This technology is a language learning model that reduces bias in conjoint experiments that identify issues or features that influence people’s decisions.
Conjoint experiments allow researchers to identify key attributes or features that influence people’s decisions. Usually, these experiments involve participants making a series of choices, revealing common characteristics in the totality of decisions. However, the experimental design often includes pre-determined or assumed dimensions that may inadvertently influence choice and introduce bias from the researcher. Therefore, a system where decision outcomes are not predetermined or handcoded would significantly improve accuracy in identifying what truly influences one’s decision.
This technology describes a conjoint experiment design that utilizes large language models to provide a tailorable survey question to determine features that influence one’s decisions. The technology first poses an open-ended, free-response question, then bases a series of product choices to determine common features. Tailorability is available for each participant based on the unique free response, and an increased number of features can be screened. This language learning model provides a non-handcoded survey with increased accuracy and decreased bias.
This technology has been validated through the identification of key voting issues.
IR CU25452
Licensing Contact: Dovina Qu
