How Do Websites Offer Personalized Recommendations?
When I'm working from home, instead of playing a tired radio station over the speakers, I'll usually click onto a website that plays a customized radio station based on songs and artists I've listened to before. Similarly, I like to stream movies and TV shows onto my television, and the program I use for that offers up recommendations for what I should watch next based on things I've watched before. It's sort of a strange thing, to have a computer recording things I like, so I set out to learn how websites like this make personal recommendations despite my not giving them all that much information.
It turns out, there's a couple of different ways that websites do this, and it's not just for things like movies and music. Online retailers can also track things you have searched for and purchased and suggest similar products that you might also like to buy. If you've ever purchased something online only to be taken to a page that suggests you buy items from "similar users," you've experienced a personalized recommendation. Let's take a look at how all of this works.
In the realm of online shopping, rules-based filtering refers to the sort of recommendations that come from things you have previously searched for or purchased. Retailers track what you've looked for and purchased and store that information in a database that identifies patterns in your behavior. Those patterns are used to make assumptions about what you might search for next. For example, if you purchased one album from a certain artist, the website might suggest purchasing another album by the same artist. Rules-based filtering is taken care of by having shoppers fill out a form when they shop that allows their information to be filed away. These forms typically include things like your name, age, address, etc. Basic information about yourself that will differentiate you from someone with the same name is all that's needed for rules-based filtering.
Collaborative filtering is a little bit different than rules-based filtering. As its name suggests, this method takes your information and patterns and collaborates with information from shoppers with similar patterns to suggest things that it thinks you will like. For example, if you and 5 other people purchased one artist's record, the website will look at other albums you've all purchased and suggest them to everyone based on the common interest of the first artist. Put simply, it's sort of like if your friend that you went to a concert with suggested a band you might like. Collaborative filtering takes the patterns and behaviors of one person makes suggestions on what to buy next based on what other people like them did.
Does It Work?
As you might imagine, sometimes the recommendation algorithm is a bit off. Some websites customize their recommendations by location when it makes sense to do so—for example, a shopper in Florida who was searching for surf equipment might not get recommendations for skiing equipment even if some other users who purchased surf equipment also bought skiing garb—there's not much skiing to be done in Florida, so the recommendation wouldn't be very effective (at least for most shoppers). It depends on how in-depth retailers want to go as far as collecting information—location, household income, your hobbies, age, and occupation can all be factors that might influence your purchasing patterns.
However, when it comes to personalizing things like social networks and ads, it's pretty clear that these so-called recommendation algorithms need a bit of work. When you do a search on a search engine, you might see ads related to the search for a few days. Some of this makes sense, but it can be pretty annoying when you search something, seek out what you were trying to do or buy, then have to look at an ad for the same thing for a week. By then, the ad is obsolete. There needs to be a better way to track what people are doing on the internet if personalization is the goal. Otherwise, retailers and marketers should back away from making everything tailored to each individual user.
How Can It Contribute to Business?
When a business offers recommendations to its customers about what they should do next, the outcome can be a bit of a gamble. Of course, there are going to be customers who purchase only what they're looking for and move on. But there will be customers who see recommended products and think that yes, they do want that product also. Theoretically, personalized recommendations will increase sales, since the company will have sold the base product that the user was going for in the beginning, plus other items they weren't necessarily in the market for.