Inference

As you might know, we have been working on sorting posts by relevance in Cadmus. In addition to filtering out similar posts in a user’s stream, we feel this is a great way to keep up with all the social media services and yet stay productive. But for the user, what we really wanted Cadmus to do was answer a simple question, “What did I miss since the last time I checked into these services?”.

And so, over the last couple of days, we have been testing out the latest version of Cadmus that does exactly that.

Not your average “sort by relevance”

There are several services out there that sort your stream by relevance. But often I find that when I sort my posts by relevance, a lot of old posts stay on top of my stream. Cadmus on the other hand approaches this problem in a unique manner. At each login, we take all the new posts since your last login and we find the most important ones. This way you won’t run into the problem of old posts showing up at the top of your stream.

The relevance algorithm itself leverages what we believe is going to be absolutely critical over the next few years; adding context to posts. We track your relationships with your friends in your social graph and use this to figure out how important the source of the content is. By tracking relationships, we can transform your simple social graph to a more weighted structure that evolves over time. This is incredibly important because the underlying social graph might not change over time, but the weighting certainly will.

We also check to see if the post itself is relevant to you. So, not only do we figure out how important the source of a post is, but we also determine how important the post is. The results of this is a Cadmus page that shows you the most important conversations since your last login.

Let’s look at an example. Below is a screenshot of my Cadmus homepage. As you can see the top post has six related posts. This means six other people are talking about the same thing. So just by looking at these first few posts I know what all my friends are talking about. I don’t have to scroll endlessly to find them; Cadmus picks up on these trends and brings them to me.

Why is this important?

The problem we are trying to solve here is that we don’t have all day to stay on top of these social media services. Most people check into these services ever so frequently. And every time they do, they have to scroll through pages of posts to find out what’s going on. From the statistics that we have collected so far, we have found that nearly 20% of all the posts in a user’s stream are similar. This means that every time you login, you have to scroll through pages of posts with low signal to noise ratios.

Cadmus fixes this problem by filtering out the similar posts and finding the relevant ones since your last login. This way you won’t feel like you are missing out on what your friends are talking about.

We are still working on improving the algorithm and handling specific cases to improve the user experience. Give Cadmus a try and let us know how we are doing with the features detailed above.

Thanks to @mliao for reading over this post.