Hierarchical NLP Agent Structure for Intent Scaling

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing natural language processing (NLP) systems have a static limit on the number of supported intents, which restricts their ability to handle more complex interactions requiring a higher number of intents, and modifying these systems is often not feasible due to technical or financial constraints.

Innovation Solution

Implementing a hierarchical agent structure, where a Master Agent and multiple lower-level agents route user input, allowing for significantly more intents to be supported without altering the existing NLP system architecture, with up to 4,002,000 intents possible in a two-level hierarchy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a hierarchical agent structure is implemented to support more intents, then the capacity of the NLP system is improved, but the device complexity increases

Engineering Contradiction:

Improveintent capacityVSAvoidsystem structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent divides the NLP system into multiple agent components (Master Agent and lower-level agents), where each agent handles a subset of intents. This segmentation allows the system to support over 4 million intents by distributing the intent-handling capacity across hierarchical levels, resolving the contradiction between intent capacity and system complexity.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a hierarchical dimension to the agent structure, organizing agents in multiple levels rather than a single flat layer. This dimensional change enables exponential scaling of intent capacity (4,002,000 intents in a two-level hierarchy) while maintaining manageable complexity at each individual agent level.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Adaptability or versatility

If more intents are supported to handle complex interactions, then the adaptability is improved, but the processing overhead increases

Engineering Contradiction:

Improveinteraction complexityVSAvoidprocessing overhead

Core Design Contradiction:

Adaptability or versatilityVSLoss of energy

Solution Approach 1:

By segmenting the intent space across multiple agents in a hierarchy, the processing overhead is distributed rather than concentrated. Each agent processes only its subset of intents, reducing the computational burden per agent while collectively handling complex interactions requiring many intents.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent assigns different functional qualities to different levels of the hierarchical agent structure. Lower-level agents handle specific, localized intent categories, while the Master Agent coordinates higher-level interactions. This local quality optimization reduces processing overhead by ensuring each agent operates within its specialized domain.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11520995B2Systems and methods for dynamically expanding natural language processing agent capacity

Publication Date: 2022.12.06 VERIZON PATENT & LICENSING INC

AI summary

A system described herein may provide for the adaptation and/or expansion of a natural language processing (“NLP”) platform, that supports only a limited quantity of intents, such that the described system may support an unlimited (or nearly unlimited) quantity of intents. For example, a hierarchical structure of agents may be used, where each agent includes multiple intents. A top-level (e.g., master) agent may handle initial user interactions, and may indicate a next-level agent to handle subsequent interactions.