Tuesday, December 10, 2019

Bots Fired: Examining Social Bot Evidence in Online Mass Shooting Conversations

Building upon our work with respect to how bots impact online conversations pertaining to global events and  health,  we have extended this research to see what role bots play in mass shooting events. In our new paper published in Palgrave Communications entitled "Bots Fired: Examining Social Bot Evidence in Online Mass Shooting Conversations" we examine four mass shooting events (i.e., Las Vegas, Sutherland Springs, Parkland, and Santa Fe) and find that social bots participate and contribute to online mass shooting conversations in a manner that is distinguishable from human contributions. Below we provide the abstract, along with some figures from the paper that highlight our methodology and main results. Finally at the bottom of the post, we provide the full reference to the paper. 

Mass shootings, like other extreme events, have long garnered public curiosity and, in turn, significant media coverage. The media framing, or topic focus, of mass shooting events typically evolves over time from details of the actual shooting to discussions of potential policy changes (e.g., gun control, mental health). Such media coverage has been historically provided through traditional media sources such as print, television, and radio, but the advent of online social networks (OSNs) has introduced a new platform for accessing, producing, and distributing information about such extreme events. The ease and convenience of OSN usage for information within society’s larger growing reliance upon digital technologies introduces potential unforeseen risks. Social bots, or automated software agents, are one such risk, as they can serve to amplify or distort potential narratives associated with extreme events such as mass shootings. In this paper, we seek to determine the prevalence and relative importance of social bots participating in OSN conversations following mass shooting events using an ensemble of quantitative techniques. Specifically, we examine a corpus of more than 46 million tweets produced by 11.7 million unique Twitter accounts within OSN conversations discussing four major mass shooting events: the 2017 Las Vegas concert shooting, the 2017 Sutherland Springs church shooting, the 2018 Parkland school shooting and the 2018 Santa Fe school shooting. This study’s results show that social bots participate in and contribute to online mass shooting conversations in a manner that is distinguishable from human contributions. Furthermore, while social bots accounted for fewer than 1% of total corpus user contributors, social network analysis centrality measures identified many bots with significant prominence in the conversation networks, densely occupying many of the highest eigenvector and out-degree centrality measure rankings, to include 82% of the top-100 eigenvector values of the Las Vegas retweet network.

Keywords: Social bots, mass shootings, school shootings, online social networks, computational social science.

 Overview of social bot analysis framework illustrating methodological steps taken to analyze social bots within online social network conversations involving mass shooting events

Overall tweet corpus volumes and suspected social bot contributions for each associated OSN mass shooting 215 event conversation.

Intra-group and cross-group retweet interaction rates among and between human (blue) and suspected social bot (red) user accounts for a one-month period following the (a) Las Vegas, (b) Sutherland Springs, (c) Parkland and (d) Santa Fe shooting events.

Social bot accounts in the top-N, where N = 1000/100/10, (a) eigenvector, (b) in-degree, (c) out-degree and (d) PageRank centrality measurement rankings within OSN mass shooting retweet networks discussing the Las Vegas (red), Sutherland Springs (green), Parkland (blue) and Santa Fe (purple) shooting events.

Full Reference:
Schuchard, R., Crooks, A.T., Croitoru, A. and Stefanidis, A. (2019) Bots Fired: Examining Social Bot Evidence in Online Mass Shooting Conversations, Palgrave Communications, 5: 158. Available at https://doi.org/10.1057/s41599-019-0359-x. (pdf)

Monday, December 09, 2019

Modeling Homeowners Post-flood Reconstruction Decisions

In the past we have developed agent-based models to explore a wide variety of applications and even to explored at humanitarian assistance after a natural disaster, however we have not explored how people might decide to rebuild or not after a natural disaster. Well that was until now. In a new paper with Kim McEligot, Peggy Brouse and myself entitled "Sea Bright, New Jersey Reconstructed: Agent-Based Protection Theory Model Responses to Hurricane Sandy" which was published as part of the 2019 Winter Simulation Conference. In the paper we use a hindcast (aka backtesting) of Hurricane Sandy’s damage to Sea Bright, NJ and explore homeowners post-flood reconstruction decisions. Below we provide the abstract to the paper, a short movie of the model running, along with a link to access the source code and data of the model, and finally a link to the full paper.

