Automatic Generation of Semantic Icon Encodings for

Figure 1. Left: Carl Sagan’s scatter plot shows the ratio of brain to body mass on a log-log scale. This ratio is a good way to measure intelligence. The Sagan view shows that labels can make it difficult to see the position of the marks. Right: Icons are a good way to indicate the semantics of marks, as the human visual system is very effective at seeing icons. Here, these icons were automatically generated by our system. Authors use icon encodings to indicate the semantics of cat-egorical information in visualizations. The default icon li-braries found in visualization tools often do not match the se-mantics of the data. Users often manually search for or create icons that are more semantically meaningful. This process can hinder the flow of visual analysis, especially when the amount of data is large, leading to a suboptimal user experi-ence. We propose a technique for automatically generating semantically relevant icon encodings for categorical dimen-sions of data points. The algorithm employs natural language processing in order to find relevant imagery from the Internet. We evaluate our approach on Mechanical Turk by generating large libraries of icons using Tableau Public workbooks that represent real analytical effort by people out in the world. Our results show that the automatic algorithm does nearly as well as the manually created icons, and particularly has higher user satisfaction for larger cardinalities of data. Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored