Dominik Jäckle

State-of-the-Art Report of Visual Analysis for Event Detection in Text Data Streams

Event detection from text data streams has been a popular research area in the past decade. Recently, the evolution of microblogging and social network services opens up great opportunities for various kinds of knowledge-based intelligence activities which require tracking of real-time events. In a sense, visualizations in combination with analytical processes could be a viable method for such tasks because it can be used to analyze the sheer amounts of text streams. However, data analysts and visualization experts often face grand challenges stemming out of the ill-defined concept of event and various kinds of textual data. As a result, we have few guidelines on how to build successful visual analysis tools that can handle specific event types and diverse textual data sources. Our goal is to take the first step towards answering the question by organizing insights from prior research studies on event detection and visual analysis. In the scope of this report, we summarize the evolution of event detection in combination with visual analysis over the past 14 years and provide an overview of the state-of-the-art methods. Our investigation sheds light on various kinds of research areas that can be the most beneficial to the field of visual text event analytics.

Temporal MDS Plots for Analysis of Multivariate Data

Multivariate time series data can be found in many application domains. Examples include data from computer networks, healthcare, social networks, or financial markets. Often, patterns in such data evolve over time among multiple dimensions and are hard to detect. Dimensionality reduction methods such as PCA and MDS allow analysis and visualization of multivariate data, but per se do not provide means to explore multivariate patterns over time. We propose Temporal Multidimensional Scaling (TMDS), a novel visualization technique that computes temporal one-dimensional MDS plots for multivariate data which evolve over time. Using a sliding window approach, MDS is computed for each data window separately, and the results are plotted sequentially along the time axis, taking care of plot alignment. Our TMDS plots enable visual identification of patterns based on multidimensional similarity of the data evolving over time. We demonstrate the usefulness of our approach in the field of network security and show in two case studies how users can iteratively explore the data to identify previously unknown, temporally evolving patterns.

Visual Soccer Analytics: Understanding the Characteristics of Collective Team Movement Based on Feature-Driven Analysis and Abstraction

With recent advances in sensor technologies, large amounts of movement data have become available in many application areas. A novel, promising application is the data-driven analysis of team sport. Specifically, soccer matches comprise rich, multivariate movement data at high temporal and geospatial resolution. Capturing and analyzing complex movement patterns and interdependencies between the players with respect to various characteristics is challenging. So far, soccer experts manually post-analyze game situations and depict certain patterns with respect to their experience. We propose a visual analysis system for interactive identification of soccer patterns and situations being of interest to the analyst. Our approach builds on a preliminary system, which is enhanced by semantic features defined together with a soccer domain expert. The system includes a range of useful visualizations to show the ranking of features over time and plots the change of game play situations, both helping the analyst to interpret complex game situations. A novel workflow includes improving the analysis process by a learning stage, taking into account user feedback. We evaluate our approach by analyzing real-world soccer matches, illustrate several use cases and collect additional expert feedback. The resulting findings are discussed with subject matter experts.

ColorCAT: Guided Design of Colormaps for Combined Analysis Tasks

Colormap design is challenging because the encoding must match the requirements of data and analysis tasks as well as the perception of the target user. A number of well-known tools exist to support the design of colormaps. ColorBrewer [HB03], for example, is a great resource to select colors for qualitative, sequential, and diverging data. PRAVDAColor [BRT95] and Tominski et al. [TFS08], for example, provide valuable guidelines for single analysis tasks such as localization, identification, and comparison. However, for solving real world problems in most practical applications, single elementary analysis tasks are not sufficient but need to be combined. In this paper, we propose a methodology and tool to design colormaps for combined analysis tasks. We define color mapping requirements and develop a set of design guidelines. The visualization expert is integrated in the design process to incorporate his/her design requirements, which may depend on the application, culture, and aesthetics. Our ColorCAT tool guides novice and expert designers through the creation of colormaps and allows the exploration of the design space of color mapping for combined analysis tasks.

