Maoyuan Sun

BiSet: Semantic Edge Bundling with Biclusters for Sensemaking

Identifying coordinated relationships is an important task in data analytics. For example, an intelligence analyst might want to discover three suspicious people who all visited the same four cities. Existing techniques that display individual relationships, such as between lists of entities, require repetitious manual selection and significant mental aggregation in cluttered visualizations to find coordinated relationships. In this paper, we present BiSet, a visual analytics technique to support interactive exploration of coordinated relationships. In BiSet, we model coordinated relationships as biclusters and algorithmically mine them from a dataset. Then, we visualize the biclusters in context as bundled edges between sets of related entities. Thus, bundles enable analysts to infer task-oriented semantic insights about potentially coordinated activities. We make bundles as first class objects and add a new layer, “in-between”, to contain these bundle objects. Based on this, bundles serve to organize entities represented in lists and visually reveal their membership. Users can interact with edge bundles to organize related entities, and vice versa, for sensemaking purposes. With a usage scenario, we demonstrate how BiSet supports the exploration of coordinated relationships in text analytics.

A Five-Level Design Framework for Bicluster Visualizations

Analysts often need to explore and identify coordinated relationships (e.g., four people who visited the same five cities on the same set of days) within some large datasets for sensemaking. Biclusters provide a potential solution to ease this process, because each computed bicluster bundles individual relationships into coordinated sets. By understanding such computed, structural, relations within biclusters, analysts can leverage their domain knowledge and intuition to determine the importance and relevance of the extracted relationships for making hypotheses. However, due to the lack of systematic design guidelines, it is still a challenge to design effective and usable visualizations of biclusters to enhance their perceptibility and interactivity for exploring coordinated relationships. In this paper, we present a five-level design framework for bicluster visualizations, with a survey of the state-of-the-art design considerations and applications that are related or that can be applied to bicluster visualizations. We summarize pros and cons of these design options to support user tasks at each of the five-level relationships. Finally, we discuss future research challenges for bicluster visualizations and their incorporation into visual analytics tools.

Bixplorer: Visual Analytics with Biclusters

A prototype visual analytics tool uses data mining algorithms to find patterns in textual datasets and then supports exploration of these patterns in the form of biclusters on a high-resolution display.

The role of interactive biclusters in sensemaking

Visual exploration of relationships within large, textual datasets is an important aid for human sensemaking. By understanding computed, structural relationships between entities of different types (e.g., people and locations), users can leverage domain expertise and intuition to determine the importance and relevance of these relationships for tasks, such as intelligence analysis. Biclusters are a potentially desirable method to facilitate this, because they reveal coordinated relationships that can represent meaningful relationships. Bixplorer, a visual analytics prototype, supports interactive exploration of textual datasets in a spatial workspace with biclusters. In this paper, we present results of a study that analyzes how users interact with biclusters to solve an intelligence analysis problem using Bixplorer. We found that biclusters played four principal roles in the analytical process: an effective starting point for analysis, a revealer of two levels of connections, an indicator of potentially important entities, and a useful label for clusters of organized information.

Visualizing Traffic Causality for Analyzing Network Anomalies

Monitoring network traffic and detecting anomalies are essential tasks that are carried out routinely by security analysts. The sheer volume of network requests often makes it difficult to detect attacks and pinpoint their causes. We design and develop a tool to visually represent the causal relations for network requests. The traffic causality information enables one to reason about the legitimacy and normalcy of observed network events. Our tool with a special visual locality property supports different levels of visual-based querying and reasoning required for the sensemaking process on complex network data. Leveraging the domain knowledge, security analysts can use our tool to identify abnormal network activities and patterns due to attacks or stealthy malware. We conduct a user study that confirms our tool can enhance the readability and perceptibility of the dependency for host-based network traffic.

AVIST: A GPU-Centric Design for Visual Exploration of Large Multidimensional Datasets

This paper presents the Animated VISualization Tool (AVIST), an exploration-oriented data visualization tool that enables rapidly exploring and filtering large time series multidimensional datasets. AVIST highlights interactive data exploration by revealing fine data details. This is achieved through the use of animation and cross-filtering interactions. To support interactive exploration of big data, AVIST features a GPU (Graphics Processing Unit)-centric design. Two key aspects are emphasized on the GPU-centric design: (1) both data management and computation are implemented on the GPU to leverage its parallel computing capability and fast memory bandwidth; (2) a GPU-based directed acyclic graph is proposed to characterize data transformations triggered by users’ demands. Moreover, we implement AVIST based on the Model-View-Controller (MVC) architecture. In the implementation, we consider two aspects: (1) user interaction is highlighted to slice big data into small data; and (2) data transformation is based on parallel computing. Two case studies demonstrate how AVIST can help analysts identify abnormal behaviors and infer new hypotheses by exploring big datasets. Finally, we summarize lessons learned about GPU-based solutions in interactive information visualization with big data.

