1 Introduction
-
skimming texts such as lead paragraphs;
-
encoding quantitative data in search lists using the novel technique of proportional encoding;
-
profiling metadata associated with topics, entities and facets via extensions to stem and leaf plots; and
-
multivariate labels for data-dense knowledge maps.
2 Existing KM and IR interfaces
2.1 Problems with KM and IR visualizations
[Compared to tag clouds] a one paragraph summary would probably be more enlightening, faster to scan, take up less screen space allowing for more items to be summarized on any given page [45].
2.2 Why font attributes
-
http://Scimaps.org is a repository of knowledge maps—only 28 % of these visualizations use font-specific attributes, and in most cases they only differentiate between compositional elements (e.g. tick labels, item labels, legend).
-
Similarly, in Hearst’s examples of IR infovis, only 29 % use font-specific attributes [27].
-
Font attributes such as bold, italic and case are not available to other glyphs.
-
Literal encoding with text is unambiguous compared to pictographic glyphs [3].
-
Letter order with text can be used to created ordered representations (i.e. alphabetic order, such as an index).
-
Glyph design can be difficult, particularly when there are many categories to encode. Corresponding text does not require a design task.
3 Font attributes across domains
4 Font attribute capabilities
-
Italic or oblique are both sloped fonts but italics have different letterforms. Sloped fonts vary in slope angle including instances of reverse italics and even vertical italics [36]. Slope can be used to encode a diverging scale, ranging from reverse, to vertical, to forward slope.
-
CASE includes UPPER, lower, Mixed and Small Caps. Uppercase is designed to standout from lowercase while small caps blend in. Case is sometimes used in cartography for ordered data, e.g. states in uppercase and counties in lowercase [34].
-
Typeface indicates font family, e.g. sans, , ,
source
, , etc. Typeface is best used to symbolize categoric information. Many typographic and cartographic references suggest using no more than two typefaces although there are examples with more (e.g. Fig. 7). Historically there was bias to using simpler sans serif fonts on screen due to limited resolutions but more detailed fonts are now used in web design, e.g. [8, 41].××× -
can be . Many typographers and cartographers recommend against underline or . Underlines on or case do not interfere with descenders. Underlines can express ordered data (e.g. as in Fig. 7).××××××
-
of a string is used in cartography to indicate the range of an area feature, and thus has possibilities for encoding quantities. Different ways to set width include:
-
are specifically designed widths of a given typeface for applications such as tight spaces. Few fonts are available in a range of widths.×
-
are horizontally scaled fonts. Typographers recommend against these distortions.×
-
S p a c i n g includes adjusting the space between letters (tracking) and between lines (leading).
× -
-
encode via size and position relative to adjacent text. They can be used to encode a high number of categories (e.g. Fig. 10 in [56]).×
-
‘Paired delimiters’ evoke enclosure by pairing the same (or mirrored) shapes, e.g. ([ ],{},“ ”,* *, etc.)
-
Alphanumeric glyphs (A, B, C, 1, 2, 3) can literally encode data and are uniquely orderable. Glyphs not native to the viewer (e.g. \(\alpha , \beta , \gamma \)) are also orderable, but symbols (e.g. \(\infty , \forall , \flat \)) are not orderable. While other font attributes may visually be perceived without active attention for quick perception, words and phrases must be actively read which is slower.
×
|
5 Fonts for knowledge maps and information retrieval
5.1 Font visualization on texts
5.1.1 Skim formatting of previews
-
“Can you install this on my iPad now?”
-
“I can see using this immediately in my own visualization research.”
-
“This is similar to how we used multiple underlines in our paper textbooks in college.”
-
“The ability to toggle is key: people who consume news all day will need to move back and forth between reading and skimming.”
-
“The technique can work well by aiding recognition of keywords instead of relying on searching (recall).”
-
“Perhaps the same technique could be used to make the words pop-out that make the text more memorable, the way that Kennedy or Martin Luther King used spoken emphasis on words.”
5.2 Font visualization on lists
5.2.1 Query result lists and proportional encoding
Variant | Normal | Dense | Sparse |
---|---|---|---|
Font weight | 1.42\(\times \)
| 0.67\(\times \)
| 0.86\(\times \)
|
Proportion | 2.68\(\times \)
| 2.22\(\times \)
| 2.07\(\times \)
|
-
The amount of bold, per movie, is indicative of the overall rating. In this sample it can be seen that Despicable Me 2 has less bold than How to Train your Dragon indicating a lower rating.
-
Within a single movie, the slope formed by the boundary between bold and non-bold text provides an indication of dispersion. Between movies, the slope can be compared, e.g. Despicable Me 2 has a shallower slope than How to Train your Dragon indicating higher dispersion for the former.
-
The initial portion of the reviewer’s opinion can be read and compared, for example, the worst review for each movie at the top; or the best reviews for each movie at the bottom.
5.2.2 Word lists and stem and leaf text plots: for facets, categories, topics, entities, keywords, descriptors, etc.
5.3 Font visualization on macro-views
5.3.1 Multi-attribute labels
Task | Choropleth map (%) | ISO code map (%) | ISO code performance |
---|---|---|---|
Identify | 15 | 65 | 4.4\(\times \)
|
Locate | 53 | 85 | 1.6\(\times \)
|
Total | 34 | 75 | 2.2\(\times \)
|
5.3.2 Multi-level labels and font attributes
-
Skupin’s self-organizing maps [53] labels with size indicating topic scale and breadth. The labels, regardless of size, do not overlap.
-
Paley’s graph of science [47] uses tiny discrete labels for specific topics overlaid on top of large labels for broad topics.
-
Boyack and Klavans [9] use colour to indicate broad topics and overlay labels to call out specific items of interest (e.g. emerging topics).
Map variant | # Countries discernable | # Attributes | Relative info density |
---|---|---|---|
Choropleth 1 var | 163 | 1 | 1.0 |
Mnemonic 1 var | 187 | 1 | 1.15\(\times \)
|
Mnemonic 4 vars | 187 | 4 | 4.6\(\times \)
|
-
Node colour represents news sentiment over the day per company (a red to green gradient).
-
Label weight indicates news volume: heavier than normal news is set in heavier font weights. Industry and sector label weights indicate net news volume for constituents.
-
Label italic angle and colour indicate sentiment. This is a double encoding: the same variable is encoded to two different visual attributes reinforcing each other.
-
Individual company labels are represented only if their news volume is over a user-defined threshold, are set on a dark background overlaid on the base map.
-
Uppercase company labels indicate companies with fresh news stories in the last 10 min.
6 Discussion
-
“This represents a whole new way of thinking about type.”
-
“These are very creative new uses of typography.”
-
“This is what typographers have always done: changing type attributes for different applications. What is new is the application of type to show quantities.”
-
“Type attributes visually work as validated over centuries of empirical refinement. These are new applications of proven techniques.”
-
“Multivariate labels are engaging and can stimulates analysis.”
-
“Mnemonic labels are interesting: if you do not recognize a code immediately, you have still got the code and adjacent codes that you can use as a cue to search your memory.”
-
“You have to be careful with the encoding. If the encoding is intuitive as in these examples, it makes the application easy to understand.”
-
“When you change the font, you change the semantics. So using the right attributes for the target application is required to express what you are trying to encode.”
-
“The skim format approach could instead encode semantics. For example, comics use conventions for shout, whisper, and so on.”
-
“Multivariate encoding works well for labels, but for prose readability may be impacted.”
-
“You have to be careful when mixing many attributes together in prose. Readability can be impacted.”
7 Conclusion
-
Paragraph and document views via skim formatting (Fig. 8).