Recommender Systems for Literature Selection: A Competition of Decision Making and Memory Models

Abstract

We examine the ability of five cognitive models to predict what publications scientists decide to read. The cognitive models are (i) the Publication Assistant, a literature recommender system that is based on a rational analysis of memory and the ACT-R cognitive architecture; (ii-iv) three simple decision heuristics, including two lexicographic ones called take-the-best and naïveLex, as well as unit-weight linear model, and (v) a more complex weighted-additive decision strategy called Franklin’s rule. In an experiment with scientists as participants, we pit these models against (vi) multiple regression. Among the cognitive models, take-the-best best predicts most scientists’ literature preferences best. Altogether, the study shows that individual differences in scientific literature selection may be accounted for by different decision-making strategies.


Back to Saturday Posters