A month ago, we released the first iteration of the our Interest Dashboard.  This is a very ambitious project, interpreting and representing back to the user an analysis of their tasks and interests.  Many online services already make some representation of this back to a user, (think of Amazon’s suggestions) but I cannot think of anyone who does this with the broad scope that the Firefox Interest Dashboard does, and at the control of the user.  Rather than something like Amazon suggestions, when I looked at my Interest Dashboard, I was more reminded of a scene from You are what you eat where I was confronted with my intake of the Web.  

Anyway: it’s often hard to understand the scope of the project and the current and planned features without the full product vision. And there were a couple of misconceptions that we evidently helped create last week which I would like to clear up.

First: we stated in our FAQ that data is stored in the client.  That is the case.  A few commentators noted that the Privacy Notice states that you may (with your express consent) share your data with other websites.  The FAQ is correct, the data is stored in the client because this functionality is not yet enabled in the add-on.  At such time as we have the functionality available, we will update the FAQ and sure sure to notify users of the add-on.  We felt the best approach was to be transparent with the privacy notice about the direction we will go, but be very clear with the current FAQ (we felt this was a better approach than updating the privacy notice with the release).   We do expect the user to be able to share their intention with Websites in the future, and as we make clear, this will be with the user’s express consent.  That is in the future.

There were two other misapprehensions about the launch last week which I would like to clear up.  I read at least one comment where a user stated they had cleared their browser’s cache but still saw data in the Interest Dashboard.  This is fairly straightforward: the dashboard is constructed from the browser’s history, not what is in cache.  Also, one commentator noted that some reviews had been removed from AMO (some are preserved here).  This was actually news to us too, and on checking in with the AMO team, we understood these reviews were removed by members of the AMO community for violating their guidelines for relevance.  

The last point that I read in the discussion was about whether or not Interest Dashboard would exacerbate the Filter Bubble problem.  It’s an excellent point, and one we think about quite a bit. In fact, I believe we can help address the issue. The Web should not seduce the user into bad or repetitious habits. A first step here (just as in You Are What You Eat) is to be presented with what you consume (“What gets measured, gets managed”).

But this raises the principle question we’d like to answer through this experimental add-on: whether it’s possible to have a user-centered recommendation system. I use the word “recommendation” loosely because not everything we get recommended online needs to be in a form of an advertisement. We intend to build this recommendation system by first providing users transparency into how an interest categorization works - this is the current state of the Interest Dashboard. We are also intensely focused in providing in-depth analysis of the user’s long-term interests and dynamic short-term interests. Most recommendation systems, including ads, only focus on the short-term intent of the user in the hope of converting the user down the purchase funnel.

Shortly, we hope to connect the user’s long and short-term interests, with their explicit consent, to a piece of content that they might find interesting. The connection part can be done several different ways: 1) use content crawler and index them by categories or 2) partner up with an aggregation service. In the case of the latter, we can either get a daily massive dump of indexed content or use an API to send the user’s interest category that will return a matching piece of content. We are internally experimenting with these options before we release it as a feature on the add-on. That’s a long way to explain a part of the experimental feature is reflected on the privacy notice.

A content recommendation system is certainly not for everyone and there are some bad examples of how companies can abuse personal data. We want to counter the bad actors by providing choice.

If the user wants it, we can provide a platform based on what/how/when the user wants. We’re hoping the Interest Dashboard is the beginning of this exploration. But this is a sensitive area.  Ideas and advice are always welcomed, as is scrutiny.  We are serious about what we are building, and the only way it will be adopted is if it is trusted, and it will only be trusted if it is scrutinized.  Please, keep the scrutiny coming.

http://kevinghim.tumblr.com/post/104767505316