Learning Model for Multidimensional Content Relevance Mapping

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing technologies are limited to generating only text-based content and cannot effectively express various kinds of creative content that humans can create, requiring a method to represent components of diverse content as multidimensional vectors.

Innovation Solution

A learning model is developed to generate, present, and evaluate content by acquiring and decomposing it into components, measuring inter-concept distances, and generating new content based on relevance indices using a learning model that outputs index values indicating the relevance of components.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If text-based language models are used to generate content, then sentence generation is automated, but the system cannot express or generate diverse creative content beyond text

Engineering Contradiction:
Improvecontent type diversityVSAvoidmodel structure complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments content into discrete components (subjects, predicates, objects, modifiers) that can be independently represented as vectors. This segmentation allows the system to handle diverse content types by composing vectors from standardized component vectors, resolving the contradiction between versatility and complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from text-only representation to multidimensional vector representation that encompasses various content dimensions (semantic, syntactic, stylistic). By embedding content components in a high-dimensional vector space, the system can represent and generate diverse content types while maintaining a unified model structure.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If components are expressed as multidimensional vectors, then creative content generation is enabled, but measurement of relevance between components becomes complex

Engineering Contradiction:
Improvecontent representation capabilityVSAvoidrelevance measurement accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent changes the parameter representation from raw text to standardized vector embeddings with specific dimensions. By transforming content components into vectors with consistent dimensional properties, the system enables precise relevance measurement through vector operations while maintaining versatile content representation capability.

Inventive Principle:
Principle #35Parameter changes

3Productivity

If a learning model outputs index values for component relevance, then content evaluation becomes automated, but the system cannot provide creative recommendations

Engineering Contradiction:
Improveevaluation efficiencyVSAvoidcreative recommendation capability
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent implements a feedback mechanism where the learning model evaluates component relevance and uses this information to generate creative recommendations. The system feeds evaluation results back into the content generation process, enabling automated evaluation while maintaining creative recommendation capability through iterative refinement.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4703982A1Learning model generation method, presentation method, evaluation method, and generation method
Publication Date: 2026.03.04 SONY GROUP CORP
  • EP4703982A1 patent drawingFigure 1
  • EP4703982A1 patent drawingFigure 2
  • EP4703982A1 patent drawingFigure 3

AI summary

The present disclosure relates to a learning model generation method, a presentation method, an evaluation method, and a generation method which are capable of providing more creative content. A plurality of components constituting each of a plurality of different pieces of content is acquired, and a learning model that outputs an index value indicating relevance of a first component with respect to a second component among the plurality of components is generated. The technique according to the present disclosure can be applied to a system related to creation support for a creator, for example.