2024-01-29: 16th International Conference on Social Computing, Behavioral-Cultural Modeling & Prediction and Behavior Representation in Modeling and Simulation (SBP-BRiMS) 2023 Trip Report

By -

The 16th International Conference on Social Computing, Behavioral-Cultural Modeling & Prediction and Behavior Representation in Modeling and Simulation (SBP-BRiMS) 2023 was a hybrid conference with the in-person event at Carnegie Mellon University, Pittsburgh, and virtual attendees joining via Zoom. The SBP-BRiMS 2023 conference took place from September 20 to 22. SBP-BRiMS is a multidisciplinary conference with a selective single paper track and poster session. The conference also invites a small number of high-quality tutorials and nationally recognized keynote speakers. The conference has grown out of two related meetings, SBP and BRiMS, which were co-located in previous years. 


Six members of the Storymodelers Lab and the Web Science and Digital Libraries (WS-DL) research group participated in the conference. Three of them were selected as speakers thanks to the acceptance of their paper proposals, while the other three presented their research in the form of posters.


Gates Hillman Complex - Carnegie Mellon University, Pittsburgh


During the days of the SBP-BRiMS conference, participants were immersed in a varied agenda of activities and sessions with topics of high social interest. The most outstanding topic was using natural language processing (NLP) to address and understand problems such as xenophobia, feelings, emotions, and others.

Day 1: September 20, 2023

Session 1: Harmful Speech and Controversy

The conference's first session focused on Harmful Speech and Controversy issues. During this session, four presentations were given, each providing a unique perspective, offering relevant information and possible solutions from a research-based perspective. Highlights and learnings from each presentation are detailed below. 

1.1. Edinam Klutse, Samuel Nuamah-Amoabeng, Hanjia Lyu and Jiebo Luo. Dismantling Hate: Understanding Hate Speech Trends Against NBA Athletes (DOI: 10.1007/978-3-031-43129-6_8)

Edinam Klutse and Samuel Nuamah-Amoabeng from the University of Rochester made the conference's first presentation. These two researchers shared their work on hate speech reactions of fans to athletes. They were motivated by societal issues about racial and gender discrimination. They relied on data collected from Twitter and used advanced language models, such as BERT and Transformer, to analyze fan responses to NBA players. In their results, Anthony Davis emerged as the player most often mentioned with hateful words in tweets. The researchers also identified factors of discrimination and racism directed toward the players. The black community was the most affected, with 4,124 hate tweets, followed by the LGBT community with 2,938 tweets. Racial discrimination was highlighted as the predominant factor, and its correlation with player performance was explored.


Essential lessons and ideas for future research emerge from this presentation. First, the findings of this presentation highlight the need to address racial and gender discrimination in sports more actively. Additionally, we can see that hate speech is an undesirable by-product of the relationship between fans and athletes. The above can affect players' performance, given the direct involvement of their emotions and feelings. On the other hand, the researchers' findings show evidence of the usefulness of artificial intelligence (AI) in understanding this relationship. Delving deeper into the subject and obtaining more information would be interesting. One thing that could be done is to analyze performance statistics before and after hate posts to see how athletes are affected. One could also expand the approach to other social networks and platforms to get a complete picture of the interactions between fans and players. From a more social and isolated computational approach, exploring effective strategies to counter discrimination and foster a more inclusive environment in the sports world could be interesting.

1.2. Kai Chen, Zihao He, Rong-Ching Chang, Jonathan May and Kristina Lerman. Anger Breeds Controversy: Analyzing Controversy and Emotions on Reddit (DOI: 10.1007/978-3-031-43129-6_5)

In this study, researchers studied the role of emotions in online conflict, focusing on the Reddit forum discussion. They used a multilingual emotion detection model. This model is based on SpanEmo, the state-of-the-art in emotion detection. Their results reveal that controversial comments express more tremendous anger and less positive emotions, influencing subsequent comments' emotional tone. The researchers highlight the predictive power of emotions in identifying controversial words, underscoring the importance of understanding emotional dynamics to manage online conversations effectively. 

