Dependency Graph NLP Pipeline Automation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current natural language processing systems are inefficient, inaccurate, and not scalable to handle complex semantic models and large teams of developers, due to cumbersome management of multiple pipelines and sensitivity to slight input variations.

Innovation Solution

The system generates a dependency graph to organize natural language processing operations based on their dependencies, automatically creating and optimizing pipelines to minimize computing resources and maximize performance, allowing for efficient processing and retraining of semantic models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple separate natural language processing pipelines are used to handle different AI/ML models, then each model can be processed with its specific requirements, but the system becomes cumbersome and technically complicated to manage

Engineering Contradiction:
Improvemodel-specific processing capabilityVSAvoidpipeline management complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent merges multiple separate NLP pipelines into a single unified pipeline that can handle multiple AI/ML models. The system consolidates tokenization, normalization, and classification operations into one integrated pipeline that processes inputs for various models (e.g., risk identification, entity extraction) sequentially, eliminating the need to manage multiple independent pipelines while maintaining model-specific processing requirements.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The unified NLP pipeline is designed with universal functionality to serve multiple AI/ML models simultaneously. It implements a single pipeline structure that can be configured to handle different model requirements through parameter settings, making the system multi-functional without requiring separate specialized pipelines for each model type.

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

2Measurement precision

If ML classifiers and text processors require exact same input types for accurate prediction, then processing accuracy is maintained, but the system becomes sensitive to slight variations and requires extensive preprocessing

Engineering Contradiction:
Improveprediction accuracyVSAvoidinput processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary normalization and standardization of input data before it reaches the ML classifiers. The unified pipeline includes preprocessing steps that convert various input formats into a standardized form early in the processing sequence, ensuring that all subsequent classifiers receive consistently formatted inputs without requiring complex individual preprocessing for each model.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts processing parameters within the unified pipeline based on the specific model requirements. It changes parameters such as tokenization depth, normalization level, and feature extraction methods according to the target model, allowing the same pipeline structure to adapt to different accuracy requirements without manual reconfiguration.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If the coverage and sophistication of the natural language model increases, then semantic understanding capability improves, but the time required for software development and administrative maintenance increases to the point that such systems are no longer cost effective

Engineering Contradiction:
Improvesemantic understanding capabilityVSAvoidsoftware development and maintenance time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system segments the NLP processing into distinct, modular stages (tokenization, normalization, classification) within the unified pipeline. Each stage can be independently developed, tested, and maintained, allowing teams to work on different segments simultaneously and reducing overall development time while maintaining sophisticated semantic understanding capabilities.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The unified pipeline incorporates automatic configuration and self-optimization capabilities that reduce administrative maintenance requirements. The system automatically adjusts processing parameters and optimizes performance based on usage patterns, minimizing the need for manual intervention and reducing maintenance time as the system scales.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3671526B1Dependency graph based natural language processing
Publication Date: 2021.11.17 ACCENTURE GLOBAL SOLUTIONS LTD
  • EP3671526B1 patent drawingFigure 1
  • EP3671526B1 patent drawingFigure 2a
  • EP3671526B1 patent drawingFigure 2b

AI summary

Examples of automatically generating natural language pipelines to process an input to generate tags, semantic or syntactic, are described. In an example, on receiving a request to process input data to generate tags, a dependency graph, based on identified dependees and further dependees may be created to satisfy the request, the dependency graph including natural language operations arranged in order of their dependencies on each other. Based on the dependency graph, a pipeline for the tags may be automatically generated, which includes a series of natural language operations such that the operations for dependee tags are processed before any of their associated depender tags. Further, the dependency graph and the automated pipeline generation allows for automated optimization of the pipeline, training, re-training, testing and regression testing of the semantic tags and supporting machine learning models and provides a framework to efficiently manage the sharing and reuse of semantic understanding operations.