Mountaz Hascoët

Cluster validity indices for graph partitioning

The aim of graph clustering is to define compact and well-separated clusters from a given graph. Clusters compactness depends on datasets and clustering methods. In order to provide evaluation of graph clustering quality, many different indices have been proposed in previous work. Indices are used to compare different graph partitions but also different clustering techniques. Moreover, some clustering techniques are based on index optimization. Indices can also be added as visual tips in graph layouts. Despite the importance of the subject, little has been done to unify the field. It results that many indices can not be easily compared or interpreted. In this paper, we provide a unified and synthetic view of indices used in graph clustering area and discuss them. We also propose several enhanced measures.

Extending drag-and-drop to new interactive environments: A multi-display, multi-instrument and multi-user approach

Drag-and-drop is probably one of the most successful and generic representations of direct manipulation in todays WIMP interfaces. At the same time, emerging new interactive environments such as distributed display environments or large display surface environments have revealed the need for an evolution of drag-and-drop to address new challenges. In this context, several extensions of drag-and-drop have been proposed over the past years. However, implementations for these extensions are difficult to reproduce, integrate and extend. This situation hampers the development or integration of advanced drag-and-drop techniques in applications. The aim of this paper is to propose a unifying implementation model of drag-and-drop and of its extensions. This model-called M-CIU-aims at facilitating the implementation of advanced drag-and-drop support by offering solutions to problems typical of new emerging environments. The model builds upon a synthesis of drag-and-drop implementations, an analysis of requirements for meeting new challenges and a dedicated interaction model based on instrumental interaction. By using this model, a programmer will be able to implement advanced drag-and-drop supporting (1) multi-display environments, (2) large display surfaces and (3) multi-user systems. Furthermore by unifying the implementation of all existing drag-and-drop approaches, this model also provides flexibility by allowing users (or applications) to select the most appropriate drag-and-drop technique depending on the context of use. For example, a user might prefer to use pick-and-drop when interacting with multiple displays attached to multiple computers, push-and-throw or drag-and-throw when interacting with large displays and possibly standard drag-and-drop in a more traditional context. Finally, in order to illustrate the various benefits of this model, we provide an API called PoIP which is a Java-based implementation of the model that can be used with most Java-based applications. We also describe Orchis, an interactive graphical application used to share bookmarks and that uses PoIP to implement distributed drag-and-drop like interactions.

Interactive graph matching and visual comparison of graphs and clustered graphs

We introduce interactive graph matching, a process that conciliates visualization, interaction and optimization approaches to address the graph matching and graph comparison problems as a whole. Interactive graph matching is based on a multi-layered interaction model and on a visual reification of graph matching functions. We present three case studies and a system named Donatien to demonstrate the interactive graph matching approach. The three case studies involve different datasets a) subgraphs of a lexical network, b) graph of keywords extracted from the InfoVis contest benchmark, and c) clustered graphs computed from different clustering algorithms for comparison purposes.

Focus dependent multi-level graph clustering

In this paper we propose a structure-based clustering technique that transforms a given graph into a specific double tree structure called multi-level outline tree. Each meta-node of the tree - that represents a subset of nodes - is itself hierarchically clustered. So, a meta-node is considered as a tree root of included clusters.The main originality of our approach is to account for the user focus in the clustering process to provide views from different perspectives. Multi-level outline trees are computed in linear time and easy to explore. We think that our technique is well suited to investigate various graphs like Web graphs or citation graphs.

Focus-based clustering for multiscale visualization

Previous works on the visualization of navigation history has provided users with overviews that facilitate the retrieval of already visited pages. Most of these techniques rely on the construction of a navigation tree without accounting much for the structure of the underlying hypertext graph. Therefore such views are heavily dependent on the order of page visited. In most cases, they are limited to one session and cannot easily represent large collections of pages. We propose a clustering method based [L. Tauscher et al., (1997)] on a focus [M. Hascoet] on the structure of the underlying hypertext graph. Our aim is to use this method to automatically organize visited pages into a hierarchically clustered graph to provide multiscale visualization of pages visited during several sessions. Our initial experiments with navigation history are promising. We believe that our clustering method is general enough to be applied to other data with an underlying highly connected graph structure.

CoViz: Cooperative Visualization to Facilitate Sense Making by Groups of Users

Coviz is a tool that uses visualization to support spontaneous construction, organization and exploration of collections. We use the term collection to identify a set of documents gathered for a given purpose by one or several individuals. Coviz facilitates creation, organization and update of collections by providing multi-scale visualization and interaction techniques. Coviz has been tested in several informative case studies involving small groups of users. In these situations, Coviz has fulfilled several important aspects of both groups and personal organization needs. Our initial experiments indicate that Coviz can even encourage unusual and interesting forms of cooperation among people.

Multi-level Exploration of Citation Graphs

In previous work, we proposed a focus-based multi-level clustering technique. It consists in computing a particular clustered graph from a given graph and a focus. The resulting clustered graph is called multi-level outline tree. It is a tree whose meta-nodes are sub-sets of nodes. A meta-node is itself hierarchically clustered depending on its connectivity. In this paper we introduce a cluster cohesiveness measure to enhance the results of the previously proposed algorithm. We further propose an optimization of this algorithm to support fluid interaction when focus changes. Finally, we report the results of a case study that consists in applying the enhanced algorithm to citation graphs where documents are considered as vertices and citation links as edges.

Synchronous cooperation and visualization for social bookmarking systems

In this paper, our aim is to facilitate synchronous and co-present interaction with social bookmarking systems for groups of related users meeting to discuss and share their collections of tags and bookmarks. Our work results in a system called Orchis that proposes a graphical user interface based on cooperative visualization and interaction as an alternative graphical user interface for social bookmarking systems. Orchis presents three major characteristics: (1) graphical overviews of collections of annotated bookmarks and tags, (2) advanced drag-and-drop interaction styles adaptable to distributed display environments and (3) support for distributed architectures possibly running different windowing systems. Our hypothesis is that by using Orchis, related users will be able to better compare and share tags and bookmarks. They will also be able to build cooperatively valuable shared collections. We expect that, in turn, this will participate in improving the overall quality of both folksonomies and social bookmarking collections.