US 6,112,186 · Granted 2000-08-29

The Microsoft Patent That Taught Computers to Read Your Mind

Imagine a friend group where everyone rates movies and books. A computer looks at whose tastes match yours most closely, then recommends something new based on what your "taste twins" loved. That's collaborative filtering — and Microsoft patented a system to do it automatically across multiple computers connected to a central server.

Curated and verified by the IsItPatented Editorial team

The plain-English version

What it protects

The claim covers a distributed system that stores profiles of users and items, calculates similarity scores between users based on their rating patterns, identifies which users have the most similar preferences to a target user, assigns weights to those "neighbor" users, and then leverages their ratings to generate personalized recommendations. What's protected here is the specific method of using correlated user preferences across a network architecture to drive automated suggestion systems.

Why it matters

Collaborative filtering became the backbone of modern recommendation engines across streaming, shopping, and social platforms. Microsoft's patent staked an early claim to one of the most commercially valuable algorithms in digital commerce — the ability to predict what you want before you know you want it. This kind of patent became a battleground in the early 2000s as companies raced to build recommendation systems.

Real-world use

Every time Netflix suggests a show you end up binge-watching, or Amazon recommends a product you actually buy, a collaborative filtering system trained on millions of users' preferences is quietly at work behind the scenes.

Original USPTO abstract

A system for facilitating exchange of user information and opinion using automated collaborative filtering includes memory elements for storing item profiles and user profiles. The data contained in those profiles is used to calculate a number of similarity factors representing how closely the preferences of one user correlate with another. The similarity factors are evaluated to select a set of neighboring users for each user which represents the set of users which most closely correlate with a particular user. The system assigns a weight to each one of the neighboring users. The system uses the ratings given to items by those neighboring users to recommend an item to a user. The system may be distributed, i.e. the system may include a number of nodes connected to a central server. The central server includes a memory element for storing user profile data and the nodes may be the type of system described above.

Patent details

Publication number: US 6,112,186
Filing date: 1997-03-31
Grant date: 2000-08-29
Assignee: Microsoft Corporation
Inventor(s): BERGH; CHRISTOPHER P., METRAL; MAX E., RITTER; DAVID HENRY, SHEENA; JONATHAN ARI, SULLIVAN; JAMES J.
CPC class: H04N21/252

View original record on Google Patents →

Want to file your own patent?

Curious how your own recommendation app idea might stack up against decades of patent filings? Run it through our free patentability scanner to see what's already locked down in this space.

Free patentability scan