From the Directed Edge Developer Base
Revision as of 12:52, 25 November 2009 by Scott (talk | contribs)
Jump to: navigation, search


Directed Edge makes integrating with our recommendations engine easy with {{{1}}}. We provide bindings that handle all of the communication with our server transparently using normal {{{1}}} objects.

Getting started

  • Introduction to Recommendations is a good starting point if you're wondering how recommendations work or what they're useful for.
  • API Concepts explains some of the basics of hour our API works and introduces the concepts of items, tags and links, also explained briefly below.
  • Grab the {{{1}}} bindings from GitHub and copy the file named {{{2}}} into your project.

Data modeling

Items and links

To model the data from your site, you'll need to figure out what your items are. Usually they're things like users, products and articles. We represent a relationship between items by links. So, if you have Bob Dylan's "Blonde on Blonde" that you want to say was bought by "John Doe", you create a link from "John Doe" to "Blonde on Blonde".


Usually we don't need to actually know the names of those items — they just need a unique identifier. Typically that's something like customer1 and product1. Most people just use the ID field from their own database. So if you have a MySQL table named products and Blonde on Blonde is at the row with ID 42 then you'd just use product42 as your identifier for that product.


Items also have tags, so John Doe would probably have a user tag and Blonde on Blonde would have a product tag. Tags don't weigh into the ranking at all — they're just used so that you can filter the sort of results you'd like to get. So if you want to show related products, you'd run a query looking for things with the tag product.

Querying for recommendations

We offer two primary sorts of recommendations, called related and recommended:


This is for example if you have one product and you want to find similar products, or a user and want to find similar users. To find products related to Blonde on Blonde you'd send a related query for that item, looking for things tagged product.


Recommended is for doing personalized recommendations, e.g. Products Bob is likely to be interested in... To do a recommended query, you would send a request for the item that corresponds to Bob looking for things that are tagged product.