You ever wonder how Google is able to figure out just what the heck you're talking about when you search for "blue album cover" and you're not talking about the first Weezer album? Google is able to use your previous search, as well as a plethora of other sources, to predict what it is you are most likely trying to find. Pretty cool stuff, once you get past how uncomfortable that may be.
Well, you can have the predictive powers of Google too. Back in March, Google launched its Prediction API. The Google Prediction API aims to help developers make their applications more intelligent. For example, instead of just using fixed ratings like "Like" and "Dislike" to learn a user's taste, the Google Prediction API can help you offer more granular options that can be automatically determined based on how you are integrating the API.
Using the Google Prediction API is somewhat of a snap but may seem a little awkward for developers that are used to simply running a cURL command on an API method URL. Essentially, the process involves uploading a CSV of 'training data' and then telling Google to 'learn' what it can. From that point on, you're back in a commonly found RESTful situation with a friendly JSON response.
Recently, Google has added a few exciting new features to the Prediction API. You can now do predictions across multiple categories, continuous output, allowing for finer tuned range of variable settings, and mixed inputs, allowing you to send numeric and text inputs. You can even combine mixed inputs with continuous inputs to build even smarter, predictive automation tools.
From there, the possibilities are tremendous. Personally, I would love to see some location game mashups to recommend me places I might like to check out. Maybe Twitter can use it to recommend I follow people that I haven't already followed and un-followed. Outside of recommendation systems, more intelligent in-app searches would be welcome, as well.