Fig. 2: Overview of the models and knowledge structures.
From: Incorporating social knowledge structures into computational models
To explore participants’ behaviour we constructed five main computational models that made use of two main knowledge structures: Reference Points (RPs) and Granularity (G). a The Reference Points represent what participants can use as a basis for estimating an average person (shown is a selection of student personality trait averages). Participants can use these RPs on each trait to compare this average rating with their current estimate for a specific person. Traits are ordered based on the Big-Five Factors (different shades of grey). Granularity (G) refers to the level of detail in the represented structure of others’ personality traits. The granularity matrix generalizes the PEs across similar items in two distinct ways: for coarse granularity it generalizes per Big-Five factor, and for fine granularity it updates every individual trait based on how correlated they are to the current trait. b Using both RPs and granularity the models can be divided into three sets, which are depicted in three different colours. First, No Learning (blue), consists of a single regression model, Model 1 [No Learning] that functions as a baseline model. Second, Coarse Granularity (pink), updates based on the (Big-Five) factor to which the current adjective belongs. Model 2 [Coarse Granularity] uses the standard Rescorla–Wagner (RW) function to update the factor estimates and Model 3 [Coarse Granularity & Population RP] combines Model 2 with information from the RP. Third, Fine Granularity (orange), consists of two models that update all adjectives based on their correlation with the current trait. Model 4 [Fine Granularity] updates all items according to the Fine Granularity and Model 5 [Fine Granularity & Population RP] combines model 4 with information from the RP (see Supplementary Fig. 1 for details on the models). P prediction, Int intercept, RP reference point, α learning rate, PE prediction error, γ weighting parameter, F (generalizes over Factor) coarse granularity, All (generalizes over All items) fine granularity, SIM similarity matrix.