Coastal flooding is the most expensive type of natural disaster in the United States. Policy initiatives to mitigate the effects of these events are dependent upon understanding flood victim responses at an individual and municipal level. Agent-Based Modeling (ABM) is an effective tool for analyzing community-wide responses to natural disaster, but the quality of the ABM’s performance is often challenging to determine. This paper discusses the complexity of the Protective Action Decision Model (PADM) and Protection Motivation Theory (PMT) for human decision making regarding hazard mitigations. A combined (PADM/PMT) model is developed and integrated into the MASON modeling framework. The ABM implements a hind-cast of Hurricane Sandy’s damage to Sea Bright, NJ and homeowner post-flood reconstruction decisions. It is validated against damage assessments and post-storm surveys. The contribution of socio-economic factors and built environment on model performance is also addressed and suggests that mitigation for townhouse communities will be challenging.
The model source code (utilizing MASON Version 17) and data is available on CoMSES.net: http://bit.ly/SEABrightABM.

Our adaptation of the Protection Motivation Theory and Protective Action Decision Model.

Full Reference:
McEligot, K. Brouse, P. and Crooks A.T. (2019), Sea Bright, New Jersey Reconstructed: Agent-Based Protection Theory Model Responses to Hurricane Sandy, in Mustafee, N., Bae, K.-H.G., Lazarova-Molnar, S., Rabe, M., Szabo, C., Haas, P. and Son, Y-J. (eds.), Proceedings of the 2019 Winter Simulation Conference, National Harbor, MD, pp 251-262 (pdf)

Wednesday, November 06, 2019

Simulating Urban Patterns of Life: A Geo-Social Data Generation Framework

At the ACM SIGSPATIAL'19 conference, Joon-Seok Kim, Hamdi Kavak, Umar Manzoor, Dieter Pfoser, Carola Wenk, Andreas Züfle and myself have a paper entitled "Simulating Urban Patterns of Life: A Geo-Social Data Generation Framework." The general idea behind the paper is that while trajectory data is being used to capture human mobility in many applications (e.g. traffic prediction, ride-sharing applications), the use of real-world trajectory data raises serious concerns with respect to the privacy of users who contribute such information. 

To overcome privacy concerns we have created a geo-social data generator by utilizing agent-based modeling. The notion behind this generator is to allow users to develop and customize the logic of agent behaviors for different applications domains (e.g. commuting around a city). Once the basic model is created, the simulation can then be run and  geo-social data is generated which can then be used as a substitute to real-world trajectory data to study human mobility. If you wish to find out more about this paper, below is the abstract to the paper, along with some figures of the framework architecture and a link to the paper. Further supplementary materials including a demo video (which is also below) and sample data can be found at: http://sigspatial19demo.joonseok.org.

Data generators have been heavily used in creating massive trajectory datasets to address common challenges of real-world datasets, including privacy, cost of data collection, and data quality. However, such generators often overlook social and physiological characteristics of individuals and as such their results are often limited to simple movement patterns. To address these shortcomings, we propose an agent-based simulation framework that facilitates the development of behavioral models in which agents correspond to individuals that act based on personal preferences, goals, and needs within a realistic geographical environment. Researchers can use a drag-and-drop interface to design and control their own world including the geospatial and social (i.e. geo-social) properties. The framework is capable of generating and streaming very large data that captures the basic patterns of life in urban areas. Streaming data from the simulation can be accessed in real time through a dedicated API. 
Keywords: Agent-based simulation, trajectory data, data generator, spatial network, human behavior.
Causality in human behavior

Architecture of framework

Layout of model builder and sample model

Full Reference:
Kim, J-S., Kavak, H., Manzoor, U., Crooks, A.T., Pfoser, D., Wenk C. and Z├╝fle, A (2019), Simulating Urban Patterns of Life: A Geo-Social Data Generation Framework, in Banaei-Kashani, F., Trajcevski, G., Güting, R.H., Kulik, L. and Newsam, S. (eds.), Proceedings of the 27th International Conference on Advances in Geographic Information Systems (ACM SIGSPATIAL 2019), Chicago, IL. (pdf)

Tuesday, November 05, 2019

New Paper: Assessing the Placeness of Locations through User-contributed Content

In the past we have written about how one can use crowdsourced data to gain a collective sense of place from Twitter contributions and also from corresponding Wikipedia entries (e.g. here). In a new paper with Xiaoyi Yuan, we extend this work to explore how user-contributed data can be used to explore if urban places are becoming inauthentic due to urban commodification and standardization by chain stores such as restaurants. To this end, at the at 3rd ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery (GeoAI) we have a paper entitled: "Assessing the Placeness of Locations through User-contributed Content"

