Adaptive User Representation Using Query-Guided Profile Retrieval

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current AI assistant systems struggle to determine which user information is most relevant for generating personalized responses, leading to inefficient use of resources and suboptimal performance due to the integration of vast amounts of irrelevant data.

Innovation Solution

An adaptive user representation (AUR) system that processes user queries to determine context and retrieves relevant user information from a repository, generating instructions for the AI assistant to provide personalized responses.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If vast amounts of user information are directly integrated into AI assistants for every query, then personalization capability is improved, but resource consumption and processing costs increase significantly

Engineering Contradiction:
Improvepersonalization capabilityVSAvoidresource consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The system pre-processes user information from multiple sources (emails, chats, preferences, digital connections) and stores it in structured user profiles before queries are submitted. This preliminary organization allows the AI assistant to access only relevant pre-processed information during query handling, rather than processing all raw user data for every query, thus reducing resource consumption while maintaining personalization capability.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system extracts and separates relevant user information from the vast amount of collected data based on query context. The architecture includes components that selectively retrieve only the necessary user profile elements needed for each specific query, excluding irrelevant information from the processing pipeline, thereby reducing computational overhead while preserving personalization.

Inventive Principle:
Principle #2Taking out (Extraction)

2Adaptability or versatility

If all user information is included with each prompt, then personalization is enhanced, but AI assistant processing time and computational resources are wasted on irrelevant data

Engineering Contradiction:
ImprovepersonalizationVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system applies different processing qualities to different parts of user information based on query requirements. The query context determining model identifies which user profile components are locally relevant to each specific query, and the information retrieval model retrieves only those localized relevant portions, rather than uniformly processing all user information, thus reducing processing time while maintaining personalization quality where needed.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

User information is pre-categorized and tagged with metadata indicating its relevance to different query types. This preliminary structuring allows the retrieval model to quickly filter and select only the necessary information for each query, avoiding the time cost of analyzing all user data from scratch for every interaction.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If extensive user information is integrated into AI assistants, then response personalization is improved, but system complexity and integration costs increase

Engineering Contradiction:
Improveresponse personalizationVSAvoidsystem complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments user information into distinct profile components (preferences, contacts, digital connections, contextual data) and processes each segment separately through specialized models. The query context determining model, information retrieval model, and prompt generation model each handle specific segments, reducing overall system complexity by dividing the integration task into manageable, modular components rather than attempting to process all information uniformly.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system introduces intermediary models (query context determining model and information retrieval model) that mediate between the raw user information repository and the AI assistant. These intermediaries translate and filter user data into standardized, query-relevant formats, reducing the complexity of direct integration while preserving personalization capabilities.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If AI assistants process vast amounts of user information for every query, then personalization accuracy is improved, but computational efficiency decreases

Engineering Contradiction:
Improvepersonalization accuracyVSAvoidcomputational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system pre-computes and stores user profile representations with embedded relevance indicators and contextual metadata. When a query arrives, the query context determining model quickly matches the query against pre-computed profile segments, and the information retrieval model efficiently retrieves the most relevant pre-processed information, maintaining high personalization accuracy without the computational burden of processing all raw user data in real-time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system applies high-processing quality only to the locally relevant user information identified for each query, while using faster, lower-quality filtering for the initial selection of candidate information. This tiered approach ensures personalization accuracy for the critical relevant data while maintaining computational efficiency by avoiding exhaustive processing of all user information.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4693068A1Adaptive user representation system
Publication Date: 2026.02.11 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4693068A1 patent drawingFigure 1
  • EP4693068A1 patent drawingFigure 2
  • EP4693068A1 patent drawingFigure 3

AI summary

An adaptive user representation (AUR) system for use with generative artificial intelligence (AI) receives queries meant for the generative AI and utilizes one or more AI models to process each query to determine query context and to identify user information from a user information repository which is relevant to the query. The system generates instructions based on the query, query context, and the relevant user information for causing the generative AI to generate a response to the query which is personalized to the user. The AUR system transforms the raw data of the query and relevant user information into a set of instructions for the generative AI which describe how to personalize the response or required searches to ensure the final response is personalized.