Bill Rand

Social Media provides a popular platform for marketers to provide content to their consumers and stakeholders. However, many organizations also fear the ability that social media gives to malignant users to spread disinformation. The fear of the potential effects of malicious influencers (e.g., bots, trolls, extremists) spreading falsehoods or attempting to radicalize the general public on social media has been recognized since as early as 2005. In recent years, these concerns have been validated and pose a significant threat to public opinion and internet word of mouth. In this project, we propose a methodology based on functional data techniques to understand the users' daily behavior on these social media platforms. We will model the user data as a ����������������curve��������������� with 0/1 values or categorical values and understand a user������������������s behavior on a social media platform by pooling across all users of the same kind. Our methods have the potential to identify the minimum number of days over which to follow users to determine their true nature (eg. genuine user or malicious). In this proposal, we will also investigate functional-data based methods that focus on the interpretability of the results, so that we can formally assess whether there are certain times during the day when the social media activity is different across multiple types of users. The methods will be deployed as packaged in R and will be applied to several social media data.

Performers at NC State and the Business Analytics Initiative will be responsible for the following tasks: 1. Collecting and collating data from Twitter, Reddit, News, and other data sources. This data will be collected using the NCSU Brandwatch license and the Twitter Academic API license. 2. Providing analysis of this data and featurizing this data by using natural language processing tools, including, but not limited to, sentiment analysis, named entity recognition, semantic analysis, topic identification and tracking, and bot detection. 3. Providing feedback and suggestions on the use of the data in UCF���s transfer entropy influence pathway identification solution. 4. Meeting with collaborators at UCF on a regular basis to help determine the scope and the overall progress of the project. 5. Presenting the analysis and work at local and international venues.

The NC State Business Analytics Initiative (BAI) will collaborate with Workplace Options to identify trends in workplace culture at companies that utilize Workplace Options services. Of particular interest are trends that indicate there is an adverse workplace cultural element present. This will allow Workplace Options to engage with organizations at risk. Additionally the BAI will look at trends across companies. This analysis will facilitate discussions Workplace Options has with companies to improve their workplace culture and organizational effectiveness.

The objective of this project is to create an informative and timely newsletter for CHAI users that highlights various trending topics from the world������������������s most read literature. The initial focus of the newsletter will be on managing the SARS-Cov-2 virus and the disease it causes, Covid-19.

NC State will support Perceptronics on this effort in the areas of: Anomaly Detection using Causal State Models, Image Analysis, Natural Language Processing.

Dr. Rand will utilize his expertise in social media, information diffusion, agent-based modeling, and evolutionary computation to advise the team on several major aspects of the prime contract. In addition, he will help with the supervision of students and / or postdocs involved with the project. In particular, Prof. Rand will facilitate the following four major areas of the project: (1) Helping with the development of the Deep Agent Framework (DAF) Modeling Markup Language (DMML) (2) Assisting in the creation of three plausible submodels for the emotional, social, and cognitive components of DMML (3) Work on the architecture and development of the Modeling for Information Spread and Evolution in Online Environments (MISEOE) (4) Collaborate on the development and articulation of the information Evaluation module within the MISEOE"

Specifically the project will attempt to answer the following questions: 1. Can we use social media data (Twitter) to identify self-suspected cases of COVID-19 in LMIC countries in Sub-Saharan Africa, and then predict future caseload? 2. What is the best way to collect and organize social media data about the concerns of Health Care Workers from LMIC countries in Sub-Saharan Africa related to the COVID-19 pandemic? And what initial information can be found in this data? 3. The last question is more exploratory in nature than the previous two. Can we start to characterize patients������������������ journeys through COVID-19 based on their Twitter timeline?

Specifically the project will attempt to answer the following questions: What is the current status of clinical risk factors that correlate to disease severity? Type of CVD or other condition, age, gender, medication use, ethnicity, type of exposure. What are the most commonly used treatments for managing COVID disease and symptoms? What do we know about how much is being used per patient? (This will help triage the WHO list.) What treatments for COVID-induced pneumonia work best?

PAIT - Statement of Work NCSU NC State will support Perceptronics in this effort in the following roles: Anomaly Detection using Causal State Models - A key component of PAIT involves determining if a particular datastream tied to an individual is exhibiting anomalous behavior. We will take the Causal State Modeling approach that has been developed for batch data and construct an online version of it, by combining this with new methods of using CSM to detect anomalies we will build an online anomaly detector for streaming data. Image Analysis - We will work with Perceptronics to develop methods for detecting threats in images. The NCSU team has experience working with Convolutional Neural Networks and processing social media images. Building on this and applying new transfer learning methods, we will work with Pereceptronics to develop new approaches to detect threats in online images. Natural Language Processing - The NCSU team has considerable experience with analyzing social media data. Though NCSU will not be developing new natural language methods, they will play an advisory role in helping with the natural language processing of the social media data used for this contract.

LAS DO1 Rand -3.3 Computational Social Sciences