It's not often that I get to write a blog post titled like a Disney movie. Anyhow, you'll recall that I've got together with the smart folk at Adaptive Lab to build a curation app for Twitter that we're calling Fraggl. The vote on whether the problem we are trying to solve is a valid one was pretty resounding with 81 votes for and 5 against, so that's good to know (thanks for the feedback).
We're now moving forward to think about how the app we build might work. Fundamental to this, as I mentioned before, will be the right combination of the key forms of curation: people (or professional curation in another context), algorithmic, and social (from the community). Our thinking is that the quality of the UX depends a lot on whether we can structure these forms of curation and then deliver or enable access to the results in the right way. We'll deal with delivery and access in a subsequent post but for now we're starting with the former.
Discussing this with the team at Adaptive Lab, we've arrived at a taxonomy of factors that can be used to determine the results:
If we were to flip this taxonomy, and think of it as our 'magic pyramid' (so to speak), it would look something like this:
The base layer of this is considering the sources of data that we are going to focus on, and those which we might apply the algorithm to. The options here vary from the widest possible context (all tweets), to just focusing on the people that I follow (as some existing services do), to taking a more focused approach and (API allowing) use Twitter lists. Part of the feedback from the initial post was around how we might incorporate the right mix of popularity and serendipity, which is something I'd also been thinking about. Another way of thinking about this is the mix between content which is popular amongst the people I follow, and content which I don't know I like but may find interesting. An early thought about this is that (API allowing) using Twitter lists could be useful here because users could plug the algorithm into their own Twitter lists, and/or they can use it to source content from pre-curated Twitter lists of people that share great content from certain areas (e.g. architecture, design, content strategy and so on).
The second layer to think about is how we construct that algorithm and the signals we will use to get us to the best content. Here we have options that range from the number of RTs a particular tweet/link has got, the number of people that have favourited a tweet, but we also have factors that might be interesting to look at such as how many followers the people doing the sharing have and so on.
Our final layer is personalisation. I'm a fan of using data (typically consumption or voting up/down) to inform content mix so that apps get smarter and more personalised over time (aka scrobbling), so it would be good to include some degree of that if we could. It's a bit like what Zite does, although the fact that after 18 months Zite has arguably got a bit too specific for me (i.e. it's become too narrow in its content mix and could benefit from widening that out slightly) is quite instructive. Any personalisation we enable therefore, needs to add enough to the mix that it enhances the curation, without detracting from the breadth of content and value of more serendipitous stuff. Perhaps a simple 'show me more of this' and 'show me less of that' is the way to start here.
We're highly unlikey to get the balance of factors absolutely right on the first go so it's inevitable that we will need to iterate. So if you want to join the beta testing group for Fraggl and get early access, you can sign up here.
In the meantime, it would be really useful to gather some feedback on this, and on the kind of curation tools and techniques that you find useful. So, a couple of key areas:
How do you currently identify valuable content? Which tools and strategies do you use?
What's good, and what's not so good about these tools and strategies?
As always, your feedback is appreciated.