10

http://code.google.com/apis/predict/

From google:

What is the Google Prediction API?
The Prediction API enables you to make your smart apps even smarter. The API accesses Google's machine learning algorithms to analyze your historic data and predict likely future outcomes. Using the Google Prediction API, you can build the following intelligence into your applications:

  • Recommendation systems (demo code)
  • Spam detection (demo code)
  • Customer sentiment analysis
  • Upsell opportunity analysis
  • Message routing decisions
  • Diagnostics
  • Document and email classification
  • Suspicious activity identification <--- I'm particularly interested in this one
  • Churn analysis
  • Language identification
  • And much more...
6
  • The book, "The Cloud Computing Bible" makes a reference to it, but I don't have a copy of that title in my collection, nor have I done any research on its writers or whatever resources they draw from. You can check there to better answer your question.
    – atdre
    Apr 14, 2011 at 23:05
  • What I have seen, though, are certain people's GMail accounts having been owned, including by Google employees in the past -- and in an unethical, perhaps illegal depending on your world location, manner. Even GApps accounts have been, obviously.
    – atdre
    Apr 14, 2011 at 23:09
  • @atdre - can you expand on that or give some examples?
    – nealmcb
    Apr 15, 2011 at 0:05
  • @nealmcb; I try to in my answer below. Kind of a difficult subject as the other answerer pointed out.
    – atdre
    Apr 15, 2011 at 4:30
  • @atdre I was asking about your claim that google employees unethically and perhaps illegally owned other people's accounts. Your answer below links to a story that simply notes that google notices where you log in to google accounts from, and alerts you (like credit card companies do) if the pattern of IP addrs seems out of character for your account. It wouldn't require anything other than a normal session cookie to support that, right?
    – nealmcb
    Apr 15, 2011 at 15:46

2 Answers 2

3

If I were to make a fairly estimated guess, I'd say it was this suspicious login notification thing.

It's not as complicated as you initially portray. At best, it's like Evercookie, Panopticlick, and Quova mixed up (at worst, it's just squid-imposter bundled into Metasploit with instructions). It's going to tell you when authenticated users using the same credentials are in two place at once, or when ones sourced from dangerous countries, or other extenuating, potentially predictable situation.

3

Thanks for the question. Interesting API. See also Use cases for Google Prediction API - Stack Overflow

As far as I can see, Google gives no information on the machine learning techniques they offer. They just say they have lots of algorithms, and use the best one for you data. Not even a hint of whether they may change in mid-stream without telling you.

You would need a good training database tagged with information about what activities were actually malicious vs what activities were benign. Perhaps we should start by brainstorming what data sets might match their model. Log files, form entry data, etc. I guess. But who has big enough well-tagged datasets relevant to their operation?

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .