How Your Device Learns Your Habits for Personalized Content Recommendations

This patent describes a system where your device creates a personal profile of your behavior, combines it with a vast online knowledge base, and uses both to give you super-personalized content recommendations.

What it covers

This patent outlines a method for a device, like a smartphone, to generate highly personalized content recommendations. It starts by collecting "log history information" about how you use the device, such as apps opened or settings changed (Claim 1). Based on this history, the device builds a "device knowledge graph" – essentially a map of your interactions, structured in a specific data format called an ontology (Claim 1). It also creates a "pattern knowledge graph" that captures your typical behavior patterns, connecting your actions to specific situations (Claim 3). This local knowledge is then combined with a larger "server knowledge graph," which is updated regularly with vast amounts of data from the internet (Claim 1). The device updates its own knowledge graph by adding your behavior patterns and connects its local insights to the server's broad knowledge. Finally, the device uses this combined, updated knowledge to provide you with content recommendations tailored to your current situation, or "context data," on its screen (Claim 1). For example, if your device learns you often listen to podcasts while commuting (a behavior pattern in a specific context), it might recommend new podcasts when it detects you're on your morning bus ride.

What it doesn't cover

• —Does not cover content recommendation systems that rely solely on a server-side knowledge graph without generating or updating a device-specific knowledge graph.
• —Does not cover systems where the device knowledge graph is not structured in an 'ontology form comprising a first entity expressed in a resource description framework (RDF)'.
• —Does not cover systems that do not generate a 'pattern knowledge graph associated with a behavior pattern of a user' based on device usage.
• —Does not cover systems where the server's knowledge graph is not 'updated periodically by the server based on big data collected through web crawling' and does not 'support a search engine'.
• —Does not cover recommendation systems that do not connect device-generated recommendation knowledge to server-generated recommendation knowledge to update the device's graph.

The clever bit

The innovation lies in the device's ability to create and continually update a detailed map of a user's behavior patterns (a 'pattern knowledge graph') locally, and then intelligently merge this personal understanding with a massive, general knowledge graph maintained on a server. This dual-layer, self-improving system allows for highly context-aware and personalized recommendations that go beyond simple preferences, adapting to how and when a user typically interacts with their device.

Why it matters

Personalized content recommendations are a cornerstone of modern digital experiences, influencing everything from what videos you watch to what products you buy. This patent matters because it describes a sophisticated way to make these recommendations even smarter by combining a user's unique, on-device behavior patterns with a vast, general knowledge base from a server. This hybrid approach allows for highly relevant suggestions that adapt to individual habits and real-time context, potentially leading to more engaging user experiences across various devices and services.

Real-world examples

• 1.Samsung Bixby recommendations
• 2.Google Discover feed
• 3.Netflix content suggestions
• 4.Amazon product recommendations
• 5.Spotify music playlists
• 6.TikTok video feed