Dynamic Interest Taxonomy Using Unsupervised Keyword Graphs

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional content understanding and curation techniques suffer from limitations such as lack of depth and granularity in taxonomies, static nature, incomplete coverage, and fragmentation across different products and content types, leading to ineffective matching of user interests and inefficient content curation.

Innovation Solution

An interest graph is constructed using unsupervised machine learning techniques to dynamically adapt to emerging interests, leveraging tailored preprocessing, unsupervised keyword extraction, and a pairwise classification model for content tagging, providing a comprehensive and agile taxonomy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional content understanding techniques use static taxonomies, then system simplicity is maintained, but adaptability to emerging interests deteriorates

Engineering Contradiction:

Improveadaptability to emerging interestsVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements dynamic taxonomies that automatically evolve and adapt to emerging user interests through continuous learning from user-generated content. The system transitions from static categorization to dynamic, self-updating interest models that respond to changing user preferences and trends in real-time.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system employs self-service mechanisms where the taxonomy automatically updates itself by learning from user interactions and content without requiring manual curation. The interest models self-adjust based on user engagement patterns, eliminating the need for continuous human intervention to maintain relevance.

Inventive Principle:

Principle #25Self-service

2Reliability

If comprehensive content analysis is performed across all content types, then coverage completeness is improved, but processing time increases

Engineering Contradiction:

Improvecoverage completenessVSAvoidprocessing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent segments content analysis into distinct processing streams tailored to different content types (text, images, video, audio). Each segment uses specialized analysis techniques appropriate to its format, allowing parallel processing that maintains comprehensive coverage while reducing overall processing time through division of labor.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system applies partial analysis actions by focusing computational resources on the most relevant features of each content type rather than exhaustive analysis of all attributes. This selective approach maintains adequate coverage for accurate interest matching while significantly reducing processing overhead.

Inventive Principle:

Principle #16Partial or excessive action

3Measurement precision

If detailed keyword extraction is performed on all content items, then taxonomy granularity is improved, but computational resources consumed increases

Engineering Contradiction:

Improvetaxonomy granularityVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies local quality by varying the depth and detail of keyword extraction based on the specific content item's characteristics, its relevance to user interests, and its position in the content hierarchy. High-value content receives detailed analysis while lower-priority content receives streamlined processing, optimizing the balance between granularity and resource usage.

Inventive Principle:

Principle #3Local quality

4Measurement precision

If content is organized into deep hierarchical taxonomies, then categorization precision is improved, but navigation complexity increases

Engineering Contradiction:

Improvecategorization precisionVSAvoidnavigation ease

Core Design Contradiction:

Measurement precisionVSEase of operation

Solution Approach 1:

The patent introduces alternative navigation dimensions beyond traditional hierarchical paths, including interest-based clustering, similarity-based grouping, and multi-dimensional tagging systems. Users can navigate content through multiple lenses simultaneously, maintaining precise categorization while providing flexible, intuitive access paths that reduce navigation complexity.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS12437010B2Automatic techniques for constructing an evolving interest taxonomy from user-generated content

Publication Date: 2025.10.07 SNAP INC

AI summary

Techniques for creating an interest graph include obtaining content items from multiple content sources and applying tailored (e.g., source-specific) preprocessing to the content items based on their respective content source. Text is extracted and salient keywords and key phrases are identified using unsupervised machine learning models. The keywords and key phrases become nodes in an interest graph, each node comprising an embedding of a keyword or key phrase in a common embedding space, with edges representing semantic similarity based on embeddings or co-engagement patterns. The graph provides an expansive, granular, and dynamic taxonomy easily adaptable to emerging interests. The interest graph overcomes limitations of conventional taxonomies that lack depth, fail to capture niche interests, and cannot adapt to reflect evolving user preferences. The described techniques construct a rich interest graph from diverse content for improved content understanding.