FanLens: Dynamic Hierarchical Exploration of Tabular Data Xinghua Lou ∗

Shixia Liu †

Tianshu Wang ‡

IBM China Research Lab

A BSTRACT Tabular data is a very popular format for storing information from various domains. However, as the data grows in size, it becomes increasingly difficult to discover its intrinsic structure or see the comparison among cells from the traditional column-row presentation. To address the problem, we propose an enhanced technique based on traditional radial, space-filling visualization (e.g. Sunburst) named FanLens, which helps users to explore tabular data by dynamically specifying the hierarchies and then visualizing them. Our work is an improvement upon existing approaches in terms of flexibility, context preservation and interaction.

• High-level start-up When visualizing the hierarchy, do not lay out the entire hierarchy initially but display only several high levels showing the summarization information (Figure 1(a)). • Expanding/collapsing mechanism Users can drill down into lower levels by expanding one branch from the higher level. The newly expanded branch will be incrementally laid out around the periphery of its parent slice, radially, and is regarded as the focus (Figure 1(b)(c)(d)).

Keywords: Tabular data visualization, dynamic hierarchy specification, radial space-filling visualization, fisheye distortion. 1

I NTRODUCTION

Tabular data is usually a matrix-like structure of columns and rows containing data cells and is one of the most popular formats for data storage. Although the table is the simplest and most straightforward way to present the tabular data, it becomes awkward when data volume increases, since there is no emphasis on its organization or the distribution. [1]. Here we present FanLens, which enables dynamic exploration of tabular data with an incremental, radial space-filling visualization technique. Compared with traditional radial space-filling visualization (e.g. Sunburst [2]), our technique can provide more flexibility and better preserve the context. The typical thin slice problem is addressed with a new fisheye distortion based interaction. 2

DATA T RANSFORMATION

Generally speaking, the tabular data can be divided into two categories, categorical data and quantitative data. Naturally, hierarchy can be structured by breaking down the tabular data in order of its categorical data. FanLens further improves this feature by allowing the users to bin the quantitative data into different ranges and impose the results on structuring the hierarchy and by allowing grouping the categorical data by the instance of its value. FanLens also supports dynamic visual data transformation, namely two separate dimensions of data can be mapped to the angle and color of the slices in the visualization, respectively. The source of the mapping could be one attribute or a mathematical expression of some selected attributes. 3

V ISUALIZATION D ESIGN

3.1 Incremental Layout Incremental layout is the primary feature of FanLens, which follows two principles: ∗ [email protected] † [email protected] ‡ [email protected]

Figure 1: Examples of layout principles. (a) High-level start-up; (b)(c)(d) Expanding/collapsing mechanism.

This incremental layout firstly brings flexibility. Users can focus on the branch of interest and also see the overview by redefining the base levels to cover the entire hierarchy (which creates the classic Sunburst visualization). The readability is also improved because of the expanding/collapsing mechanism which offers the users with a clear view of the exploration path and structure of the focus. 3.2 Zooming and Picking Zooming is used to deal with the thin-slice problem and is implemented by enlarging the sweep angle of the focus so all the thin slices in it are enlarged as well. Former designs [2, 3] applying this method have difficulties with preserving the context. Our solution follows the same idea but, benefiting from the expanding/collapse mechanism, preserves the context better because the focus is enlarged where it is and need not be repositioned (Figure 2). Former solution to picking is zooming before picking, which surely works but also lower the efficiency. Sometimes users need to locate one slice quickly and precisely to find some detailed information or drill down from it. We brought an interesting solution to this problem by applying fisheye transformation to distort the angles of slices in the focus. The goal is to ensure that users can have a clear view of this current selection. The transformation algorithm is prepared when the mouse is moving within the focus, but is only executed when it gets close to or into a thin slice (Figure (3)).

Figure 2: An example of focus zooming.

Figure 5: Explore the 3-point shooting abilities and find the special pattern of the Phoenix Suns team.

Figure 3: (a) Picking without fisheye transformation. (b) Picking with fisheye transformation.

4 C ASE S TUDY The case contains the statistics of NBA players for one season. 4.1 Overall Evaluation Figure 4 shows the result of structuring the data in order of Conference, Division, Team and Player and defining the base levels to cover the entire hierarchy. The slice angle and slice color are mapped to player’s offensive ability and defensive ability, respectively. This overall evaluation indicates that, even though player’s offensive and defensive abilities vary a lot, the league is quite balanced in all levels of Conference, Division and Team. That is one important reason why NBA game is usually exciting because it is always a close matchup.

4.3 Hypothesis Testing We may also use Figure 4 to analyze player’s scoring ability (PPG) and mistakes (TO, Turnover); however, this visualization is not really intrinsic but guides us to hypothesize that players with high scoring ability also have more turnovers. To test this hypothesis, we specify the hierarchy by ranging the players into several categories according to their PPG (see the following table) and visualizing the new hierarchy with the same visual presentation configuration, as shown in Figure 6(b). This new visualization proves our hypothesis that players with stronger scoring ability also give more turnovers.

Figure 6: Testing the hypothesis on players’ scoring ability and turnovers.

Figure 4: Use overview to evaluate the balance of the NBA league.

4.2 Special Pattern Discovery Figure 5 shows the same hierarchy but is dedicated to analyze the 3point shooting ability. The angle is mapped to 3PM(3-Points Made per game) and the color is mapped to 3P%(3-Point shooting Percentage). The large angle of Pacific Divison guides us to explore it and find the team Phoenix Suns which is actually wild about shooting 3-points. In addition, a special pattern is noted that most of its players have close 3-Point shooting percentage which explains why they dare shooting so many 3-point baskets.

5 C ONCLUSIONS We have introduced the FanLens, an approach for dynamic and hierarchical exploration of tabular data. Our primary contribution is an incremental, radial space-filling visualization technique which better preserves the context, provides extra flexibility and has an unique fisheye transformation supported picking interaction. In the future, we plan to enrich the interactions, e.g. Drag and Drop operation on the display to specify the hierarchy and multiple expansion to compare different branches. We also intend to support automatically hierarchy specification using methods such as SOMs and clustering. R EFERENCES [1] G. Chintalapani, C. Plaisant, and B. Shneiderman. Extending the utility of treemaps with flexible hierarchy. In Proc. of the Eighth International Conference on Information Visualisation, pages 335–344, 2004. [2] J. Stasko and E. Zhang. Focus+context display and navigation techniques for enhancing radial, space-filling hierarchy visualisations. In IEEE Symposium on information visualisation, pages 9–12, 2000.