Visual abstraction of complex motion patterns

Today’s tracking devices allow high spatial and temporal resolutions and due to their decreasing size also an ever increasing number of application scenarios. However, understanding motion over time is quite difficult as soon as the resulting trajectories are getting complex. Simply plotting the data may obscure important patterns since trajectories over long time periods often include many revisits of the same place which creates a high degree of over-plotting. Furthermore, important details are often hidden due to a combination of large-scale transitions with local and small-scale movement patterns. We present a visualization and abstraction technique for such complex motion data. By analyzing the motion patterns and displaying them with visual abstraction techniques a synergy of aggregation and simplification is reached. The capabilities of the method are shown in real-world applications for tracked animals and discussed with experts from biology. Our proposed abstraction techniques reduce visual clutter and help analysts to understand the movement patterns that are hidden in raw spatiotemporal data.

Integrated Spatial Uncertainty Visualization using Off-screen Aggregation

Visualization of spatial data uncertainties is crucial to the data understanding and exploration process. Scientific measurements, numerical simulations, and user generated content are error prone sources that gravely influence data reliability. When exploring large spatial datasets, we face two main challenges: data and uncertainty are two different sets which need to be integrated into one visualization, and we often lose the contextual overview when zooming or filtering to see details. In this paper, we present an extrinsic uncertainty visualization as well as an off-screen technique which integrates the uncertainty representation and enables the user to perceive data context and topology in the analysis process. We show the applicability and usefulness of our approach in a use case.

Interpretation of Dimensionally-reduced Crime Data : A Study with Untrained Domain Experts

Dimensionality reduction (DR) techniques aim to reduce the amount of considered dimensions, yet preserving as much information as possible. According to many visualization researchers, DR results lack interpretability, in particular for domain experts not familiar with machine learning or advanced statistics. Thus, interactive visual methods have been extensively researched for their ability to improve transparency and ease the interpretation of results. However, these methods have primarily been evaluated using case studies and interviews with experts trained in DR. In this paper, we describe a phenomenological analysis investigating if researchers with no or only limited training in machine learning or advanced statistics can interpret the depiction of a data projection and what their incentives are during interaction. We, therefore, developed an interactive system for DR, which unifies mixed data types as they appear in real-world data. Based on this system, we provided data analysts of a Law Enforcement Agency (LEA) with dimensionally-reduced crime data and let them explore and analyze domain-relevant tasks without providing further conceptual information. Results of our study reveal that these untrained experts encounter few difficulties in interpreting the results and drawing conclusions given a domain relevant use case and their experience. We further discuss the results based on collected informal feedback and observations.

Leaf Glyph - Visualizing Multi-dimensional Data with Environmental Cues

In exploratory data analysis, important analysis tasks include the assessment of similarity of data points, labeling of outliers, identifying and relating groups in data, and more generally, the detection of patterns. Specifically, for large data sets, such tasks may be effectively addressed by glyph-based visualizations. Appropriately defined glyph designs and layouts may represent collections of data to address these aforementioned tasks. Important problems in glyph visualization include the design of compact glyph representations, and a similarityor structure-preserving 2D layout. Projection-based techniques are commonly used to generate layouts, but often suffer from over-plotting in 2D display space, which may hinder comparing and relating tasks. We introduce a novel glyph design for visualizing multi-dimensional data based on an environmental metaphor. Motivated by the humans ability to visually discriminate natural shapes like trees in a forest, single flowers in a flower-bed, or leaves at shrubs, we design a leaf-shaped data glyph, where data controls main leaf properties including leaf morphology, leaf venation, and leaf boundary shape. We also define a custom visual aggregation scheme to scale the glyph for large numbers of data records. We show by example that our design is effectively interpretable to solve multivariate data analysis tasks, and provides effective data mapping. The design also provides an aesthetically pleasing appearance, which may help spark interest in data visualization by larger audiences, making it applicable e.g., in mass media.

Ambient Grids : Maintain Context-Awareness via Aggregated Off-Screen Visualization

When exploring large spatial datasets, zooming and panning interactions often lead to the loss of contextual overview. Existing overview-plus-detail approaches allow users to view context while inspecting details, but they often suffer from distortion or overplotting. In this paper, we present an off-screen visualization method called Ambient Grids that strikes the balance between overview and details by preserving the contextual information as color grids within a designated space around the focal area. In addition, we describe methods to generate Ambient Grids for point data using data aggregation and projection. In a use case, we show the usefulness of our technique in exploring the VAST Challenge 2011 microblog dataset.