Although we could not capture pictures or upload evidence to Twitter, we were in person at the presentation, where we learned about the importance of emotional analysis in online conflict management. This study opens the door to future research to explore how the relationship between emotions and online controversy could generate online community management practices, fostering a more constructive and collaborative user environment.

1.3. Jennifer Golbeck. Misogyny, Women in Power, and Patterns of Social Media Harassment (DOI: 10.1007/978-3-031-43129-6_1)

Jennifer Golbeck, University of Maryland, shared her research to answer the question, "Does the act of posting misogynistic slurs at women in power predict a higher general rate of slur use on social media?" To answer this question, she extracted data from Twitter from January to May 2022, and she was particularly interested in tweets that mentioned Kamala Harris, which already contained a considerable number of tweets—4265 to be precise—that included mentions of Kamala Harris with derogatory language, such as the word "bitch." Dr. Golbeck was determined to expand their analysis by identifying more individuals engaging in similar behavior. Her mission was clear: collect an additional 200 tweets from the users responsible for these misogynistic tweets.


Dr. Golbeck uncovered some intriguing patterns. Kamala Harris appeared to be a frequent target of misogynistic language on the platform, making her one of the most frequently mentioned women slurs and harassment on Twitter during that time. Curiosity led her to explore whether the behavior differed on alternative social media platforms. She turned her attention to Parler, a platform known for attracting users from the far-right political spectrum who had migrated from Twitter. Surprisingly, they found a stark contrast; there were zero misogynistic comments directed at Republican figures on Parler. Not satisfied with the results, Dr. Golbeck broadened her search to encompass Kamala Harris and Nancy Pelosi, another prominent female political figure. For this, she used a "slur detection" method, leveraging a slur list provided by the University of Sheffield. Her narrative became a fascinating exploration of how political discourse and misogyny intertwined on social media platforms, leading to thought-provoking questions about online behavior and the dynamics of different platforms. Dr. Golbeck was determined to uncover more insights from the ever-evolving digital communication landscape.


1.4. Rong-Ching Chang, Jonathan May and Kristina Lerman. Feedback Loops and Complex Dynamics of Harmful Speech in Online Discussions (DOI: 10.1007/978-3-031-43129-6_9)

A diverse team of researchers across multiple languages set out to investigate harmful speech and its connection to political discourse on social media. They employed the XLM-R model for language analysis, revealing a link between negative emotions, harmful speech, and the political context of online discussions. They used a Naive Bayes classifier for text classification and explored system dynamics approaches for more comprehensive insights. Adding additional variables to their model was also under consideration by them. 


Panel 1: Simulation Challenges with Human-Machine Teaming 

In a thought-provoking panel discussion, experts gathered to explore the dynamic world of human-machine collaboration and its unique challenges. The panel was moderated by Geoff Dobson, who expertly guided the conversation among the following distinguished speakers:

  1. Anita Wolley, Ph.D.: As the panel began, Dr. Anita Woolley, also known as Katie, took the stage and shared her insights on a fascinating topic using a social media emulator. With visual aids displayed on slides, she delved into the intricacies of this technology, shedding light on its impact on human-machine interactions.
  2. Leslie Blaha, Ph.D.: Dr. Blaha addressed a crucial question: "How do we design a good teammate, human or machine?" She delved into the role of individual differences in shaping human interactions with technology. Highlighting the relationship between the use of smart devices and trust in the technology, she outlined the key characteristics necessary for someone or something to be considered a valuable teammate. The audience was intrigued to discover that older individuals often anthropomorphize technology, referring to it as "he" or "she" rather than "it."
  3. Lingfei Wu, Ph.D.: Dr. Wu took the stage to explore an intriguing concept: Are ideas increasingly challenging to discover? His research suggested that the missing piece in the puzzle could be the team itself. By associating the team's size with progress, he explained how larger teams are more likely to achieve groundbreaking discoveries. He backed up his theories with data from scientific papers spanning over 3.5k cities, discussing two categories of interdisciplinary teams and the benefits of remote collaboration. The main takeaway from his presentation was the call to "Reinvent team science to find new ideas" and to consider whether AI could help generate ideas or provide invaluable support in data analysis.
  4. Dustin Updyke: The panel was enriched by Dustin Updyke's insights on open-source tools. He discussed these tools and their significance in human-machine teaming.

