Sporting events attract high volumes of people, which in turn leads to increased use of social media. In addition, research shows that sporting events may trigger violent behavior that can lead to crime. This study analyses the spatial relationships between crime occurrences, demographic, socio-economic and environmental variables, together with geo-located Twitter messages and their 'violent' subsets. The analysis compares basketball and hockey game days and non-game days. Moreover, this research aims to analyze crime prediction models using historical crime data as a basis and then introducing tweets and additional variables in their role as covariates of crime. First, this study investigates the spatial distribution of and correlation between crime and tweets during the same temporal periods. Feature selection models are applied in order to identify the best explanatory variables. Then, we apply localized kernel density estimation model for crime prediction during basketball and hockey games, and on non-game days. Findings from this study show that Twitter data, and a subset of violent tweets, are useful in building prediction models for the seven investigated crime types for home and away sporting events, and non-game days, with different levels of improvement.

Continuous passive sensing using smartphone embedded sensors can drain the battery quickly, interrupting other usages of the device. In order to improve the energy efficiency in continuous mobile sensing applications, we propose a new adaptive sensing framework using reinforcement learning to optimize the sensing timing. We model our adaptive sensing problem as a Markov Decision Process and dynamically change the sensing timing of targeted sensor(s) so that they are only operating in desired contexts (e.g. collect accelerometer data only when the phone is moving). Using accelerometer data continuously collected from 220 participants for over two weeks, we show that our approach is able to save the energy while attaining high accuracy and data coverage. Specifically, our strategy attains energy saving of 62.4 % at an accuracy of 80.9% and data coverage of 67.4%, which outperforms two baseline strategies, including a random strategy and a strategy using learning automata technique.

Activists have used social media during modern civil uprisings, and researchers have found that the generated content is predictive of off-line protest activity. However, questions remain regarding the drivers of this predictive power. In this paper, we begin by deriving predictor variables for individuals' protest decisions from the literature on protest participation theory. We then test these variables on the case of Twitter and the 2011 Egyptian revolution. We find significant positive correlations between the volume of future-protest descriptions on Twitter and protest onsets. We do not find significant correlations between such onsets and the preceding volume of political expressions, individuals' access to news, and connections with political activists. These results locate the predictive power of social media in its function as a protest advertisement and organization mechanism. We then build predictive models using future-protest descriptions and compare these models with baselines informed by daily event counts from the Global Database of Events, Location, and Tone (GDELT). Inspection of the significant variables in our GDELT models reveals that an increased military presence may be predictive of protest onsets in major cities. In sum, this paper highlights the ways in which online activism shapes off-line behavior during civil uprisings.

Increasingly large document collections require improved information processing methods for searching, retrieving, and organizing text. Central to these information processing methods is document classification, which has become an important application for supervised learning. Recently the performance of traditional supervised classifiers has degraded as the number of documents has increased. This is because along with growth in the number of documents has come an increase in the number of categories. This paper approaches this problem differently from current document classification methods that view the problem as multi-class classification. Instead we perform hierarchical classification using an approach we call Hierarchical Deep Learning for Text classification (HDLTex). HDLTex employs stacks of deep learning architectures to provide specialized understanding at each level of the document hierarchy.

The burden of entry into mobile crowdsensing (MCS) is prohibitively high for human-subject researchers who lack a technical orientation. As a result, the benefits of MCS remain beyond the reach of research communities (e.g., psychologists) whose expertise in the study of human behavior might advance applications and understanding of MCS systems. This paper presents Sensus, a new MCS system for human-subject studies that bridges the gap between human-subject researchers and MCS methods. Sensus alleviates technical burdens with on-device, GUI-based design of sensing plans, simple and efficient distribution of sensing plans to study participants, and uniform participant experience across iOS and Android devices. Sensing plans support many hardware and software sensors, automatic deployment of sensor-triggered surveys, and double-blind assignment of participants within randomized controlled trials. Sensus offers these features to study designers without requiring knowledge of markup and programming languages. We demonstrate the feasibility of using Sensus within two human-subject studies, one in psychology and one in engineering. Feedback from non-technical users indicates that Sensus is an effective and low-burden system for MCS-based data collection and analysis.

Kernel density estimation is a popular method for identifying crime hotspots for the purpose of data-driven policing. However, computing a kernel density estimate is computationally intensive for large crime datasets, and the quality of the resulting estimate depends heavily on parameters that are difficult to set manually. Inspired by methods from image processing, we propose a novel way for performing hotspot analysis using localized kernel density estimation optimized with an evolutionary algorithm. The proposed method uses local learning to address three challenges associated with traditional kernel density estimation: computational complexity, bandwidth selection, and kernel function selection. We evaluate our localized kernel model on 17 crime types from Chicago, Illinois, USA. Preliminary results indicate significant improvement in prediction performance over the traditional approach. We also examine the effect of data sparseness on the performance of both models.

