Conversation Control Unit Model Selection for Response Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing conversation systems struggle to generate appropriate responses based on the content of the conversation, leading to ineffective human-computer interactions.

Innovation Solution

The proposed solution involves an information processing apparatus with a model storage unit, a conversation history storage unit, and a conversation control unit. This apparatus stores multiple models, including large-scale language models and task-specific models, and uses conversation history to select the appropriate models for generating response messages.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a single large-scale language model is used for all conversation tasks, then the system has high versatility, but the response precision for specific tasks deteriorates

Engineering Contradiction:
Improveconversation task coverageVSAvoidresponse accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent divides the single large-scale language model into multiple specialized models, each trained for specific conversation tasks. The conversation control unit segments the conversation analysis into different dimensions (intention, sentiment, topic) and selects appropriate specialized models for each dimension, thereby improving response precision while maintaining versatility through the model selection mechanism.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a universal conversation system that can handle multiple task types by equipping it with a suite of specialized models. Each model serves a specific function (intention recognition, sentiment analysis, topic classification), but together they provide universal coverage across diverse conversation scenarios. The conversation control unit acts as a dispatcher that routes different conversation aspects to the appropriate specialized model.

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

2Measurement precision

If multiple specialized models are used for different tasks, then the response precision improves, but the device complexity increases

Engineering Contradiction:
Improveresponse accuracyVSAvoidmodel management complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The conversation control unit serves as an intermediary that manages the complexity of multiple specialized models. It receives the conversation input, analyzes it across different dimensions, and selectively invokes only the necessary specialized models for each conversation turn. This mediator approach allows the system to maintain high response precision through specialized models while hiding the complexity from the overall system architecture.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent implements a dynamic model selection mechanism where the conversation control unit determines which specialized models to activate based on the current conversation context, history, and detected dimensions. This dynamic approach allows the system to use only the necessary models for each specific conversation scenario, reducing the effective complexity at any given moment while maintaining the capability to handle diverse tasks.

Inventive Principle:
Principle #15Dynamics

3Reliability

If conversation history is analyzed in detail to select appropriate models, then the response appropriateness improves, but the processing time increases

Engineering Contradiction:
Improveresponse appropriatenessVSAvoidmodel selection time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary analysis of the conversation history to identify key dimensions (intention, sentiment, topic) before model selection. The conversation control unit pre-processes the conversation context to extract relevant features and determine which specialized models will be needed, thereby reducing the time required during the actual response generation phase while maintaining high response appropriateness.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4564271A1Information processing apparatus, information processing method, and conversation system
Publication Date: 2025.06.04 RICOH CO LTD
  • EP4564271A1 patent drawingFigure 1
  • EP4564271A1 patent drawingFigure 2
  • EP4564271A1 patent drawingFigure 3

AI summary

An information processing apparatus (11) includes a model storage unit (105, 106), a conversation history storage unit (104) and a conversation control unit (102). The model storage unit (105, 106) stores a plurality of models including a large-scale language model and a task-specific model different from the large-scale language model. The task-specific model have been acquired by machine learning specialized in a specific task. The conversation history storage unit (104) stores a history of conversations in which a conversation agent participates. The conversation control unit (102) selects one or more models from among the plurality of models based on the history of conversations and creates a response message of the conversation agent using output data of the selected model.