The discussion touched on the challenges and opportunities in human-machine collaboration, sparking curiosity and encouraging further exploration of this intriguing field. Also, Joe Lyons, Ph.D., was part of the panel discussion.

Session 2: Media and Language Use

2.1. Brian Llinas, Guljannat Huseynli, Erika Frydenlund, Katherine Palacio, Humberto Llinas and Jose J. Padilla. Assessing Media's Representation of Frustration Towards Venezuelan Migrants in Colombia (DOI: 10.1007/978-3-031-43129-6_13)

We addressed the challenge of migration from Venezuela to Colombia using a comprehensive approach that combined qualitative and quantitative elements. Our study analyzed media coverage, explicitly examining news stories that expressed concerns about migration, infrastructure, governance, and geopolitics.


To carry out their research, we used trajectory models to explore the relationship between the flow of Venezuelans migrating to Colombia due to the economic collapse and the frustrations generated by how the media addressed specific issues in our news. The results of the study yielded significant conclusions. In particular, we observed that the increase in the Venezuelan population in Colombia harmed local infrastructures, such as hospitals and schools, generating more significant tensions. In addition, the study revealed that this population increase increased tensions and frustrations between the local population and the migrants and towards the government. However, it was interesting to note that, in contrast, frustrations related to geopolitics decreased.


This research allows us to generate reflections on how this methodology could be implemented in other countries to compare the experiences of frustration from different perspectives in migrant-receiving nations.



2.2. Wenjia Hu, Zhifei Jin, and Kathleen M. Carley. Vulnerability Dictionary: Language Use During Times of Crisis and Uncertainty (DOI: 10.1007/978-3-031-43129-6_11)


In this presentation, the researchers explored the expression of vulnerability, considering it a psychological state that people experience in times of uncertainty or emotional crisis. Their approach was based on a comprehensive review of the literature related to vulnerability, and they presented a psycholinguistic dictionary addressing common themes in this context. Validation of the dictionary was carried out by using AI in combination with human labeling to compare and contrast the results, thus ensuring the reliability and accuracy of their findings.


One of the things that could be interesting to explore based on this presentation would be a multicultural validation of the psycholinguistic dictionary. You could investigate how this dictionary behaves in different cultures and languages. We can ask questions such as: Are there significant differences in the expression of vulnerability across cultures? How can the dictionary be adapted to be more effective in different contexts, how can the dictionary be adapted to be more effective in other contexts, and how can the dictionary be adapted to be more effective in different contexts?


2.3. Jose Mari Luis Dela Cruz and Maria Regina Justina Estuar. Classifying Policy Issue Frame Bias in Philippine Online News. (DOI: 10.1007/978-3-031-43129-6_7)


This was a virtual presentation where Jose Dela Cruz presented his study on the impact of the media in shaping public opinion through the selection and presentation of policy issue frames in Philippine online news articles. Using a BERT model, he observed that over time, there was a trend towards more excellent uniformity in policy issue framing across media, even on initially ambiguous issues, highlighting the presence of framing biases in news coverage in the Philippine media.

2.4. Xuanlong Qin and Tony Tam. Stereotype Content Dictionary: A semantic space of 3 million words and phrases using Google News Word2Vec embeddings (DOI: 10.1007/978-3-031-43129-6_2)


This was also an online presentation, where Xuanlong Qin expressed that his study covered an efficient method to expand the stereotyped content of texts. He developed a Stereotyped Content Dictionary based on Google News Word2Vec embeddings. His methodology was based on a semantic differential model that reduced the semantic space to two dimensions. Their result showed that the Stereotyped Content Dictionary significantly outperformed other models in predicting personality traits, thus presenting a valuable tool for investigating stereotypes from large datasets.

