Abstract:
- In this project, we present a new
algorithm for visualization of sets, focusing on the law case data. The
data consist of a collection of cases, each referring to a set of
cases.Thus, the aim is the cluster the cases based on the cases they
refer to.
Clustering of data is mainly grouping of data items, to get high intra-cluster similarity. In this case, the similarity is a measure of the common references of the cases. We present a top-down algorithm to cluster the cases, which builds a tree, starting at the root which is a cluster of all the cases and recursively dividing the clusters. The clusters at each level are created such that the common references of the cluster increase by one at each level.
Thus, cases referring to 'd' common cases will be grouped into one cluster at depth 'd' in the tree. The nodes at any level can be a clustering result since the clusters at each level are disjoint set of elements. The clustering can be seen in the treeview visualization. We also provide two more visualizations as a part of this project. The histogram view and a co-citation matrix. The co-citation matrix shows the count of] co-citation of every two cases.
Project Report
Project presentations:
Download Application
Video Demo produced with Camtasia
Screen shots: (click to enlarge)
Figure 1 - A screenshot of the CaseCluster interface, with the TreeView activated. Each node in the tree is labeled with the common references that are unique for this node and shows in parenthesis the number of cases that have these common references. Moving from the root on the left to the farthest node on the right the number of common citations is increasing with each level. Clicking on a node displays all the cases citing the common references on this node in the Case Information panel in the lower right. The nodes in gray contain cases that have less common references than the level they are attached, they never have children. Looking at the tree it becomes obvious that there are fewer nodes that represent more common citations, by seeing that the number of nodes decreases with each depth level. The sliders on the upper right of the interface can be used to modify properties of the tree.
Figure 2 - A screenshot of the Histogram view, the case with the highest citation count is selected (colored marker) and the cases which reference it are shown in the detail view on the lower right. Looking at the distribution of the cases in this view starting from the left we can see two cases with very high citation frequency, then two with a little less, then four group together, then six and then we have a big collection of cases that are less frequent cited.
Figure 3 - A screenshot of the CoCitationMatrix view for 64 cases, one entry is selected (markers in row and column) and the cases which reference these two cases together are shown in the detail view on the lower right. A tool tip shows the exact co-citation frequency. The darker a square is the higher us its co-citation frequency. From this view we can see that if a case has a high co-citation frequency then not only for one other case but often for most of its column. For clarity only the left triangle of the matrix is shown since it is identical to the right triangle.