In the paper we attempt to understand the relationship between restaurants and urban identities via user-contributed content. We extracted and analyzed information from over 3 million Yelp reviews from 37,000 restaurants using a Convolutional Neural Network (CNN) model in order to study places from the bottom up. Specifically we were interested to what extent cities share similarities or differences in their Yelp restaurant reviews. Furthermore, we wanted to explore how opinion aspects (i.e. what reviewers care about the most) are mentioned differently in urban chain and independent restaurants. Through the analysis of the Yelp reviews we find that online geo-tagged text data is fruitful for understanding places and aspect-based sentiment analysis helps us understand the large volumes of text. Not only did we discover that cities show homogeneity in terms of restaurant reviews, but for chain restaurants, “location” often emphasizes the differences between different stores of the same chain whereas for independent restaurant reviews, the aspect “location” reflects the characteristics of the places the restaurants are situated. If this is of interest to you, below we provide the abstract to the paper, along with some of the key findings and a link to the paper.

Previous research has argued that urban places are becoming “placeless” and inauthentic. Many local policies have also proposed to encourage more independent stores in order to restore urban identity. Others argue, however, that chain stores provide affordable merchandise and different locations of the same chain may have different meanings to an individual. The research presented in this paper uses a Convolutional Neural Networks model to extract opinion aspects from more than 3 million user-contributed Yelp restaurant reviews. The results show high homogeneity among cities in terms of the average proportions of aspects in restaurant reviews. In addition, for fast food chains, “location” is the only aspect category reviewed proportionally higher than independent fast food restaurants. An analysis of the co-occurrences of “location” indicates that the identity of chain restaurants stems from the comparison between the same chain of different locations whereas the identity of the independent restaurants is more diverse, implying the intricacies of placeness of urban stores. This research demonstrates that fine-grained sentiment analysis (i.e., opinion aspect extraction and analysis) with geo-tagged text data is fruitful for studying nuanced place perceptions on a large scale.
KEYWORDS: Urban Places, Convolutional Neural Networks, Aspect-based Sentiment Analysis
Figure 1: Illustration of an example of a CNN layer.
Figure 3: Mapping restaurants in NV, AZ, PA, NC, WI, IL. Not all cities are shown in each state. Only cities have data that accounts for the majority of the restaurants in that state are mapped, for the sake of visual clarity.
Figure 6: Average proportions of aspect categories for chain and independent fast food restaurants for two kinds of cuisine (American, Mexican) in Las Vegas, Phoenix, and Charlotte, normalized by dividing the mean for comparison.
Yuan X. and Crooks A.T. (2019), Assessing the Placeness of Locations through User-contributed Content, in Gao, S., Newsam, S., Zhao, L., Lunga, D., Hu, Y., Martins, B., Zhou, X. and Chen, F. (eds.), Proceedings of the 3rd ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery (GeoAI), Chicago, IL. pp. 15-23. (pdf)

Thursday, October 31, 2019

Talk: Utilizing Agent-based Models and Open Data to Examine the Movement of People and Information

Earlier this month I was invited to give a talk as part the Criminal Investigations and Network Analysis Center (CINA) Distinguished Speaker Series. As readers of the blog might expect, I chose to talk about how open data (e.g. OpenStreetMap, Twitter) can be utilized in agent-based models to study a variety of applications (many of which can be found over on my research page). The talk itself was entitled: "Utilizing Agent-based Models and Open Data to Examine the Movement of People and Information: A Gallery of Applications." Below you can read the brief abstract of the paper and if this peaks your interest, CINA recorded my talk and highlighted (short) version  is given below (while the full talk can be found at: https://youtu.be/iIvSnE-IBZI).

Today we are awash with many new forms of open data (e.g. crowdsourced, social media), but we are still challenged with how individuals make decisions and how this leads to more aggregate patterns emerging. One way to explore how individuals make decisions, or are impacted by information and their resulting consequences, is via agent-based modeling. Agent-based modeling allows for simulating heterogenous actors and their decision-making processes within complex systems. Through a series of example applications ranging from the small-scale movement of pedestrians over seconds, to that of the movement of people over borders over hours and days, I will demonstrate how open data can be leveraged within the agent-based building process. Specifically, the examples will show that by focusing on individuals, or groups of individuals and the networks that connect them, more aggregate patterns emerge from the bottom up.