We introduce an analytical framework for analyzing tweets to (1) identify and categorize fine-grained details about a disaster such as affected individuals, damaged infrastructure and disrupted services; (2) distinguish impact areas and time periods, and relative prominence of each category of disaster-related information across space and time. We first identify disaster-related tweets by generating a human-labeled training dataset and experimenting a series of deep learning and machine learning methods for a binary classification of disaster-relatedness. We employ LSTM (Long Short-Term Memory) networks for the classification task because LSTM networks outperform other methods by considering the whole text structure using long-term semantic word and feature dependencies. Second, we employ an unsupervised multi-label classification of tweets using Latent Dirichlet Allocation (LDA), and identify latent categories of tweets such as affected individuals and disrupted services. Third, we employ spatially-adaptive kernel smoothing and density-based spatial clustering to identify the relative prominence and impact areas for each information category, respectively. Using Hurricane Irma as a case study, we analyze over 500 million keyword-based and geo-located collection of tweets before, during and after the disaster. Our results highlight potential areas with high density of affected individuals and infrastructure damage throughout the temporal progression of the disaster.

We present the results of a user study comparing variants of commonly used line symbolizations for directed origin–destination flow maps. Our design and evaluation consisted of five line symbolizations that employ a combination of following visual variables: arrowheads, origin–destination coloring (color hue, and value), line shortening, line width, tapered edges (varying width from wide to narrow, and narrow to wide), and curvature asymmetry and strength. To guide our evaluation, we used a task-by-type typology and chose four representative tasks that are commonly used in flow map reading: identifying dominant direction of flows, flows with the highest magnitude (volume), spatial focusing of long flows toward a destination, and clusters of high net-exports (net-outflow). We systematically analyzed user responses and task performance which we measured by task completion time and accuracy. We designed a web-based flow mapping and testing framework and recruited the participants from Amazon Mechanical Turk. To demonstrate the application and user experiment, we used 16 commodity flow data sets in the United States from 2007 and systematically rotated the layouts to evaluate the effect of layout orientation. From this study, we can conclude that there is potential usefulness for all of the five symbolizations we tested; however, the influence of the design on performance and perception depends on the type of the task. Also, we found that data and layout orientation have significant effects on performance and perception of patterns in flow maps which we attribute to the change in visual saliency of node and flow patterns in relation to the way users scan the map. We recommend that the choice of line symbolization should be guided by a task taxonomy which end users are expected to perform. We discuss various design trade-offs and recommendations and potential future work for designing and evaluating line symbolizations for flow mapping.

Thanks to recent advances in high-performance computing and deep learning, computer vision algorithms coupled with spatial analysis methods provide a unique opportunity for extracting human activity patterns from geo-tagged social media images. However, there are only a handful of studies that evaluate the utility of computer vision algorithms for studying large-scale human activity patterns. In this article, we introduce an analytical framework that integrates a computer vision algorithm based on convolutional neural networks (CNN) with kernel density estimation to identify objects, and infer human activity patterns from geo-tagged photographs. To demonstrate our framework, we identify bird images to infer birdwatching activity from approximately 20 million publicly shared images on Flickr, across a three-year period from December 2013 to December 2016. In order to assess the accuracy of object detection, we compared results from the computer vision algorithm to concept-based image retrieval, which is based on keyword search on image metadata such as textual description, tags, and titles of images. We then compared patterns in birding activity generated using Flickr bird photographs with patterns identified using eBird data—an online citizen science bird observation application. The results of our eBird comparison highlight the potential differences and biases in casual and serious birdwatching, and similarities and differences among behaviors of social media and citizen science users. Our analysis results provide valuable insights into assessing the credibility and utility of geo-tagged photographs in studying human activity patterns through object detection and spatial analysis.

Patient mobility can be defined as a patient's movement or utilization of a health care service located in a place or region other than the patient's place of residence. Mobility provides freedom to patients to obtain health care from providers across regions and even countries. It is essential to monitor patient choices in order to maintain the quality standards and responsiveness of the health system, otherwise, the health system may suffer from geographic disparities in the accessibility to quality and responsive health care. In this article, we study patient mobility in a national health care system to identify medical regions, spatio-temporal and service characteristics of health care utilization, and demands for patient mobility. We conducted a systematic analysis of province-to-province patient mobility in Turkey from December 2009 to December 2013, which was derived from 1.2 billion health service records. We first used a flow-based regionalization method to discover functional medical regions from the patient mobility network. We compare the results of data-driven regions to designated regions of the government in order to identify the areas of mismatch between planned regional service delivery and the observed utilization in the form of patient flows. Second, we used feature selection, and multivariate flow clustering to identify spatio-temporal characteristics and health care needs of patients on the move. Medical regions we derived by analyzing the patient mobility data showed strong overlap with the designated regions of the Ministry of Health. We also identified a number of regions that the regional service utilization did not match the planned service delivery. Overall, our spatio-temporal and multivariate analysis of regional and long-distance patient flows revealed strong relationship with socio-demographic and cultural structure of the society and migration patterns. Also, patient flows exhibited seasonal patterns, and yearly trends which correlate with implemented policies throughout the period. We found that policies resulted in different outcomes across the country. We also identified characteristics of long-distance flows which could help inform policy-making by assessing the needs of patients in terms of medical specialization, service level and type. Our approach helped identify (1) the mismatch between regional policy and practice in health care utilization (2) spatial, temporal, health service level characteristics and medical specialties that patients seek out by traveling longer distances. Our findings can help identify the imbalance between supply and demand, changes in mobility behaviors, and inform policy-making with insights.

Interpersonal communication on online social networks has a significant impact on the society by not only diffusing information, but also forming social ties, norms, and behaviors. Knowing how the conversational discourse semantically and geographically vary over time can help uncover the changing dynamics of interpersonal ties and the digital traces of social events. This article introduces a framework for modeling and visualizing the semantic and spatio-temporal evolution of topics in a spatially embedded and time-stamped interpersonal communication network. The framework consists of (1) a topic modeling workflow for modeling topics and extracting the evolution of conversational discourse; (2) a geo-social network modeling and smoothing approach to projecting connection characteristics and semantics of communication onto geographic space and time; (3) a web-based geovisual analytics environment for exploring semantic and spatio-temporal evolution of topics in a spatially embedded and time-stamped interpersonal communication network. To demonstrate, geo-located and reciprocal user mention and reply tweets over the course of the 2016 primary and presidential elections in the United States from 1 August 2015 to 15 November 2016 were analyzed. The large portion of the topics extracted from mention tweets were related to daily life routines, human activities, and interests such as school, work, sports, dating, wearing, birthday celebration, music, food, and live-tweeting. Specific focus on the analysis of political conversations revealed that the content of conversational discourse was split between civil rights and election-related discussions of the political campaigns and candidates. These political topics exhibited major shifts in terms of content and the popularity in reaction to primaries, debates, and events throughout the study period. While civil rights discussions were more dominant and in higher intensity across the nation and throughout the whole time period, election-specific conversations resulted in temporally varying local hotspots that correlated with locations of primaries and events.