Session 3: Tech Demos

During session 3, there were many presentations that we were unable to attend because we had a dessert break. However, we used this time to meet new people and establish contacts with researchers from other universities. We especially enjoyed meeting Tobin South, a Ph.D. student at MIT who connected us with some of his colleagues who align with our computer science and social science research. One of these contacts was Isabella Loaisa, a Colombian who works in social science from computational approaches.

Keynote #1: Dr. Nathaniel Persily


Dr. Nathaniel Persily's keynote was about the impact of AI on elections and democracy. Dr. Persily began by acknowledging the uncertainty that shrouds the influence of AI in these domains. He emphasized that AI accelerates the strategies of both good and bad actors in the ever-evolving social media landscape. This phenomenon, he suggested, has far-reaching implications that extend beyond the boundaries of the information ecosystem. However, he cautioned that while technology certainly plays a role, the larger sociopolitical forces at play are equally, if not more, significant.


One startling revelation Dr. Persily shared was the massive exodus of election officials, with retirements reaching an unprecedented 30-40%. This alarming trend adds a layer of complexity to the already intricate challenges facing democracy.


Moving from the familiar territory of Twitter, he alluded to the cryptic "X," hinting at the ever-expanding influence of technology and AI in shaping the political landscape. He noted that the legal terrain surrounding these developments is still being determined, further complicating the path forward.


Dr. Persily went on to highlight the disinformation problem exacerbated by AI. As these technologies become more sophisticated, spreading false information becomes a pressing concern.


He then steered the conversation toward reevaluating the relationship between AI and democracy. He presented a framework for action, discussing key aspects such as disclosure, transparency, and verification. Auditing for political and racial bias and ensuring accuracy was deemed crucial by Dr. Persily. To counter the misuse of AI, mainly through the employment of chatbots, became a central theme.


Hyperpolarization and a palpable lack of trust in the political discourse were also brought into focus. Dr. Persily pointed out that elites often propagate falsehoods in this environment, undermining the foundations of democracy.


As he concluded his keynote, Dr. Persily left the audience with much to ponder. The intricate web of AI's influence on elections and democracy was unveiled, urging policymakers, technologists, and citizens alike to confront these challenges head-on while preserving the core principles of democracy and truth.

Day 2: Thursday, September 21

Session 1: Modeling Human Performance

1.1. Robert Thomson and William Frangia. Investigating the Use of Belief Bias to Measure Acceptance of False Information (DOI: 10.1007/978-3-031-43129-6_15)


The first presentation of Session 1 was titled "Investigating the Use of Belief Bias to Measure Acceptance of False Information" by Robert Thomson and William Frangia of the United States Military Academy in West Point, New York. In their study, Thompson and Frangia explored how people's prior beliefs can affect their judgment of whether an argument makes sense. The analysis collected data by having the United States Military Academy cadets evaluate the validity of statements related to the Afghanistan pull-out from various sources of authority. The authors found that when presented with an argument, cadets were more likely to reject it if it contained false information, even if the overall opinion was logically correct. The authors discussed how this bias in thinking was influenced by where the argument came from, such as military or civilian sources, and was also connected to certain personality traits. Overall, their study shows that our existing beliefs and the source of information can impact how we evaluate arguments.

1.2. Jeongkeun Shin, Kathleen M. Carley, and Richard Carley. Integrating Human Factors into Agent-Based Simulation for Dynamic Phishing Susceptibility (DOI: 10.1007/978-3-031-43129-6_17)


The second presentation of Session 1 was titled "Integrating Human Factors into Agent-Based Simulation for Dynamic Phishing Susceptibility" by Jeongkeun Shin, Kathleen M. Carley, and Richard Carley of the Center for Computational Analysis of Social and Organizational Systems at Carnegie Mellon University. Their study aims to develop a computer simulation framework that allows modelers to tailor it to their organization's needs. Using this framework, individual agents can be configured to respond to various cyber attack scenarios, primarily focusing on phishing attacks (often associated with human errors within the organization). These simulations are made to replicate real-life situations that will help explore how people can fall for phishing attacks. Incorporating individuals' personalities and training into these simulations would enhance their realism, ultimately helping strengthen our cybersecurity defenses.

