimer news? Try:,
If you don't have a Pocket plugin installed, you can use the Pocket button in the upper-right corner of each card to save items to Pocket. This opens a new window. If you're already signed into Pocket you can have these windows close themselves after 1 second. Depending on your device, this may or may not be enough time for it to save. YMMV! If you aren't already signed in they will open and close without saving. So, only check the box below if you are signed in.
Since users of Mastodon use different servers/instances, we need to collect that information to enable sharing (i.e., so we can queue your post on the right server). If you'd like to use this feature, enter your server here:
Display content from the last 7 days. Refresh page to see results.
You can limit how much you see in your timeline by adjusting the slider above. If you clear you reading history, however, it will remove older articles, not just hide them. This will effect your recommendations since they are based on your interactions with old articles (did you vote them up or down).
FWIW, you can download the data stored in your browser to a file. This is helpful for backups and if you want to move your history et al. between devices.