Toffer Winslow: We take a knowledge of both user interests -- and by that we mean behavioral patterns that are fairly stable over a long time -- as well as user intentions, which are much more volatile. A user's intentions vary substantially from session to session. And so we take those inputs and based on a user's particular behavior in a given session will discern both their intent and look at users who have demonstrated similar intents and what they ultimately tried to do in their sessions. We will also in parallel look at that individual user as longer terms interests, their more stable behavioral profile. We marry those two together as a way to push the kinds of recommendations to the user they would be interested in.
BI: What are the differences between serving recommendation on the Web and onto a set top box?
Winslow: The Web world is a pretty standard tried and true approach to implementation. You take an existing Web applications, put tags or beacons on those Web pages that may fire off events about both user behavior and about the catalog the user is interacting with back to our system. Over a period of time, often as short as a week and sometime as long as a month, our system ingests and processes a sufficient amount of data so it can start deriving actionable insights from that data. They are delivered back in the form of recommendations to the user, which are implemented on the specific pages they are on. In the set top box world of television there are more modern architectures now that provide a comparable amount of data about users. A lot of the IP based TV networks out there right now similar to the ones that Verizon and AT&T are rolling out are able to provide a wealth of insight about what TV viewers are doing, what shows they are clicking on. It allows for certain interactivity, to even rate shows in some cases or to download certain VOD titles. All of that behavioral data can be fed back into our analytical engine and transformed into content recommendations that the user would enjoy seeing on TV.
BI: What stage are you at in the set top box world? Doing tests?
Winslow: We have a couple of customers who have made significant investments in our technology and are rolling out set top box based implementations later this year and early next. We were with Microsoft at the Consumer Electronics Show early in January where we were demonstrating a prototype of our recommendation engine being plugged into the Microsoft Media Room IPTV platform. We are well past the prototyping and beta stage and into heavy implementation stage with a couple of big TV providers both in the IPTV world and the direct satellite broadcasting world.
BI: What does it look like from the viewer's perspective?
Winslow: You can offer a tabbed browsing or viewing interface, or at least a whole bunch of selectable options. One of those options would be My Recommendations. You would click on that and see a list of things being presented to you that either are on now or can be saved to DVR for future viewing.
BI: Are there specific and different privacy issues for the set top box?
Winslow: Typically no, we find. Most companies have privacy policies with their users that allow them to collect data from them whether is it is data they acquire in interaction with a Web application or via the set top box. In some cases there are subtleties around whether a user needs to explicitly opt in or if they are assumed to be part of personalized service unless they opt out. By and large the two media are comparable. At the end of the day our clients have the right to use the data they collect from their customers interaction with their services, be it a Web application or a TV service, and use that data to refine the offerings they are giving them.
BI: Any fundamental difference between erecting a recommendation engine or its use on a set top box as opposed to the way they see it online?
Winslow: You have to be sensitive to the unique qualities of the user interface in both. The Web is an interface, [so] people have become much more accustomed to having many options to select on a given page. If you just think about your portal home page you probably have a hundred links on that page to select from and you have a user experience with a keyboard and a point device that enables you to move very efficiently through those hundreds of possible selections. TV is a different animal. Historically it hasn't been nearly as interactive as the Web. While the new generation of TV interfaces from the likes of Verizon or AT&T are much more interactive, you don't have the same precision or capabilities as a computer yet.
BI: I imagine one of the Web behaviors one would expect for a recommendations engine on a set top box is more sampling of content. Can we get features like that yet? Once you start initiating Web-like behaviors on the set top box it almost becomes necessary to pile on other features users come to expect online.
Winslow: Yeah and I think that is fair. To be sure, most TV services aren't there yet. I am a Comcast subscriber and pretty infrequently do I have the opportunity to watch a real trailer for something I am considering spending $6.99 of my money on. Online that typically would be standard practice. That said, as these TV architectures become more flexible, then the differences between the Web-based experience and the set-top-based experience will start to decrease over time.