1.3. Aadhar Gupta, Mahavir Dabas, Shashank Uttrani and Varun Dutt. Modeling human actions in the cart-pole game using Cognitive and Deep Reinforcement Learning approach (DOI: 10.1007/978-3-031-43129-6_19) 


A collaborative research project by Aadhar Gupta, Mahavir Dabas, Shashank Uttrani, and Varun Dutt explored the modeling of human actions in the cart-pole game using a blend of Cognitive and Deep Reinforcement Learning. They engaged 42 participants in their study, employing Instance-based Learning (IBL) integrated with the ACT-R cognitive architecture. The decision-making process revolved around the activation of instances, the probability of retrieval, and blended values. Mental noise and similarity thresholds were fine-tuned with hyperparameters. They also incorporated a Deep Q-Learning Network to map the quality of actions at different states, addressing the limitations of traditional methods.


The researchers evaluated their models using data from ten trials and the F1 metric, scrutinizing 126 models. Their results showed that the Deep Q-Learning Network achieved a mean F1 score of 0.8228, outperforming IBL, which scored 0.8091. Despite the simplicity of the cart-pole game with only two actions, the researchers plan to continue using IBL and DQN for more complex datasets, further bridging the gap between cognitive and reinforcement learning.

1.4. Tristan J. Calay, Basheer Qolomany, Aos Mulahuwaish, Liaquat Hossain and Jacques Bou Abdo. CCTFv1: Computational Modeling of Cyber Team Formation Strategies (DOI: 10.1007/978-3-031-43129-6_20) 


A collaborative research project by Tristan J. Calay, Basheer Qolomany, Aos Mulahuwaish, Liaquat Hossain, and Jacques Bou Abdo focused on CCTFv1, a study involving Computational Modeling of Cyber Team Formation Strategies. Drawing inspiration from sports, they explored the concept of a "Galacticos" team. Their investigation delved into the role of teams as collective intelligence, examining leadership dynamics and student design.


The project comprised attacker and defender roles, each following specific steps. The probability of success was a key consideration. They developed a simulation model with nine parameters and defined metrics to measure the effectiveness of strategies.


The research yielded results on Attacker and Defender Strategies, shedding light on their performance. The team outlined future directions, including empirical and theoretical validation of their model and pursuing more intelligent strategies for forming cyber groups. This research contributes to the evolving cybersecurity strategy and team dynamics field.


Panel 2: Humanistic Artificial Agents

  • Gwyneth Sutherlin, National Defense University
  • Laurie Fenstermacher, AFRL
  • Julie Marble, Northeastern University
  • Erin Hahn, JHU APL National Security Analysis Dept
  • Brian Hutler, Temple
  • Bill Lawless, Paine College
  • Co-Editor: Joey Jebari, JHU Berman Institute of Bioethics
  • Moderator: AM Greenberg, JHU/APL

This panel focused on the interaction between humans and autonomous machines. Experts discussed the importance of developing efficient algorithms that facilitate communication between humans and AI systems. They highlighted the need to create artificial agents that can effectively understand and respond to human needs. Additionally, they mentioned the importance of striking a balance between technological innovation and human control in the development of AI.

Session 2: Transportation and Location

1.1. Obed Domson, Jose Padilla, Guohui Song and Erika Frydenlund. A Bayesian approach to predicting the movement of internally displaced persons (DOI: 10.1007/978-3-031-43129-6_24)


The first presentation of Session 2 was titled "A Bayesian approach of predicting the movement of internally displaced persons" by Obed Domson, Jose Padilla, and Erika Frydenlund of Storymodelers Lab at Virginia Modeling, Analysis & Simulation Center, ODU, and Guohui Song of the Department of Mathematics and Statistics, ODU. The authors examined the migration patterns of internally displaced people (IDPs) in the Democratic Republic of Congo as a reaction to conflict by employing a Bayesian model to explore potential destinations beyond IDP camps. They established a metric for the appeal of each destination, and this "attractiveness" rating for a destination is computed using elevation, distance from the starting point, and the landscape. Using data from North Kivu province in Congo, the results of their model showed that distance emerges as the most significant factor influencing the choice of destination, followed by the impact of landscape and elevation.


