New Tab Site Suggestions

Over the past weeks, we’ve talked about ways to analyze data in Firefox and to share that data out of the browser while keeping users in control. Today we’re putting those pieces together with a proof-of-concept: Site Suggest. This simple add-on locally analyzes your browsing history to pick an ODP category that you seem to have interest in then makes a secure and privacy-protected remote request for a site suggestion to then display on your new tab page.

A site suggestion with why it was suggested

Conceptually, the flow is much like the add-on update mechanism in Firefox. Users install add-ons, then Firefox makes a request to see if any of those add-ons have updates, and then updated add-ons can change the behavior of Firefox. From a high level, here are the similarities of Site Suggest and add-on updates:

  1. The user creates activity in Firefox (browsing pages vs installing add-ons)
  2. Firefox analyzes that behavior (finding ODP categories vs listing add-ons)
  3. Firefox securely sends the data (sent to Site Suggest vs add-ons servers)
  4. Firefox updates with the response (new tab content vs add-on functionality)

One thing we want to be explicitly clear is that the ODP category sent to the server is not tracked with the user. The site suggestion server does not use cookies or any way to identify the request to then use previous requests to personalize future requests. This simulates the terms of use on the user’s data where the server is only allowed to use the information in the request for that one response. This also means that if your interests change, the server is always using the freshest data to personalize the site suggestion.

For this proof-of-concept, we didn’t bother creating a large number of suggestions, so we don’t expect people to get high quality recommendations. In fact, we probably could have put all the suggestions in the add-on, and the site suggestion on new tab experience would be exactly the same. For many users, there isn’t a distinction between local suggestions and remote suggestions, and these users trust Firefox to do the right thing.

But by having this secure and privacy-protected remote request, the quality of suggestions can improve without requiring a Firefox or add-on update. And this is especially important as the number of potential suggestions increase, so instead of having millions of unused suggestions built into Firefox that can become outdated, Firefox can focus on just the newest suggestions the user cares about.

We would like your input on what you think about this privacy-protected flow of data that improves users’ Firefox experience. So install Site Suggest then open a new tab, and see if that triggers any ideas for you. As always, you can check the source on Github, provide feedback, and submit issues or suggestions!

Ed Lee on behalf of the Prospector team

4 responses

  1. martin wrote on :

    The link to this blog entry on the addon page (“About this Add-on”) is broken. Clearly it’s not 2013, but 2012. 😀

    1. Edward Lee wrote on :

      Nice catch. I’ve fixed the link on AMO. Thanks!

  2. Caspy7 wrote on :

    If we’re going to do client-side analysis why don’t we use the current time + visit frequency together to anticipate likely destinations?

    Joe gets home from work around 5:15pm every day. After arriving he usually checks a particular news site and web comic.
    Maybe he checks the news site frequently so it already has a high frecency rating, but the humor site is a once a day web comic so it doesn’t have as high a rating. But we can easily determine that on weekdays, during the 5 o’clock hour Joe is very likely to visit that comic. So when he opens a new tab it is suggested at the top of the list.

    Joe spends a good chunk of time on work related sites, but after work they’re the last things he wants to think about, so their lack of visits after 5 actually downgrade their suggestion priority during that time period.
    He also checks the weather before going to bed (around 10:30) and a sports website on Saturday mornings.

    These are all patterns we should be able to analyze & anticipate. And just like with the location bar, patterns are updated as new ones arrive (e.g. students during the school year vs the summer).

    Actually, now that I think about it, this behavior may also be applicable in the location bar too.

    1. Edward Lee wrote on :

      Absolutely. There are many sources of information within Firefox that can help understand the user’s context. An old Prospector experiment looked into displaying suggested links to open tabs based on the pages that you frequently open at a certain time on certain days. One of the original use cases was to open up wiki meeting pages in preparation for a meeting — very similar to your example of opening news or comics at a regular time on week days.

      If Firefox understood your context, which is even more important on portable computers such as laptops and especially smartphones, there are so many things Firefox could do to improve your experience. And potentially even more interesting is if that type of information could be used on web sites in a privacy protected manner.