Marshall Taylor

Numerous studies have been conducted to identify the factors that predict trust/distrust in science. However, most of these studies are based on closed-ended survey research, which does not allow researchers to gain a more nuanced understanding of the phenomenon. This study integrated survey analysis conducted within the United States with computational text analysis to reveal factors previously obscured by traditional survey methodologies. Even after controlling for political ideology—which has been the most significant explanatory factor in determining trust in science within a survey framework—we found those with concerns over boundary-crossing (i.e. concerns or perceptions that science overlaps with politics, the government, and funding) were less likely to trust science than their counterparts.

Distances derived from word embeddings can measure a range of gradational relations—similarity, hierarchy, entailment, and stereotype—and can be used at the document- and author-level in ways that overcome some of the limitations of weighted dictionary methods. We provide a comprehensive introduction to using word embeddings for relation induction, and demonstrate how such techniques can complement dictionary methods as unsupervised, deductive methods.