Computational approaches for semantics-aware typographical choices

Typographic signals carry strong semantic connotations, e.g., they may convey excitement, anger or even sweetness, which empowers them to affect almost any aspect of life, from perception of an email, to the perceived sweetness of a cup of coffee. This thesis explores some of the possibilities that can be offered by computational approaches to support users in understanding and taking advantage of this impact. More specifically, the focus is on learning font semantics from crowdsourced and Web data, and using this information to facilitate font search and recommendation. Among the novel contributions are the use of CNN-based embeddings to represent fonts in attribute learning, leveraging emotion theories and lexical relations to infer font semantics, a multimodal font search method that allows specifying a reference font together with the target semantic additions, enabled by a cross-modal representation of fonts and words, and the proposal of affect-aware word clouds that let users specify a target emotion, which is used to recommend fonts and color palettes with congruent affective connotations.Ph.D.Includes bibliographical reference