Predicting Semantic Signatures of Fonts

Towards the aim of semantic font recommendation, we first analyze the relationship between fonts and semantic attributes using a crowdsourced dataset. We deepen this analysis at the level of font categories and font styles, including via a series of interactive visualizations of the relationships between multiple dimensions of this data. Subsequently, we induce semantic signatures for a large number of fonts by computationally predicting attribute values using a k-NN based approach. We evaluate the effectiveness of this approach, and, through a visual exploration, categorize semantic attributes into three groups based on their potential to be conveyed by the fonts. The resulting resource is made available, and we aim for the findings, visualizations, and data to benefit studies that computationally support the challenging but critical process of font selection.