Use cases for Google Prediction API

Question

Has anyone here been using the Google Prediction API? For what? Did it "work"?

Answer 1

If you're looking for a real-world case check out this one which automatically assigns priority to bugs.

I've also created a couple test projects to get a feel for how I might use it in production.

One of them suggests tags for Stackoverflow questions. The model for this was trained with questions (from the data dump) that only had 1 tag. This helps it recognize particulars of a specific tag. Code and prose were split into separate features, as one of them may have greater impact on the result. When given a question, it returns the top 10 tag suggestions; actual tags are included for comparison.

Structure was a CSV file:

"tag","code blocks(200 chars)","body text(200 chars)"

Some characters were filtered out of code that caused training errors: []^|~. Not sure which of those was causing problems.

If you'd like to see a certain question added to the test list, let me know. Obviously, questions that have code predict better.

The other project predicts movie ratings based on IMDB data and director/actors. Unlike the tag suggester, this one is live so you can experiment with different combinations to see what it would predict.

Structure here was:

rating,"directorId","actorId actorId actorId"

Both are running on Google App Engine so python is the back end. I'm not using a particular API; just followed some of Nick Johnson's example code.

Answer 2

I have not used this API. But their website lists the below usecases

• 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
• Churn analysis
• Language identification