The number of possible visualisation techniques that can be applied to different appli- cations is big. However, there are several visualisation methods that have been well investigated and applied successfully to a wide variety of tasks. These methods include graphs, histograms, circle graphs, self-organising maps, hyperbolic trees, treemaps, fish- eye graphs and menus, scatterplots, as well as hybrid forms (Card et al., 1999; Feldman and Sanger, 2007; Heer et al., 2005). Card et al. (1999) is a useful resource on information visualisation research. Further, Noah Iliinsky (2012) provides valuable guidelines on best uses of visual encodings and their properties given the nature of the data (see Appendix D) and on the design of data visualisations (Iliinsky and Steele, 2011), as well as examines various case studies and their approaches to projects from a variety of perspectives (Steele and Iliinsky, 2010). Below, we briefly discuss several visualisation practices on different tasks.
In recent years, several studies have emerged that involve visualisation of natural lan- guage, and are perhaps more related to our research than others; for example, Lyding
et al. (2012) use Structured Parallel Coordinates (SPCs) (Culy et al., 2011) to visualise diachronic changes in academic discourse, in terms of lexicogrammatical features of regis- ters. SPCs involve the use of axes, each representing different data dimensions, while data points represented on the axes are connected with lines to visualise their relationships. Van Ham et al. (2009) introduce Phrase Net, a system that analyses unstructured text by taking as input a predefined pattern and displaying a graph whose nodes are words and whose edges link the words that are found as matches. Users can interactively specify a pattern or choose from a list of default ones. Patterns can take the simple form of Xand Y or Xor Y , or be defined using regular expressions. Another visualisation technique which is popular for representing information in texts is “word clouds”. Vi´egas et al. (2009) investigate Wordle, a tool for making “word clouds”, that is, graphic statements in which words are packed tightly and can be placed vertically, horizontally or diagonally. Further, the colour and size of the words can be used to represent different types of information, for example, frequent words may be given more prominence via using larger fonts.
Collins (2010) in his dissertation addresses different visualisation techniques for nat- ural language processing (NLP) research. The Bubble Sets visualisation draws secondary set relations around arbitrary collections of items, such as a linguistic parse tree. Vis- Link provides a general platform within which multiple visualisations of language (e.g., a force-directed graph and a radial graph) can be connected, cross-queried and compared. Moreover, he explores the space of content analysis using DocuBurst, an interactive vi- sualisation of document content, which spatially organises words using an expert-created ontology (e.g., WordNet). Parallel Tag Clouds combine keyword extraction and coor- dinated visualisations to provide comparative overviews across subsets of a faceted text corpus. Recently, Rohrdantz et al. (2011) proposed a new approach to detecting and investigating changes in word senses by visually modelling and plotting aggregated views about the diachronic development in word contexts.
Visualisation techniques have been successfully used to support humanities research (e.g., Plaisant et al., 2006 and Don et al., 2007), as well as genomics (e.g., Meyer et al., 2010a and Meyer et al., 2010b). For example, Don et al. (2007) have developed a system that visualises the distribution of frequent patterns found in text collections, displays their context and supports analysis of their correlations. Plaisant et al. (2006) have built a user interface which aids the interpretation of literary work by integrating text mining algorithms. Their system allows visual exploration of documents, preparation of training sets and reviewing of classification algorithm results. Meyer et al. (2010a) present a system that supports the inspection and curation of data sets showing gene expression over time, in conjunction with the spatial location of the cells where the genes are expressed.
Graph-based visualisations, which we adopt in our work, have been used effectively in various areas. As Herman et al. (2000) note, “The area of graph visualization has reached a level of maturity in which large applications and application frameworks are being developed. However, it is difficult to enumerate all the systems because of the sheer quantity”. An overview on graph visualisation methods and systems is beyond the scope of this thesis. However, recent examples that are similar to ours, design-wise, include the analysis of domains such as social networks to allow for a systematic explo- ration of a variety of Social Network Analysis measures (SNA). Gao et al. (2009) present MixVis and Perer and Shneiderman (2006) SocialAction, two systems designed to help structural analysts examine social networks (e.g., a terrorism network). Both tools allow
systematic exploration of SNA measures17 by linking together the statistical and visual components of a network. Heer and Boyd (2005) have implemented Vizster, a visuali- sation system for the exploration of on-line social networks (e.g., Facebook) designed to facilitate the discovery of people, promote awareness of community structure, and so on. VisualComplexity.com is a unified resource space of projects regarding a variety of graph/network visualisation methods across different domains (Lima, 2011). Examples include visualising the Bible, Wikipedia, computer systems, food webs, semantic net- works, topic shifts, and so forth. Useful resources on more technical details on graph drawing, visualisation and layout algorithms include Battista et al. (1994, 1998); Eades and Sugiyama (1990); Gansner et al. (1993); Herman et al. (2000).