The curation algorithm sits at the heart of News Curator. Its AI-based machine learning pipeline has been significantly reworked and improved. Our algorithms are now better at extracting the actual content of a news story from web pages. And we have carefully evaluated the signals used by the curation algorithm, removing signals that have proven too noisy, and increasing our efforts around reliable ones. The result is a massive increase in overall curation quality while maintaining that element of serendipity that is needed to prevent proverbial tunnel vision.
We have also revisited our search engine and changed the underlying technology stack. The result is state-of-the art search quality. While search is not at the core of News Curator, it is a useful function in case the curation algorithm needs a couple more hints to automatically come up with specific stories in the future.
Next, we have cut back on the amount of content and signals we ingest from social media. This was again driven by signal-to-noise trade-offs. On social media, we now mostly focus on actual posts of the user to passively learn about his interests.
|Overall data flow in News Curator.|
Finally, we have discontinued the automated posting pipeline. There are excellent solutions focused on this particular task, such as Buffer. Admittedly, it was a hard decision to discontinue our posting pipeline. But doing so let's us focus on our key task which is the automated curation of news stories. As a positive side effect, this leads to a simpler, easier to use app.
What do you think about our update? Don't be shy to drop us a comment in the blog, or use the feedback function in the app.