How Does Human Preference Vary Across Semantic Categories?

Background

Words are representations of mental structures and allow effective communication. Human preference is encoded in word meanings and varies on an individual and group level as people use word choices to exhibit their social identity, which is based on several factors. Two of these are gender and personality which shape peoples’ everyday experiences and the language they use. Estimating the preferences of others involves taking the allocentric (other) perspective, which is based on one’s knowledge and preconceptions of other groups of people.

Research Questions / Hypotheses

This study aims to explore a) the extent of the variation of preferences based on gender and extraversion demographics; b) people’s accuracy in taking the allocentric perspective by both demographic and category; c) the capacity of Large Language Models (LLM’s) to accurately predict human behaviour using the allocentric perspective.

Participants

848 participants completed this study, and after cleaning, data from 619 participants was used. The exclusion criteria was for participants who did not have english as their first language as there are known differences in lexical choices and preferences based on language and cultural background.

Methods

This study involved 2 tasks, the first one was an item ranking task where participants were given 25 items per category and were asked to place them on a scale based on their own preferences. The second task was a forced choice task in which they were given an item from the previous list and asked to pick if it would be preferred by a male, female, introvert or extrovert. After this they were asked a personality questionnaire which consisted of the bfi-2 and BFAS in which their extraversion was determined.

Results

Preliminary data analysis showed significant discrepancies between categorical preferences on a group level for both egocentric and allocentric perspectives.

Implications

This contributes to an understanding of how human preference is encoded in semantic knowledge based on the egocentric and allocentric perspective, thusly reflecting the accuracy of LLM’s in showcasing human behaviour.