Digital Assistant Sample Expression Generation From Application Metadata

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Creating diverse and high-quality sample expressions for intent classification in digital assistants is inefficient and costly, requiring significant human expertise and time, especially when modeling numerous intents and actions across multiple applications.

Innovation Solution

Automate the generation of sample expressions using a generative machine learning model, leveraging application metadata to produce structured prompts that guide the model to generate varied and comprehensive expressions, which are then processed and integrated into a digital assistant's configuration.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If human experts manually create sample expressions for intent classification, then data quality and accuracy are improved, but time consumption and costs increase significantly

Engineering Contradiction:
Improvedata qualityVSAvoidtime consumption
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The system uses a generative machine learning model to automatically generate sample expressions by learning patterns from existing sample expressions and application metadata. Instead of manually creating each sample expression, the model copies and adapts linguistic patterns to produce diverse, high-quality training data at scale, resolving the contradiction between data quality and time consumption

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The system enables self-service data generation where the machine learning model autonomously creates sample expressions without requiring continuous human intervention. The model uses application metadata and existing samples to self-generate training data, reducing reliance on human experts while maintaining data quality

Inventive Principle:
Principle #25Self-service

2Manufacturing precision

If human experts manually create sample expressions, then data quality is improved, but costs increase significantly

Engineering Contradiction:
Improvedata qualityVSAvoidcosts
Core Design Contradiction:
Manufacturing precisionVSLoss of energy

Solution Approach 1:

The generative model copies linguistic patterns from existing high-quality sample expressions to generate new samples automatically. This eliminates the need to pay human experts for each sample creation while preserving data quality through pattern-based generation

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The system replaces expensive human expert labor with automated machine learning generation. The computational resources required for model inference are significantly cheaper and more scalable than human expertise, enabling cost-effective large-scale data generation

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

3Adaptability or versatility

If the number of intents and actions to be modeled increases, then digital assistant functionality is improved, but complexity of data generation increases

Engineering Contradiction:
ImprovefunctionalityVSAvoiddata generation complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The generative machine learning model serves as a universal data generation system that handles multiple intents and actions across different applications. By learning general linguistic patterns from diverse application metadata, the model can generate sample expressions for any intent or action without requiring separate manual processes for each, reducing complexity while supporting expanded functionality

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Measurement precision

If the number of sample expressions per intent is increased, then intent classification accuracy is improved, but data generation time increases

Engineering Contradiction:
Improveclassification accuracyVSAvoiddata generation speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The generative model enables continuous, automated generation of sample expressions without interruption. Once the model is trained, it can generate unlimited numbers of diverse sample expressions on demand, maintaining high classification accuracy while eliminating the time constraints associated with manual creation. The system can generate samples in parallel and at scale, achieving both high quantity and high quality simultaneously

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4589429A1Automated training data generation for digital assistant
Publication Date: 2025.07.23 SAP SE
  • EP4589429A1 patent drawingFigure 1
  • EP4589429A1 patent drawingFigure 2
  • EP4589429A1 patent drawingFigure 3

AI summary

Example systems and methods described herein relate to the automated generation of sample expressions. Metadata is accessed for each of a plurality of applications. The metadata includes a functional description of each application. Prompt data is provided to a generative machine learning model. The prompt data includes the metadata for each of the plurality of applications and an instruction to generate, for each of the plurality of applications, a plurality of sample expressions corresponding to user input provided to a digital assistant to invoke an action related to the application. One or more responses that were generated by the generative machine learning model based on the prompt data are processed to obtain output data including the plurality of sample expressions for each of the plurality of applications in a structured format. The output data is used to configure the digital assistant.