Use Metadata to Enhance Personalization

Those of you who’ve been with me since the beginning (anyone?) remember my first claim regarding personalization: It doesn’t have to be expensive to be effective. My early columns presented low-cost personalization solutions and ideas. Today, a return to that theme and a subject near and dear to my and my clients’ hearts: metadata.

Metadata

Metadata is “data about data,” such as a book’s price or author. A digital picture’s metadata tells when the photo was taken, the exposure, and focal range. Metadata saved with a news article identifies category (e.g., world news) and key ideas (or keywords) in the text: health, fitness, diet, weight loss.

There’s also metadata about users and customers. This metadata tells a company that user “Jack Aaronson” is male, lives in New York, and has bought many products with the company before, including X, Y, and Z.

Personalization is best when it executes on the metadata level. Metadata helps answer why (versus what). The fact I bought products X, Y, and Z is less important than understanding why I bought them. Two different people can buy the same three products for very different reasons. Personalization for those two (e.g., recommending more products) must take into account why they were interested in those products, not just look statistically at what they bought.

How in the world can a computer guess why someone buys a given product? The answer lies in the product’s metadata, as well as metadata in other products that person buys. Products need a fine level of detailed metadata to accomplish this seemingly clairvoyant personalization feat.

Taxonomies and Ontologies

A taxonomy can be thought of as a dictionary of terms used throughout a metadata tree. A well-conceived taxonomy helps maintain a “clean” list of metadata tags, ensuring you don’t use two different words to mean the same thing (e.g., tagging one exercise book “health” and a similar one “fitness”).

An ontology goes a step further, organizing the taxonomy into a hierarchy. In a metadata tree lower-leaf metadata inherits the path of metadata that’s above.

A consumer goes to a store for a new computer mouse. She looks at the sign that points to “electronics.” She follows signs to “computers,” the “computer accessories” section, then the “pointing devices” aisle. Finally, she finds the area with only mice. Metadata associated with that mouse, in hierarchical order, are electronics > computers > computer accessories > input devices > pointing devices > mouse.”

In this real-world example, what should the salesperson (who’s on commission, so hovering close by) recommend in addition to that mouse? How can he know at this point why the consumer’s buying the mouse? How does he know what to show her next?

He doesn’t. He’ll likely ask a question or watch to see what the consumer does next. It’s certainly possible to add interactivity to your site to ask users what types of things they seek, but I promised we’d talk about personalization on the cheap. That’s not a cheap solution. Instead, let’s watch and see what the customer does next.

To make this more emphatic (and confuse everyone), let’s assume two customers just picked up the same mouse and are walking around the store. There are two money-grubbing salespeople hovering over their shoulders waiting to help with their next purchase. (See where I am going with this? Both customers have the same product, each for a different reason. “What” is the same, “why” is different.)

The first customer goes to the keyboard area of the “computer accessories” section and looks at keyboards, then at other add-ons. Metadata for the keyboard are electronics > computers > computer accessories > input devices > keyboards. The other add-ons all start with the same three tags, then branch to other computer accessories sub-trees. The eager salesman thinks, “This person is looking at various different types of computer accessories, so I’ll show her more keyboards, maybe a joystick for gaming, a USB port enhancer, and other things you could add to a new computer to make it more powerful and cool.”

The second customer stays in the mouse aisle and looks at other mice. These all have the same metadata as the original mouse she picked up. Her salesperson thinks, “This person is trying to find the best mouse-type device. Or maybe she needs a few different mice for several computers or specific mouse features, like one with buttons to control PowerPoint presentations.”

Watching this process makes everything seem like common sense. The first consumer stayed in the “computer accessories” section but is interested in many different types of accessories. The second consumer is focused just on pointing devices. The salespeople picked up on that. Although they don’t know the exact reasons why these consumers look at what they do, the salespeople pretty quickly figure out the pattern and know what products to start recommending.

Back to the Online World

It’s easy for a salesperson to think this through, and it’s just as easy for computers to do the same, without tons of expensive technology. Look at product metadata. The first product (the mouse) brought us down a path of metadata (the online equivalent of department, section, and aisle).

Further products also brought us somewhere along the same path, then diverged. A computer, as easily as a human salesperson, can look at metadata streams and count how often the same metadata is viewed. Once metadata is counted, it’s easy to locate the pieces that occur most frequently and pinpoint which frequently occurring pieces sit on the lowest branches of the tree.

For the first shopper, the most frequently occurring piece of metadata (at the lowest spot in the tree) is “computer accessories.” A salesperson (online or off-) can recommend any products that fall under that general heading. The customer is looking for any number of different types of things to buy, all having to do with things to add to a computer.

The second customer should only be recommended other pointing devices, as the lowest-level metadata she looks at is “pointing devices.” All pointing devices (trackballs, infrared pointers, etc.) are fair game.

Getting Started

The first (and hardest) step is defining a clear, precise, and detailed metadata taxonomy and ontology for your product space. There are many products on the market and many consulting companies that spend a lot of time creating these metadata worlds for their clients. After this is accomplished, recommending products and other items of interest to users becomes a lot easier.

Product recommendation based on browsing or buying behavior is just the tip of the iceberg. Many personalized services can be created based on this taxonomy. If this topic interests you, let me know. I’ll share more metadata-based personalization ideas you can add to your site.

Until next time…

Jack