Context Aware Search

This is the holy grail of search. This is where you take into account context about the user making the search and use that to customize the search results based on that information.

Let's say you run an e-commerce company that sells clothes, and you know the user just bought a pair of leather shoes. You could recommend leather belts that match, perhaps a matching watch band.

Context can be anything you know about the user:

• Purchase history
• Search/Browsing history
• Profile information
• Geo location

I’ll admit the privacy advocate in me hates this, but as a guy who buys stuff online a lot, I love it.

I bought some patio furniture and a big umbrella this summer. It never occurred to me to buy a cover for the umbrella to keep it safe during our brutal Wisconsin winters, but a context aware recommendation showed me a cover for the umbrella. I didn’t even know that such a product existed.

How can this be achieved?

Services like AWS Personalize can build models fairly quickly. You can always train a classifier model, but I would wager you could get better results with a Vector Index to query tangentially related products.

I personally am testing all 3 methods to see which one will give us the biggest ROI on our investment.

Check out the Schematical Group Coaching Community where I help people like you design systems like this that will scale up in a cost-effective way.