Dynamic Agent System for Natural Conversation via Bayes Estimation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional yes-or-no question systems, such as binary tree formats, require numerous questions to determine user preferences, leading to irrelevant and time-wasting interactions, poor user experience, and difficulties in maintaining and updating questions and items, while failing to efficiently utilize user history data.

Innovation Solution

An agent system that dynamically selects statistically meaningful questions using Bayes Estimation based on user answers, allowing for natural and efficient conversations, flexible maintenance, and autonomous evolution with user history data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If a binary tree format with predetermined questions is used, then the system structure is simple and easy to implement, but many questions are required leading to irrelevant and time-wasting interactions

Engineering Contradiction:
Improveease of system implementationVSAvoidtime required for question interactions
Core Design Contradiction:
Ease of manufactureVSLoss of time

Solution Approach 1:

The system transitions from a static binary tree structure to a dynamic question selection mechanism that adapts based on user history data and statistical analysis. Questions are selected dynamically using Bayes Estimation to determine the most informative questions, reducing the total number of interactions while maintaining personalization effectiveness.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system incorporates feedback loops where user answers are continuously analyzed to update probability estimates for different user profiles. This feedback mechanism allows the system to learn from user responses and optimize question selection, reducing unnecessary questions while improving recommendation accuracy over time.

Inventive Principle:
Principle #23Feedback

2Adaptability or versatility

If many questions are statically associated in advance in a binary tree, then comprehensive coverage of user circumstances is achieved, but the conversation becomes unnatural and irrelevant to user context

Engineering Contradiction:
Improvecoverage of user circumstancesVSAvoidnaturalness of conversation
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The system changes the parameter of question selection from fixed static associations to dynamic probability-based selection. By calculating conditional probabilities using Bayes Estimation, the system determines which questions are most likely to provide useful information given the current user context, making conversations more natural while maintaining comprehensive coverage.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The system replaces the mechanical binary tree structure with a statistical probability model. Instead of following predetermined question paths, the system uses Bayes Estimation to dynamically select questions based on user history data, enabling more natural and context-appropriate conversations while achieving the same goal of user profiling.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Stability of the object's composition

If the binary tree structure is used to ensure systematic question progression, then question organization is maintained, but maintenance and updates of questions and items become difficult

Engineering Contradiction:
Improvequestion organization structureVSAvoidease of question and item maintenance
Core Design Contradiction:
Stability of the object's compositionVSEase of repair

Solution Approach 1:

The system segments the question bank into independent, loosely-coupled units that can be individually added, removed, or modified without affecting the overall system structure. Questions are stored as discrete records with associated metadata, allowing administrators to maintain and update content flexibly without reconstructing the entire question framework.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system creates a universal question management framework that works with any question set without requiring structural reconfiguration. The Bayes Estimation-based selection mechanism is agnostic to the specific content or organization of questions, allowing easy updates and additions while maintaining systematic question progression through probability-based routing.

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

4Ease of manufacture

If a fixed binary tree configuration is used, then initial system setup is straightforward, but usage history data from many users cannot be flexibly reflected

Engineering Contradiction:
Improveease of initial system setupVSAvoidability to reflect user history data
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The system enables itself to automatically learn and adapt from user history data without requiring manual reconfiguration. The Bayes Estimation mechanism automatically processes user responses, updates probability distributions for different user profiles, and optimizes question selection based on accumulated data, allowing the system to self-improve while maintaining its straightforward operational framework.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP2885755B1Agent system, agent control method and agent control program with ability of natural conversation with users
Publication Date: 2019.09.25 ACCENTURE GLOBAL SERVICES LTD
  • EP2885755B1 patent drawingFigure 1
  • EP2885755B1 patent drawingFigure 2A~2D
  • EP2885755B1 patent drawingFigure 3A~3B

AI summary

An object of the present invention is to provide an agent system and an agent control method for providing appropriate advice, proposals, recommendations, etc. to individual users. The system executes and the method includes the steps of: storing, into a database, a plurality of question response data acquired from a plurality of prior users, wherein the plurality of question response data is indexed in association with each of a plurality of questions and in association with each of a plurality of recommendations; identifying a set of candidate recommendations from the plurality of recommendations based on a first predefined criterion; repeating a sub-method until determining that a predefined condition is satisfied, the sub-method including the steps of selecting a question from the database based on a second predefined criterion, acquiring an answer from the user to the question selected in the step of selecting a question, updating probabilities of all the recommendations for the user based on the answer from the user, and updating the set of candidate recommendations based on the updated probabilities of the recommendations; and outputting a recommendation selected from the set of candidate recommendations, in a case of determining that the predefined condition is satisfied.