The Effect of Edge Bundling and Seriation on Sensemaking of Biclusters in Bipartite Graphs

Exploring coordinated relationships (e.g., shared relationships between two sets of entities) is an important analytics task in a variety of real-world applications, such as discovering similarly behaved genes in bioinformatics, detecting malware collusions in cyber security, and identifying products bundles in marketing analysis. Coordinated relationships can be formalized as biclusters. In order to support visual exploration of biclusters, bipartite graphs based visualizations have been proposed, and edge bundling is used to show biclusters. However, it suffers from edge crossings due to possible overlaps of biclusters, and lacks in-depth understanding of its impact on user exploring biclusters in bipartite graphs. To address these, we propose a novel bicluster-based seriation technique that can reduce edge crossings in bipartite graphs drawing and conducted a user experiment to study the effect of edge bundling and this proposed technique on visualizing biclusters in bipartite graphs. We found that they both had impact on reducing entity visits for users exploring biclusters, and edge bundles helped them find more justified answers. Moreover, we identified four key trade-offs that inform the design of future bicluster visualizations. The study results suggest that edge bundling is critical for exploring biclusters in bipartite graphs, which helps to reduce low-level perceptual problems and support high-level inferences.

Be the Data: Embodied Visual Analytics

With the rise of big data, it is becoming increasingly important to educate groups of students at many educational levels about data analytics. In particular, students without a strong mathematical background may have an unenthusiastic attitude towards high-dimensional data and find it challenging to understand relevant complex analytical methods, such as dimension reduction. In this paper, we present an embodied approach for visual analytics designed to teach students about exploring alternative 2D projections of high-dimensional data points using weighted multidimensional scaling. We propose a novel concept, Be the Data, to explore the possibilities of using humans embodied resources to learn from high-dimensional data. In our implemented system, each student embodies a data point, and the position of students in a physical space represents a 2D projection of the high-dimensional data. Students physically move within the room with respect to each other to collaboratively construct alternative projections and receive visual feedback about relevant data dimensions. In this way, students can pose hypotheses about the data to discover the statistical support as well as learn about complex concepts such as high-dimensional distance. We conducted educational workshops with students in various age groups inexperienced in complex data analytical methods. Our findings indicate that Be the Data provided the necessary engagement to enable students to quickly learn about high-dimensional data and analysis processes despite their minimal prior knowledge.

Interactive Discovery of Coordinated Relationship Chains with Maximum Entropy Models

Modern visual analytic tools promote human-in-the-loop analysis but are limited in their ability to direct the user toward interesting and promising directions of study. This problem is especially acute when the analysis task is exploratory in nature, e.g., the discovery of potentially coordinated relationships in massive text datasets. Such tasks are very common in domains like intelligence analysis and security forensics where the goal is to uncover surprising coalitions bridging multiple types of relations. We introduce new maximum entropy models to discover surprising chains of relationships leveraging count data about entity occurrences in documents. These models are embedded in a visual analytic system called MERCER (Maximum Entropy Relational Chain ExploRer) that treats relationship bundles as first class objects and directs the user toward promising lines of inquiry. We demonstrate how user input can judiciously direct analysis toward valid conclusions, whereas a purely algorithmic approach could be led astray. Experimental results on both synthetic and real datasets from the intelligence community are presented.

Be the Data: Social Meetings with Visual Analytics

Social meetings provide important venues for people to get connected. However, it is challenging to explore reasons of social gathering, identify its key impact factors, and further use it to support peoples social activities. In this paper, we present an embodied visual analytics system, which highlights analyzing and displaying social-cluster related information in real time. In the system, each user represents a data point in a high-dimensional dataset, and their positions reflect a 2D projection of the dataset, by using weighted multidimensional scaling. As users move and socialize with others, the 2D projection is dynamically updated, and relevant information of user clusters is visually analyzed and presented through dimension reduction techniques. We conducted informal social meetings with participants who were a mix of strangers and friends. We found that there are 3 stages of social gathering, corresponding to different interactions in the system. Our results also suggest that the system assists social gathering with dimension reduction visualizations.