Affinity-based browsing, the way I envision it, goes beyond the shallow targeting of Netflix, Amazon, or, from what I can tell, Glue. These programs base recommendations on, for example, people who viewed Ice Age also viewed Cars, so the system recommends Cars. Stumbleupon seems to address general similarities, but doesn’t approach the type of targeting I’m suggesting, because its “classification” and “clustering” engines are primarily based on subject matter and lifestyle interests, which is a start, but isn’t enough to truly serve targeted information. Stumbleupon also collects sex and age information, but it doesn’t know other demographic information, such as where I live or my education level. Thus, the groups I’ll be placed in will be very broad, even though organized around a particular interest.

I’d like more variables in the formula. Advertisers seem to know a lot about us, since ads often seem to be well-targeted (feel free to test this claim by entering any well-advertised pharmaceutical product into your search engine, or use “diet” as a keyword and then pay attention to the ads you’re served). Google Adwords, for example, definitely picks up text from Gmail to serve ads, as does Facebook Connect. Both have deep demographic information about us. So, instead of the Amazon/Netflix model, where recommendations are based on a single selection (e.g., users who purchased the APA Style Guide also bought a similarly boring book titled “x”) or the Stumbleupon example, where recommendations do combine variables, but still remain relatively broad (e.g., 27 year old males interested in entertainment and philosophy found these websites interesting), I’m picturing a formula like the following:

A person who (1) drives a Volkswagen + (2) owns a Mac + (3) shops at Trader Joe’s will like the following sites based on users with similar sociographic patterns. But wait, this person’s browsing behaviors indicate she spends time on teaching websites and downloads educational research articles…so, now the affinity group gets smaller and likely better defined. Combined with demographic information Google and others collect, when this person conducts a search on laptops & classrooms, the results could be better tailored to individual interest (e.g., instead of showing results for laptops for sale, Edutopia and other relevant educational sites would appear in the top 5). If a menu were provided where the user could further define their interests (Stumbleupon has the beginnings of this, but we need more than subject/lifestyle preferences), we could really select how our information is served.

[Special thanks to Christine M. for the VW example.]