Programming Social Applications

Privacy and Security

One particular challenge with user clustering involves privacy and security, and developers should keep both concepts foremost in their minds when exploring this area.

Simply clustering users does not advance your goal of engaging them. It may be an interesting metrics technique, but in the end you must apply clustering purposefully to maximize user engagement, increase time on site, target advertising, and build revenue. You do this by automatically targeting conversations between clusters, sharing information with certain clusters and not others (much like we do in our real lives). In this way, you protect user privacy and help to build a trust relationship between users and the social sites that they are using.

Let’s look at a practical example of a use case that was a failure point for previous services that endeavored down this path, and which you should avoid at all costs. Let’s say you are using clustering to identify the people a user communicates with the most to give them privileged access to more information about one another, such as feeds from other services they use. This sounds like a good clustering target since these are the people a user has chosen to interact with the most; the logic is fairly sound. The problem here is that relationships (and thus the clusters) change with each interaction (or lack of interaction); for example, a husband and wife (in the family cluster) might go through a particularly bad breakup. The service is now broadcasting ...