Curated side-bars on Facebook and LinkedIn -> NLP

The other day, I was browsing LinkedIn casually and noticed that there was a section titled What people are talking about now on the side bar of my home-page. Maybe it was there before but maybe it wasn’t.

What people are talking about now on LinkedIn

Clicking into it, got me a curated set of posts by various organizations as well as individuals talking about the specific topic that I clicked into. In addition to it, there was a well crafted summary of the topic itself.

Now, curated side-bars aren’t new. Facebook has had the trending sidebar on the right-hand side of the newsfeed for a while. But I’ve never clicked on the any of the trending topics on Facebook ever.

The trending section on Facebook that I’ve never clicked

What made me click on the sidebar in LinkedIn but not on Facebook? Really what enticed me to click on the side-bar in LinkedIn were two things:

  • More descriptive titles
  • Larger dedicated screen real-estate

How are these titles generated? Is there something fancy going on behind the scenes? Turns out that the generation process and curation is pretty simplified today. LinkedIn says the following about how the topics are selected and surfaced: These topics are selected and curated by our news editors, leveraging data on and off LinkedIn. Similarly, Facebook’s process today for the generation is based off of data, but are still curated manually by a team.

A potential for NLP and automation: It is not far fetched to imagine the topic selection and the curation process to be automated in the near future, especially the push on ML/AI at these companies. First, there is the classification of posts/articles into topics, the quantification of the magnitude of activity on individual topics and the display to users based on relevance. Secondly the generation of titles can be optimized and tested in real-time based on user engagement data (the testing across different options might already be done today).

In other words, this would really just be a recommendation system that is more NLP based and based on the semantic understanding of the topics of unstructured data around text, images and videos in posts and articles, all of which are continually getting better.