The Many-Faced Plot: Strategy for Automatic Glyph Generation

The usage of glyphs related to the presented data is believed to help interpretation. We propose a strategy that allows the production of multi-purpose glyphs related to the data thematic (read the paper).


Data glyphs are used in data visualization for the representation of multidimensional data. They can be described as graphical objects that possess visual features, which can be assigned to data variables to produce a visualization. When considering iconic glyphs (e.g. faces, cars, or even flowers, see Fig. 1), they can be unrelated to the data thematic (e.g. a face glyph representing forest fires data) or be related in a literal (e.g. a face glyph representing data on facial features) or a metaphoric way (e.g. a face glyph representing non-facial anthropometric data).


Some authors state that using data-relatedness has perceptual advantages, which leads to easier interpretation and better accuracy. Despite this, glyphs are often used unrelated to the represented data. We address this issue by presenting an approach to implement an automatic glyph generation system. The approach takes advantage of the properties of emoji system and our ultimate goal is to make the glyph design process simple and effective, facilitating the communication of complex information


Figure 1

Examples of glyphs obtained with our system, their deconstruction, usage examples and comparison with glyphs used by other authors



Approach and resources


Glyphs are often produced using specialized programs, which have parametrizable functions to draw geometric shapes. It has been pointed out that it would be advantageous for a glyph-based visualization tool to have different types of glyphs, which could be chosen by the user and allow a better match to the data. Such tool is normally considered difficult to implement as it would require a large repository of glyphs prepared for data-representation. We believe that emoji have several properties which make them adequate for this task.


We propose an approach for a tool which can be used to generate ready-to-use glyphs for data visualization. In order to achieve this, we take advantage of the emoji connection between pictorial representation and associated semantic knowledge. This work builds upon the system behind Emojinating platform, which automatically retrieves emoji that represent concepts previously introduced by the user (see this post for a more detailed description). It uses the following resources:


  • the vector emoji images from Twitter1;
  • semantic knowledge of EmojiNet2;
  • and exploration of ConceptNet3.


Our end goal is to develop a semi-automatic system in which three general steps occur: (i) the user introduces the thematic; (ii) the system presents the user with possible glyphs for the introduced thematic, their visual variables and suggested variation ranges; (iii) the user selects the glyph and configures the assignment of data variables to visual variables, according to his/her preferences, and/or the data semantics.


System Architecture


The following 4-step pipeline is used:


  1. Identifying the topic: identification of the topic to be searched. This user-provided information is used to gather emoji to be glyph candidates. Depending on the type of data, the introduced topic may be a thematic (e.g. functioning mode a, described in subsection III-C) or categories of the data itself (e.g. functioning modes b and c);
  2. Glyph generation: generation of the glyphs to be used. This step can be divided into the following tasks: (i) Emoji deconstruction, (ii) Identification of visual variables, (iii) Removal of inadequate emoji, and (iv) Evaluation of visual variables;
  3. User configuration: the user is able to configure the assignment of visual variables, as well as, establish the variation limits, with the help of the suggestions of the system;
  4. Setup of the display: configuring the final view of the visualization, i.e. how the glyphs should be organized.


The system can be used in different ways, depending on data type. We considered three functioning modes (see Fig. 2): (a) Glyph as single emoji, (b) Glyph as set of emoji, and (c) Glyph as combination of emoji parts.


Figure 2

The three different functioning modes of the system



Analysis and Discussion


In order to analyze the potential of our approach in terms of usefulness in information visualization, we compared it with current glyph techniques. To do this, we conducted a bibliographic research and collected the following information from the gathered papers: type of glyph(s) used (e.g. car), number of total glyph visual variables, number of visual variables used, thematic of the dataset used and data variables represented.


Comparison with existing glyphs: we are able to obtain similar number of visual variables to the ones from existing glyphs and we consider our versions more visually appealing (see Fig. 1). Moreover, whereas implementation using functions allows greater flexibility, our approach allows higher variability of glyphs.


Thematic representation: we were able to obtain data-related candidates for all the thematics gathered, ranging from literal (e.g. a fire icon to represent fires) to metaphoric (e.g. a magnifying glass to represent Google search results). The system is even able to generate several possible glyphs for the same thematic.


Figure 3

Glyphs obtained for dataset thematics used by other authors


Open issues: we still have some issues that require further analysis regarding, for example, the transformation of the visual variables,  the suggestion of variation limits and assessment of salience and complexity.


Despite these open issues, in our opinion the major advantage of our approach is the automatic proposal of data- related glyphs, which we believe our system achieves. It is also important to mention that the emoji dataset can be easily changed, leading to different glyphs.


Read more about it!






[2] Wijeratne, S., and Balasuriya, L., Sheth, A., Doran, D. 2017. EmojiNet: An Open Service and API for Emoji Sense Discovery. In 11th International AAAI Conference on Web and Social Media (ICWSM 2017). Montreal, Canada; 2017.

[3] Speer, R., and Havasi, C. 2012. Representing general relational knowledge in conceptnet 5. In LREC, 3679–3686.




  • J. M. Cunha, E. Polisciuc, P. Martins, and P. Machado, “The Many-Faced Plot:Strategy for Automatic Glyph Generation,” in Proceedings of the 22st International Conference Information Visualisation (IV), 2018, 2018.