Star Glyph Insets for Overview Preservation of Multivariate Data

Exploring vast spatial datasets often requires to drill down in order to inspect details, thus leading to a loss of contextual overview. An additional challenge rises if the visualized data is of multivariate nature, which we encounter in various domains such as healthcare, nutrition, crime reports, or social networks. Existing overview-plus-detail approaches do provide context but only limited support for multivariate data and often suffer from distortion. In this paper, we dynamically integrate star glyphs as insets into the spatial representation of multivariate data thus providing overview while inspecting details. Star glyphs pose an efficient and space saving method to visualize multivariate data, which qualifies them as integrated data representative. Furthermore, we demonstrate the usefulness of our approach in two use cases: The spatial exploration of multivariate crime data collected in San Francisco and the exploration of multivariate whisky data. Introduction Multivariate data accompanies us in our day-to-day life. Prominent examples represent data from healthcare, nutrition, crime reports, or social networks, among others. We typically use spatial representations in order to determine patterns and correlations among dimensions. An example represents the exploration of a huge set of malt whiskies: Each whisky is assigned to the geo-location of its distillery and has several diverse taste categories. The task can be either to seek correlations between particular taste categories and geo-locations, or to find patterns of whiskies for certain taste categories. The latter case can be achieved by applying dimension reduction methods which project the data to a lower dimensional space. When exploring such vast amounts of spatial data, at some point we use zooming and panning interactions to focus on certain regions of interest to obtain a detailed view. However, due to the limited size of the display screen, zooming and panning interactions lead to an inevitable loss of the contextual overview. Overview can be regained by zooming out resulting in a continuous trade-off between overview and detail. Jerding and Stasko argue that the limited size of the display makes it difficult to create efficient global views [25]. Existing Overview-and-Detail and Focus-plus-Context approaches provide comprehensive methods that typically operate in image space. Overview-and-Detail techniques attach a second viewport to the visualization. Although overview is provided, the user is forced to split his attention, which can result in increased cognitive load [19]. In contrast, Focus-plus-Context techniques integrate overview and detail, but use image-based distortion which restricts the interface by means of zooming levels [36]. In this paper, we propose a novel data-driven Off-Screen visualization technique for spatial multivariate data. More specifically, we contribute a dynamic integration of star glyphs as efficient visual insets for the representation of multivariate off-screen data objects. To do so, we augment the viewport with a dedicated border region including star glyph insets. A result of our approach is depicted in Figure 1. The remainder of this paper is organized as follows: First, we discuss related work. Then, we introduce the design of our approach and show the usefulness in two use cases, before we conclude and outline future work. Related Work In order to preserve the overview of multivariate data during exploration, we need to consider the potentials of both multivariate data visualization and overview preserving visualization. Following, we discuss related work of these areas. Multivariate Data Visualization Visual analysis of multivariate data has the objective of allowing the user to identify correlations and patterns among dimensions. Dimensions in multivariate data are not supposed to be considered independently but simultaneously, because they typically provide combined information that contributes to the overall understanding of the data [33]. Various techniques have been presented to visualize multivariate data. Prominent examples of geometric projections are parallel coordinates [22], Andrew curves[1], or star coordinates [27]. Pixel-oriented techniques include recursive patterns [28] and pixel barcharts [29]. However, aforementioned techniques are not optimal to be integrated as space efficient inset giving a coarse overview of dimensions; glyph-based techniques such as Chernoff faces [7] or star glyphs [5] meet these requirements. Integration of Overview and Detail In order to allow efficient navigation and provide support for data analysis, the integrated preservation of the contextual overview is crucial. In this paper, the term context refers to the overview of the multivariate nature of the data including information about location and in some cases topology of the data. Following, we give a brief overview of integrated techniques, namely Focus-plus-Context and Off-screen Visualization techniques. Distortion-oriented Techniques The pioneering approach of Apperley et al. [2] provides a maximum focus region while all surrounding areas are distorted. Variations of this approach apply the technique for example to one-dimensional visualizations [32, 39]. Furnas [14] further introduced the degree-of-interest (DOI) function as basis for the wellknown Focus-plus-Context systems [40, 4]. Additional Focus-