LLM Context Graph Generation for User Research
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for developing context of use models are costly and time-consuming, relying heavily on user research that is often limited by access and expense, and result in unstructured information that is difficult to synthesize into comprehensive models.
Innovation Solution
A system that automatically generates context graphs based on queries submitted to large language models (LLMs), using an ontology to map user queries, parse responses, and iteratively build out the context graph by submitting subsequent queries and parsing sub-nodes from LLM outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional user research methods (site visits, interviews) are used to gather Context ofUse information, then comprehensive user understanding can be achieved, but the process becomes expensive and time-consuming
Solution Approach 1:
The patent creates synthetic user profiles that copy and represent real user characteristics, behaviors, and needs without requiring actual user interaction. These synthetic personas capture essential user attributes while eliminating the time-consuming nature of traditional research methods.
Solution Approach 2:
The system enables self-service generation of Context ofUse models by automatically analyzing product data, user feedback, and behavioral patterns to create comprehensive user understanding without requiring external research intervention.
2Loss of information
If traditional user research methods are used to gather information, then user insights can be obtained, but the information remains unstructured and difficult to synthesize
Solution Approach 1:
The patent segments user understanding into structured components including synthetic personas with defined attributes, contextual scenarios, and organized behavioral patterns. This segmentation transforms unstructured research data into manageable, organized elements that are easy to synthesize and apply.
Solution Approach 2:
The system changes the state of information from unstructured text to structured parameters by defining specific attributes for each synthetic persona (demographics, behaviors, needs, goals). This parameterization enables systematic organization and retrieval of user insights.
3Productivity
If LLM chat interfaces are used for knowledge acquisition, then information can be gathered quickly, but the output lacks coherent organization from the perspective of user requirements
Solution Approach 1:
The patent introduces an intermediary layer between LLM knowledge generation and final output that structures the information according to user requirement perspectives. This intermediary process transforms unorganized LLM responses into coherent, requirement-aligned information structures.
Solution Approach 2:
The system creates a universal framework that can organize diverse LLM outputs into a consistent structure applicable to various user requirements. This multi-functional approach allows the same organizational structure to handle different types of knowledge while maintaining coherence with user needs.
4Loss of information
If extensive user research is conducted to build comprehensive knowledge base, then broad understanding is achieved, but duplication of efforts increases difficulty and time needed
Solution Approach 1:
The patent merges multiple sources of user understanding (product data, feedback, behavioral patterns, LLM knowledge) into a unified synthetic persona framework. This consolidation eliminates duplication by integrating diverse information streams into a single coherent structure.
Solution Approach 2:
The system performs preliminary organization of user insights by creating structured synthetic personas before analysis or application is needed. This advance structuring eliminates the need for time-consuming organization later in the process.
Data Source
AI summary
A system performs automatic context graph generation based on LLM (large language model) queries. The system submits a user query to the LLM based on an ontology for a context of the user query. The system parses a response from the LLM system to automatically identify a context graph node from the response based on the ontology. The system automatically builds the context graph based on the identified node, and builds out the context graph by iteratively submitting subsequent queries and parsing out sub-nodes from subsequent responses. Repeating the submitting of LLM prompts and identifying child nodes from the LLM output based on the identified ontology builds out the graph until boundary conditions are met. The system builds out the graph with an accumulated context represented in the form of the existing nodes.