Interpretation and trust

摘要: Statistical topic models can help analysts discover patterns in large text corpora by identifying recurring sets of words and enabling exploration topical concepts. However, understanding validating the output these itself be a challenging analysis task. In this paper, we offer two design considerations - interpretation trust for designing visualizations based on data-driven models. Interpretation refers to facility with which an analyst makes inferences about data through lens model abstraction. Trust actual perceived accuracy analyst's inferences. These derive from our experiences developing Stanford Dissertation Browser, tool exploring over 9,000 Ph.D. theses similarity, subsequent review existing literature. We contribute novel similarity measure collections notion "word-borrowing" that arose iterative process. Based literature review, distill set recommendations describe how they promote interpretable trustworthy visual tools.