Jieqiong Zhao

VASA: Interactive Computational Steering of Large Asynchronous Simulation Pipelines for Societal Infrastructure.

We present VASA, a visual analytics platform consisting of a desktop application, a component model, and a suite of distributed simulation components for modeling the impact of societal threats such as weather, food contamination, and traffic on critical infrastructure such as supply chains, road networks, and power grids. Each component encapsulates a high-fidelity simulation model that together form an asynchronous simulation pipeline: a system of systems of individual simulations with a common data and parameter exchange format. At the heart of VASA is the Workbench, a visual analytics application providing three distinct features: (1) low-fidelity approximations of the distributed simulation components using local simulation proxies to enable analysts to interactively configure a simulation run; (2) computational steering mechanisms to manage the execution of individual simulation components; and (3) spatiotemporal and interactive methods to explore the combined results of a simulation run. We showcase the utility of the platform using examples involving supply chains during a hurricane as well as food contamination in a fast food restaurant chain.

TimeFork: Interactive Prediction of Time Series

We present TimeFork, an interactive prediction technique to support users predicting the future of time-series data, such as in financial, scientific, or medical domains. TimeFork combines visual representations of multiple time series with prediction information generated by computational models. Using this method, analysts engage in a back-and-forth dialogue with the computational model by alternating between manually predicting future changes through interaction and letting the model automatically determine the most likely outcomes, to eventually come to a common prediction using the model. This computer-supported prediction approach allows for harnessing the users knowledge of factors influencing future behavior, as well as sophisticated computational models drawing on past performance. To validate the TimeFork technique, we conducted a user study in a stock market prediction game. We present evidence of improved performance for participants using TimeFork compared to fully manual or fully automatic predictions, and characterize qualitative usage patterns observed during the user study.

Evaluating Social Navigation Visualization in Online Geographic Maps

Social navigation enables emergent collaboration between independent collaborators by exposing the behavior of each individual. This is a powerful idea for web-based visualization, where the work of one user can inform other users interacting with the same visualization. Results from a crowdsourced user study evaluating the value of such social navigation cues for a geographic map service are presented. Results show significantly improved performance for participants who interacted with the map when the visual footprints of previous users were visible.

ParkAnalyzer: Characterizing the movement patterns of visitors VAST 2015 Mini-Challenge 1

The 2015 VAST challenge features movement tracking (Mini- Challenge 1 (MC1)) and communication information (Mini- Challenge 2 (MC2)) datasets of all visitors in an amusement park over a three-day weekend. The data includes around 25 million individual movement records, along with 4 million communication records. Analyzing and exploring such large-scale datasets require intelligent data mining methods that characterize the overall trends and anomalies, as well as interactive visual interfaces to support investigation at different spatiotemporal granularities. The objective of MC1 was to characterize the behavior of different groups of visitors, compare different activity patterns over the three days, and discover anomalies or unusual behavior patterns that relate to the crime that occurred during the weekend. We utilized both movement data provided in MC1 and communication data provided in MC2 to answer the questions asked in MC1.

TimeFork: Mixed-initiative time-series prediction

We present TimeFork, an analytics technique for predicting the behavior of multivariate time-series data originating from modern disciplines such as economics (stock market) and meteorology (climate), with human-in-the-loop. We identify two types of machine-generated predictions for such datasets: temporal prediction that predicts the future of an attribute; and spatial prediction that predicts an attribute based on the other attributes in the dataset. Visual exploration of this prediction space, constituting of these predictions of different confidences, by chunking and chaining predictions over time promises accurate user-guided predictions. In order to utilize TimeFork technique, we created a visual analytics application for user-guided prediction over different time periods, thus allowing for visual exploration of time-series data.

Big Data Visualizations in Organizational Science

Visualizations in organizational research have primarily been used in the context of traditional survey data, where individual data points (e.g., responses) can typically be plotted, and qualitative (e.g., language data) and quantitative (e.g., frequency data) information are not typically combined. Moreover, visualizations are typically used in a hypothetico-deductive fashion to showcase significant hypothesized results. With the advent of big data, which has been characterized as being particularly high in volume, variety, and velocity of collection, visualizations need to more explicitly and formally consider the issues of (a) identification (isolating or highlighting relevant data pertaining to the phenomena of interest), (b) integration (combining different modes of data to reveal insights about a phenomenon of interest), (c) immediacy (examining real-time data in a time-sensitive manner), and (d) interactivity (inductively uncovering and identifying new patterns). We discuss basic ideas for addressing these issues and provide illustrative examples of visualizations that incorporate and highlight ways of addressing these issues. Examples in our article include visualizing multiple performance criteria for police officers, publication network of organizational researchers, and social media language of Fortune 500 companies.

Visual analytics of heterogeneous data for criminal event analysis VAST challenge 2015: Grand challenge

We developed a visual analytics system to analyze the provided heterogeneous 2015 VAST Challenge data. This system utilized several analytic models and visualization techniques. Currently, the underlying data models and clustering techniques have limitations in processing the large volume of data in real time. Therefore, for future work, we will improve the scalability of our system to support real time interactivity and analysis.

Visual analytics for detecting communication patterns

Our visual analytics system that we developed as part of the 2015 VAST Challenge provided us with insights to characterize the communication patterns of visitors.

Real-time identification and monitoring of abnormal events based on microblog and emergency call data using SMART

This article describes a real-time visual analytics process based on microblog and emergency call data to solve VAST 2014 Mini Challenge 3. We extended SMART system (Social Media Analytics and Reporting Toolkit), developed by the U.S. Department of Homeland Securitys VACCINE Center. Our system consists of multiple linked views to allow the analyst monitor topic evolution, identify influential microblog users, observe geo-location patterns and examine correlations among different data sources. Extensions to our previous work include a time series view, a reply/retweet networks view, and integration of emergency call data.

AnnotatedTimeTree, Dodeca-Rings Map & SMART: A geo-temporal analysis of criminal events

The 2014 VAST Grand Challenge required us to find victims, suspects, and criminal motivations based on three separate dataseis. We developed three VA tools (AnnotatedTimeTree, Dodeca-Ring Map and SMART) to facilitate the understanding of heterogeneous multivariate dataseis. These tools were integrated to gain insights into the source data and find connections among complex information (Fig. 1). AnnotedTimeTree aims to identify the cause clues and timeline of kidnapping based on analysis of text and network data. Dodeca-Rings Map allows analysts to interact with geospatial, temporal, and card transaction data to find suspicious personal behaviors and social networks. SMART is a visual analytics tool that enables text stream analysis by dynamically visualizing microblog data on the map over time.