1.2. Tobin South, Nick Lothian, Takahiro Yabe, and Alex Pentland. Building a healthier feed: Private location trace intersection driven feed recommendations (DOI: 10.1007/978-3-031-43129-6_6)


Tobin South, Nick Lothian, Takahiro Yabe, and Alex Pentland studied enhancing social media feeds with a focus on privacy and healthier user interactions. In contrast to many platforms prioritizing capturing user attention, they aimed to create a more meaningful and less intrusive user experience.


The team proposed a methodology centered around defining and measuring user attention more thoughtfully to achieve this. Their approach aimed to foster genuine connections between individuals. This involved identifying shared data points, including leveraging Bluetooth proximity data, to estimate the frequency of interactions between people.


Their research explored innovative ways to build a healthier social media feed by prioritizing privacy and fostering authentic human connections.

1.3. Kathleen Salazar-Serna, Lorena Cadavid, Carlos Franco, and Kathleen M. Carley. Simulating Transport Mode Choices in Developing Countries (DOI: 10.1007/978-3-031-43129-6_21)

In this presentation, Kathleen Salazar discussed using agent-based simulations to analyze transport mode decisions, specifically focusing on developing countries. She developed a model incorporating motorcycles, one of the preferred modes of transport in these economies. Her study, applied in a Colombian city, revealed that, without interventions, a continued increase in motorcycles and private cars is expected, aggravating congestion and traffic accidents. In addition, she found that factors such as travel time and personal safety influence transport preferences, highlighting the need for targeted interventions to promote sustainable and efficient modes of transport.

Keynote #2 - Dr. Charles "Chick" Macal

The second keynote speaker was Dr. Charles "Chick" Macal, the chief scientist for Argonne National Laboratory's Decision and Infrastructure Sciences division, whom Dr. Kathleen M. Carley introduced as a pioneering figure in agent-based modeling. In his keynote address, he discussed the challenges in modeling the COVID-19 pandemic by highlighting the remarkable response of the modeling community during that time. Dr. Macal emphasized how it initially focused on forecasting the number of affected individuals and the pandemic's duration of the pandemic but eventually shifted towards interventions such as what can be done to help. He pointed out the high demand for models during the pandemic and directed the audience to the COVID Scenario Modeling Hub, which became a valuable resource for COVID-19 teams and models. Additionally, he discussed the CityCOVID model, developed by his team at Argonne, which used agent-based modeling to understand how the virus spread in the context of the city of Chicago. Dr. Macal emphasized that extensive preparations are underway for future pandemics, with funding from various sources. He concluded his speech by offering guidance on what modelers should be aware of and the actions they must take in preparation for future pandemics, underscoring the importance of individual protective behaviors and the potential effectiveness of contact tracing programs. 

Summary


SBP-BRiMS was an enriching experience in many ways. We enjoyed every stage, from the presentation process to the trip and the presentations. We learned new techniques and approaches that are at the forefront today. We highlighted the rise of natural language processing and the use of large linguistic models to analyze and solve social problems. We also highlighted the use of simulation models to explore behaviors that allow us to have an approximation to real life. We learned that sometimes there are papers that could be more publishable but interesting enough to be seen at a conference. This is valuable because it is an avenue and an opportunity to get feedback that can move the work to the next category. From the presentations, we realized that presenting papers is not only about sharing conclusive results but also about the valuable feedback that arises during discussions and interactive sessions. 


The direct interaction with colleagues and experts gave us new perspectives and innovative approaches that will enrich our future research. We have yet to determine when and where the next SBP-BRiMS conference will be, but we know a new path to orient our prospective study. We hope to contribute significantly in social and technical areas from computational approaches, implementing natural language processing techniques on issues that still need to be addressed to provide knowledge from research for a better society. As we reflect on our participation, we appreciate the value of such conferences in fostering collaboration, staying abreast of cutting-edge research, and strengthening our ties with the broader academic community.


Published by Brian (bllin001), Jhon (Jhon_gbm12), and Himarsha (HimarshaJ)






Comments

Post a Comment