Knowledge Representations in Health Judgments

AbstractIn the present paper, we introduce a novel computational approach for uncovering mental representations underlying healthiness judgments for food items. Using semantic vector representations derived from large-scale natural language data, we quantify the complex representations that people hold about foods, and use these representations to predict how both lay decision makers and experts (trained dietitians) judge the healthiness of food items. We also successfully predict the impact of behavioral interventions (e.g. the provision of nutrient content information or “traffic-light labels”) on healthiness judgments for food items. Our models are highly general, and are capable of making predictions for nearly any food item. Finally, these models outperform competing models based on factual nutritional content, suggesting that health judgments depend more on complex (semantic) knowledge representations than on quantified nutritional information. The results in this paper illustrate how methods from cognitive science and computational linguistics can be combined with existing theories in psychology, to better predict, understand, and influence health behavior.


Return to previous page