The term “visualisation”, as used in this thesis, refers to techniques to illustrate the findings of data mining activities. Visualisation often acts as a powerful analytical tool to communicate and present data mining results. This section describes some related work on visualisation in DM in Sub-section 2.8.1. Then Sub-section 2.8.2 introduces the Visusetvisualisation tool, which was extended to support the work described in this thesis.
2.8.1 Visualisation in DM
The objective of visualisation tools in DM is to support user understanding of the end results. The output of DM processes can be complicated to comprehend. One of the often identified issues in DM that needs to be addressed is the visualisation of discovered knowledge [28]. According to a survey conducted by Rexer Analytics [4], a group of data miners noted that they faced challenges in explaining the essence of DM results. Han and Gao [50] also pointed out that effective and efficient visualisation tools should be investigated further to determine how they might support the analysis of DM results.
It is generally acknowledged that the visualisation of DM results should serve to enhance the users’ understanding [123]. The users’ need to understand the context
of the discovered “hidden” knowledge within the datasets, for example what is the relationship and causal association between attributes? Thus, the visualisation or rep- resentation of data mining output should be meaningful. Also, users should be able to interact with the visualisation so as to get further clarification of the results. There are a number of DM software systems, for example WEKA, that include facilities to allow data miners to visualise DM results [136]. MineSet is a DM visualisation tool developed by Silicon Graphics in 1996 [73]. The JUNG programming toolkit was introduced to provide visualisation support for social network mining [93]. JUNG stands for Java Universal Network/Graph, and is a Java library that provides several algorithms that allow social network miners to visualise dynamic graphs by adding or removing nodes and links.
There is also some reported work on data visualisation of temporal data and trends [6] and cluster change [36]. Jung [60] proposed a technique for the visual illustration of recommender system to help users to make more effective decisions. Another study, Rossolet al. [111], recommended the usage of a 3D framework for real-time geospatial temporal visualisation by evaluating livestock movement data for tracking and simu- lating the spread of epidemic diseases. The significance of the latter is that live stock tracking is one of the exemplar applications considered in this thesis.
The work described in this thesis, implements three visualisation methods for: (i) the visualisation of large numbers of frequent pattern trends in terms of trend clusters, (ii) visualising communities of clusters in the context of trend migration and (iii) vi- sualising the predicted migration of trends. This thesis utilisesVisuset to illustrate the outcomes from the proposed social network mining.
2.8.2 VISUSET
Visuset is a 2-D visualisation software tool that was developed for chance discovery [90]. It represents node communities, using a 2-D drawing area, based on the Spring Model [120]. It highlights which nodes are connected directly and indirectly with other nodes in detected communities which are depicted as “islands”. Nishikidoet al. [98] presented Visuset as an animation interface to illustrate change points in keyword relationship networks. This was considered to be a chance discovery tool because it discovered significant candidates (keywords) that benefited the utilisation and selection process. Visuset provides a clear animation of communities of clusters to highlight which clusters connect to which clusters. The orginal Visuset software was developed by the Hiroshima City University research group. The version of Visuset used with respect to the work described in this thesis is a version that the author has extended, in collaboration with Wataru Sunayama of Hiroshima City University, to provide animation and visualisation support for trend cluster analysis that illustrates significant dynamic cluster changes in sequences of data. Visuset is available for research purposes by contacting the author.
There are alternative visualisation tools that could have been adopted with respect to the work described in this thesis. For example, Kandogan [62] developed a system to display multi-dimensional data in a two dimensional surface as a scatter plot. However, no indication was given of the inter relationships between data points. Visuset groups data into “islands”, data within an island is closely linked according to co-relationship values. Visuset thus highlights the nature of the groupings that exist and how the data is correlated. Havre et al. [54] described a technique for displaying thematic changes as river flows, so that changes of topics can be observed. However, unlike Visuset, the relationships between topics are not considered. Chen [26] described a system to visualise a network so as to identify emerging trends. However, the network is displayed with respect to a specific time stamp. Therefore changes in trends cannot be easily observed. The extension of Visuset described later in this thesis illustrates trend transitions in the form of an animation so as to demonstrate how trends change over a given period of time. Robertsonet al. [109] introduced a system to also show trends by animation. Their method illustrated changes in the data in the form of traces, but their changes are considered independently. In the proposed Visuset approach, trends are correlated against one another so that observers can see how groups of trends change with time.