Interactive Persona Design From Survey-Based Synthetic Participants
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current persona generation systems, particularly those using large language models (LLMs), fail to provide a mechanism to understand and explore user behavior implications and are prone to dataset bias, unable to connect responses to actual data.
Innovation Solution
A method and system that leverage previous studies to create a multi-task learned representation for personalization vectors and individual task models for each survey question, enabling the prediction of synthetic persona responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If LLM-based systems are used for persona generation, then synthetic participants can be created to represent user types, but the systems amplify dataset bias and cannot connect responses to actual data
Solution Approach 1:
The patent creates synthetic participants by copying and transforming real participant data through a systematic process. Real participants complete surveys with multiple questions, and the system generates synthetic versions that maintain the essential characteristics and response patterns of real users while preserving the connection to actual survey data, thereby avoiding LLM bias while maintaining synthetic participant utility
Solution Approach 2:
The patent introduces an intermediary data transformation layer between real participant data and synthetic participants. This intermediary process systematically transforms real survey responses into synthetic participant representations while maintaining the link to original data, serving as a bridge that preserves data connectivity without relying on biased LLM-generated content
2Measurement precision
If new consumer surveys are fielded to understand user behavior, then current product ideas can be tested, but time and resources are consumed that could be used for other initiatives
Solution Approach 1:
The patent performs preliminary data collection and transformation work by creating a library of synthetic participants based on previous survey data. This preliminary action enables the system to quickly generate responses to new survey questions without fielding completely new studies, as the synthetic participants are pre-prepared and can be queried immediately for new product ideas
Solution Approach 2:
The system enables self-service by allowing designers and researchers to query synthetic participants for new survey questions without requiring manual survey fielding. The synthetic participants automatically generate responses based on their trained characteristics, providing immediate insights for new product ideas without consuming additional time for survey administration
3Loss of information
If comprehensive survey questions are asked to capture all user behaviors, then detailed user insights are obtained, but the complexity of data collection and analysis increases
Solution Approach 1:
The patent segments the complex survey data into distinct components: real participant responses, synthetic participant representations, and query-specific task models. This segmentation allows the system to handle comprehensive survey questions by breaking them down into manageable units that can be processed independently, reducing overall system complexity while maintaining detailed user behavior insights
Data Source
AI summary
A method for generating an interactive persona design system is described. The method includes creating a persona description, by a designer/marketer, representing a synthetic person for which an interview is desired regarding a survey of new questions. The method also includes translating, using a personalization model, the persona description into a personalization vector, in which the personalization vector represents the synthetic person. The method further includes creating a query vector including an individual task model for each question of the survey of new questions. The method also includes training a choice model based on the personalization vector and the individual task models for each survey question of the query vector to predict a response of the synthetic persona for each of the survey of new questions based on the persona description.