The convergence of public data and statistical modeling has created opportunities for public safety officials to prioritize the deployment of scarce resources on the basis of predicted crime patterns. Current crime prediction methods are trained using observed crime and information describing various criminogenic factors. Researchers have favored global models (e.g., of entire cities) due to a lack of observations at finer resolutions (e.g., ZIP codes). These global models and their assumptions are at odds with evidence that the relationship between crime and criminogenic factors is not homogeneous across space. In response to this gap, we present area-specific crime prediction models based on hierarchical and multi-task statistical learning. Our models mitigate sparseness by sharing information across ZIP codes, yet they retain the advantages of localized models in addressing non-homogeneous crime patterns. Out-of-sample testing on real crime data indicates predictive advantages over multiple state-of-the-art global models.

Prior work in statistical crime prediction has not investigated micro-level movement patterns of individuals in the area of interest. Geotagged social media implicitly describe these patterns for many individuals; however, methods of extracting such patterns and integrating them into a statistical model remain undeveloped. This paper presents methods and experiments that begin to fill this gap. We investigate the use of spatiotemporally tagged Twitter posts for inferring micro-level movement patterns, and we use real crime data to develop and test a model informed by such patterns. Our results indicate improved performance for 15 of the 20 crime types studied, when comparing our model with a baseline that does not use micro-level movement patterns.

The Internet's increasing use as a means of communication has led to the formation of cyber communities, which have become appealing to violent extremist (VE) groups. This paper presents research on forecasting the daily level of cyber-recruitment activity of VE groups. We used a previously developed support vector machine model to identify recruitment posts within a Western jihadist discussion forum. We analyzed the textual content of this data set with latent Dirichlet allocation (LDA), and we fed these analyses into a variety of time series models to forecast cyber-recruitment activity within the forum. Quantitative evaluations showed that employing LDA-based topics as predictors within time series models reduces forecast error compared with naive (random-walk), autoregressive integrated moving average, and exponential smoothing baselines. To the best of our knowledge, this is the first result reported on this forecasting task. This research could ultimately help assist with efficient allocation of intelligence analysts in response to predicted levels of cyber-recruitment activity.

Care variation studies often use large amounts of data but approaches developed for such research are either scalable but not locally applicable or locally applicable but not scalable. We present a method that is scalable and locally applicable while being statistically significant. Using a population of patients diagnosed with both congestive heart failure and myocardial infarction, we developed and tested measures of care variation on data derived from hospital billing records. Our metrics yielded statistically significant results. Computing time for the method was found to increase linearly allowing for the desired scalability. In the future, our care variation metrics be used to gain insight into local conditions that correlate with outcomes of interest like visit charges or morbidity rates.

The processes of developing, monitoring, and maintaining transportation systems produce large volumes of information. Human fieldworkers are often responsible for gathering this information, and despite their best efforts, they will inevitably introduce errors into the collected data. This is a critical problem since: 1) the collected data are used to justify key infrastructure maintenance and development decisions; and 2) the volume of unstructured information (e.g., plain text) makes manual quality control prohibitively expensive. We introduce a solution to this problem in the example domain of vehicle accident reports. First, we analyzed a sample of accident reports and confirmed the existence of many data entry errors. Second, we developed and evaluated a statistical language processing approach that automatically identifies reports containing data entry errors. We tested a variety of system configurations on real-world data and compared their performance with multiple baseline methods. The best configuration achieved a performance score of 84%, far outperforming the baseline methods. Our results and analyses have quality control implications for any data source that pairs structured text (e.g., coded fields) with unstructured text.

Personal and property crimes create large economic losses within the United States. To prevent crimes, law enforcement agencies model the spatio-temporal pattern of criminal incidents. In this paper, we present a new modeling process that combines two of our recently developed approaches for modeling criminal incidents. The first component of the process is the spatio-temporal generalized additive model (STGAM), which predicts the probability of criminal activity at a given location and time using a feature-based approach. The second component involves textual analysis. In our experiments, we automatically analyzed Twitter posts, which provide a rich, event-based context for criminal incidents. In addition, we describe a new feature selection method to identify important features. We applied our new model to actual criminal incidents in Charlottesville, Virginia. Our results indicate that the STGAM/Twitter model outperforms our previous STGAM model, which did not use Twitter information. The STGAM/Twitter model can be generalized to other applications of event modeling where unstructured text is available.