Universal Interest Space for Content Recommendation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional content recommendation systems fail to accurately capture users' interests due to fragmented user profiling, reliance on short-term interests, and inability to integrate interests across different applications, leading to incomplete and reactive content recommendations that do not account for long-term user interests or broader interest ranges.
Innovation Solution
A system and method that utilize a universal interest space defined by concept archives like Wikipedia to create a baseline interest profile, combining short-term and long-term user interests, and employing probing content to discover unknown interests, while integrating content from various sources and dynamically updating the content pool based on user interactions and context.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional user profiling methods are used based on declared interests and CTR, then user interests can be identified in isolated application settings, but the user profile becomes fragmented and cannot capture a broad range of overall user interests
Solution Approach 1:
The patent creates a universal interest space that maps user interests across multiple applications and contexts using a standardized taxonomy. This allows the same user profile structure to serve different applications (news, social media, e-commerce) while capturing diverse interest categories, resolving the fragmentation problem by making the profiling system universally applicable across contexts
Solution Approach 2:
The patent segments user interests into distinct categories within a taxonomy structure (e.g., politics, sports, entertainment with subcategories). This segmentation allows precise measurement of specific interests while the collective segments cover the broad spectrum of user interests, enabling both accuracy and comprehensiveness
2Productivity
If content recommendation systems rely on observed short-term user interests from past interactions, then they can provide reactive content matching, but they fail to discover unknown long-term interests of users
Solution Approach 1:
The patent performs preliminary actions by proactively presenting content from under-explored interest categories before users have naturally encountered them. Instead of waiting for users to independently discover these interests through random browsing, the system anticipates potential interests based on the universal taxonomy and presents relevant content in advance, enabling discovery of unknown long-term interests
Solution Approach 2:
The patent implements feedback loops where user responses to proactively presented content (clicks, time spent, shares) are continuously monitored and used to refine the understanding of user interests. This feedback mechanism allows the system to distinguish between genuine long-term interests and temporary curiosities, progressively building a more accurate comprehensive interest profile
3Ease of manufacture
If applications create their own isolated content pools based on application-specific user interests, then content can be optimized for that application, but a coherent content pool that serves broader user interests cannot be developed
Solution Approach 1:
The patent merges content from multiple applications into a unified content pool organized by the universal interest space taxonomy. Content is tagged with interest category identifiers that map to the standardized taxonomy, allowing content to be efficiently retrieved and recommended across different applications while maintaining application-specific optimization. This combining approach creates both simplicity through standardization and versatility through cross-application applicability
Data Source
AI summary
The present teaching relates to discovery of user unknown interests. In one example, information related to a user is retrieved from a user profile. The information indicates one or more known interests of the user. At least one known interest of the user is identified based on the information. One or more supplemental interests with respect to each identified at least one known interest of the user are identified. The one or more supplemental interests do not overlap with the one or more known interests of the user. Supplemental content associated with the one or more supplemental interests are identified. Each piece of content in the supplemental content is ranked. At least one piece of content in the supplemental content is selected based on the ranking. The selected at least one piece of supplemental content is used to discover unknown interest of the user.