Microsoft patents system to match your online habits with content you will use
This filing describes a method for training a computer model to guess what topics interest a user based on their past actions. It cleans up the user's history to remove random noise, then tests different topic labels to see which ones best find content the user actually uses.
In the filing’s own words
A technique trains an interest-generating model that maps activity data to a set of candidate interest key phrases. The activity data describes prior action in interacting with computer-implemented services. Once trained, the interest-generating model is used to select and serve items. The technique for training reduces noise in its output results by first filtering the activity data using a filtering model, to produce filtered activity data. The technique then trains the interest-generating model to produce key phrases that are effective in both (1) expressing the interests underlying the filtered activity data and (2) retrieving items from a retrieval engine that are likely to consumed when served, given the identified interests. The technique accomplishes the above goals using reward-based training, which involves scoring candidate interest key phrases based on their effectiveness in retrieving relevant items, and fine-tuning the interest-generating model based on the scored candidate interest key phrases.
What exactly they claim (claim 1)
1. A method for training an interest-generating model, comprising: receiving initial activity data from one or more data stores that describes interactions with one or more computer-implemented services; filtering the initial activity data using a machine-trained filtering model to remove a subset of the initial activity data that represents noise, to produce filtered activity data; mapping the filtered activity data to a set of candidate interest key phrases; assigning scores to the candidate interest key phrases using a machine-trained reward system, to produce scored candidate interest key phrases, each score assigned to a particular candidate interest key phrase describing an effectiveness of the particular candidate interest key phrase in retrieving items from a retrieval engine that will be selected when served; and performing reward-based training of the interest-generated model based on the scored candidate interest key phrases, upon completion of the reward-based training, the interest-generating model aligning an information space associated with the filtered activity data with an information space associated with an index of the retrieval engine and associated consumption metrics that describe prior consumption of available items in the index.
Facts
- Company
- Microsoft (Microsoft Technology Licensing, LLC)
- Official title
- Reward-Based Training of an Interest-Generating Model for use in Retrieving Items
- Publication number
- US20260289394A1
- Application number
- 19087412
- Filed
- Mar 21, 2025
- Published
- Sep 24, 2026
- Inventors
- 3
- Claims
- 20
Classified under
G06N 20/00 Physics & computingG06F 18/214 Physics & computingG06Q 30/0246 Physics & computingMore from Microsoft in AI & software
- Microsoft patents method to check AI answers against verified facts Oct 20, 2026
- Microsoft patents AI that presents PowerPoint slides for you Sep 3, 2026
- Microsoft patents method to check if LLM answers are based on real data Sep 10, 2026
- Microsoft patents system to blend multiple AI agents into one chat Oct 1, 2026
- Microsoft patents system that answers questions using your screen history Oct 15, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.