During the perinatal period, psychosocial health risks, including depression and intimate partner violence, are associatedwith serious adverse health outcomes for both parent and child. To appropriately intervene, healthcare professionals mustfirst identify those at risk, yet stigma often prevents people from disclosing the information needed to prompt an assess-ment. We use techniques from natural language processing to indirectly identify psychosocial risks in the perinatal period.We apply latent Dirichlet allocation (LDA) and latent semantic indexing (LSI) to categorize themes from brief diary entriesby pregnant and postpartum women and apply sentiment analysis to characterize affect, then perform regularized regres-sion to predict diagnostic measures of depression and emotional intimate partner violence. Journal text entries quantifiedthrough sentiment analysis and topic models show promise for improved identification of depression and intimate partnerviolence, both stigmatized risks. Such methods may serve as an initial or complementary screening approach.

In this paper we demonstrate how to use graph matching touncover heterogeneity in the structure of preferences acrossa population of decision-makers. We propose a novel non-parametric approach to formally capture the concept of pref-erence structure using preference graphs, thereafter clusteringdecision-makers based on graph embedding methods. We ex-plore the approach with simulated choice and empirical datafrom the most common classes of economic and psychologicalmodels. The approach uncovers heterogeneity in preferencestructure across a variety of dimensions, without requiring anyprior knowledge of those structures.