Pocket Digest: Your personalized feed without information overload

This is a mail I sent to Pocket earlier, about some weird idea about Pocket. Sadlly, I didn’t receieve any response, but I still think it’s a weird idea worth sharing.

Screenshot:

demo: https://productchaseapp.herokuapp.com/pocket_digest

Before I get into the details of the demo, I want to clarify that the one of visions of Pocket that I understand, is to make article easier to read (and thus help people to read).

Now let’s discuss the problems related to use cases of Pocket, and their solutions.

1. Problems and Solutions

Problem #1: 
In general, blogs don’t have good content every post. As a reader, I’m more interested in its good content and don’t care about its filler. This is one of the typical problems in RSS reader, that when you have many subscriptions, your feed’s quality drops considerably. There are too many stories to be read, yet the gems are buried.

Solution #1:
However, with the Pocket data, we can monitor the popularity/relative value of an article from its blog. And we can tell if an article is of the 30% good article it has produced (i.e., peak over threshold method [1]). This means that we can use the Pocket to filter the feed, and only read the good part of them.

And this further means that I can subscribe much more blogs than before, yet read more valuable content and use less time.

So this is the feature #1: user can use Pocket to follow the topic/blogs they like. Somewhat like an enhanced version of RSS reader.

Problem #2:
The most content I mark with Pocket is long posts. And they are hard to read. How can we make it easier to read?

Solution #2:
When people do read an article, they can highlight the sentences they like most (this is already being popularized by Medium, and recently being launched by Instapaper).

So it would be great if we can show the highlights to others, and thus help people to quickly find the valuable points in a post.

And we can take this a step further, which is to put the top highlight at the beginning of the blog to help people to grasp the key points. Something like the highlights in ScienceDirect, Elsevier.

And this is feature #2: Pocket can display the top highlights to other people to help them understand the post at start.And I think this fits the vision of Pocket better (help people to read more easily).

The #1 feature is something I have personally used a while ago, as you can see in the link. It’s not mainstream, but it’s a kinda lovely pet product for me.

The #2 feature is possible to lay a foundational dataset for AI. Given the input (article), we have the output (highlights). This is the basic of supervised learning. Will us create an AI to create summary of articles for us? This would be too early to talk about it, but this can be strategic thing.

Combing these 2 features together, I call the demo Pocket Digest, because it can save us much time reading things, and help us read better.

Finally, as a PM-to-be, I always want to launch something that can increase the user base (attract new users). Theoretically, by giving a slight different value proposition (Pocket Digest), I think it might be possible, but I’m not very experienced in this kind thing.

That’s about it. Thanks for spending time reading my rambling, and it would be great if you can give any feedback/criticism on this hypothetical product management exercise. And I’, sorry that I only find the business mail that I can send this post.

All the best to the Pocket team.

[1] Peak over threshold method for Pocket data

The Pocket data is how many people saved an article to read later. It can be seen as a measure of how good the content is (otherwise people simply won’t pocket it). From my personal experience, it’s much better than the sharing stats on Twitter/Facebook. One possible reason is that serious content does not get shared as much as jokes in social network.

The core assumptions for using the Peak Over Threshold method (POT) are:

  1. As a consumer, we do not care about every post a blog produced. We only care about the good content.
  2. Most blogs do not have good content ever post.

So for a given blog, its posts and Pocket data would look like this:

It’s possible that the post with high Pocket data would interest us most. Or put it in another way, the post quality follows a Power-law distribution, and we’re interest in teh tails rather than the head.

Still, this is just a very rough model. We can certainly mix it with recommendation algorithim. Also, to compare posts from different sources, we can use percentile of the Pocket data, rather than the absolute value. Other wise the blogs with most popularity would simple overshadow the others.