Wednesday, September 16, 2020

Learning-based Actor-Interpreter State Representation

While in previous posts we have discussed machine learning (ML) with respect to social media analysis, we have also been exploring how one can use it in agent-based modeling. One of the first examples of this is a new paper with Paul Cummings which has been accepted at the upcoming 2020 International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation (or SBP-BRiMS 2020 for short).

In the paper entitled "Development of a Hybrid Machine Learning Agent Based Model for Optimization and Interpretability" we discuss the growth of ML within agent-based models and present the design of the hybrid agent-based/ML model called the Learning-Driven Actor-Interpreter Representation (LAISR) Model. LAISR's attempts to: "a) generate an optimal decision-making strategy through its training process, including a more constrained parameter space, and b) describe its behaviors in a human-readable and interpretable approach." To demonstrate the LAISR's model we use a simple wargaming example, that of a tactical air and ground warfare as an experiment and discuss areas of further research and applications.

If this is of interest to you, below we provide the abstract to the paper, along with a high level overview of the LAISR's model, its tactical experiment diagram and some results from the wargaming experiment. This is followed by a short movie of a representative model run that Paul has created. Finally at the the bottom of the post you can see the full reference and a link to the paper itself.

Abstract.
The use of agent-based models (ABMs) has become more wide-spread over the last two decades allowing researchers to explore complex systems composed of heterogeneous and locally interacting entities. How-ever, there are several challenges that the agent-based modeling community face. These relate to developing accurate measurements, minimizing a large complex parameter space and developing parsimonious yet accurate models. Machine Learning (ML), specifically deep reinforcement learning has the potential to generate new ways to explore complex models, which can enhance traditional computational paradigms such as agent-based modeling. Recently, ML algorithms have proved an important contribution to the de-termination of semi-optimal agent behavior strategies in complex environments. What is less clear is how these advances can be used to enhance existing ABMs. This paper presents Learning-based Actor-Interpreter State Representation (LAISR), a research effort that is designed to bridge ML agents with more traditional ABMs in order to generate semi-optimal multi-agent learning strategies. The resultant model, explored within a tactical game scenario, lies at the intersection of human and automated model design. The model can be decomposed into a format that automates aspects of the agent creation process, producing a resultant agent that creates its own optimal strategy and is interpretable to the designer. Our paper, therefore, acts as a bridge between traditional agent-based modeling and machine learning practices, designed purposefully to enhance the inclusion of ML-based agents in the agent-based modeling community.

Keywords: Agent-Based Modeling, Machine Learning, Explainable Artificial Intelligence.

LAISR model

LAISR tactical experiment diagram (A). Actor finite state machine (B).

Heat map representations of actor agents.



Full Reference:
Cummings, P. and Crooks, A.T. (2020), Development of a Hybrid Machine Learning Agent Based Model for Optimization and Interpretability, 2020 International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation, Washington DC.
 If you would like a pre-print of  paper, just let us know and we can email you one. 

Friday, September 11, 2020

Utilizing ABMs for The Human Resource Management

In a previous post from a few years ago we looked at how the workplace the layout might impact subordinates interactions with managers. Now turning to work employee satisfaction within the workplace, at the forthcoming International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation (or SBP-BRiMS for short) we have a paper entitled "The Human Resource Management Parameter Experimentation Tool".

In this paper we have created a model called the "Human Resources Management-Parameter Experimentation Tool" or HRM-PET for brevity, which is based on Herzberg et al. (1959) Two-Factor Theory. This theory has been used and tested for decades in human resource management as it can capture the interaction between a work force’s motivation and their environment’s hygiene. Hygiene in this context relates to policies and administration, supervision-technical, relationship-superior, working conditions, and salary which together moderate job dissatisfaction. While the theory has been extensively used it has not been explored via an agent-based model until now. By utilizing agent-based modeling, it allows us to test the empirically found variations on the Two-Factor Theory and its application to specific industries or organizations.

If you are interested in finding out more about this work, below we provide the abstract to the paper, an annotated graphical user interface of the model along with the basic decision making process for the agents. This is followed by some of the results from the model and movie of a representative model run. At the bottom of the post we provide the full citation to the paper, along with that of Herzberg et al. (1959). The model itself which was created in NetLogo 6.1, can be found along with a detailed Overview, Design concepts, and Details plus Decision making (ODD+D) document at http://bit.ly/HMR-PET. The rationale for utilizing the ODD+D and for sharing the model is that it allows broader dissemination of the model and its methodology.

Abstract:
Human resource management (HRM) draws on the field of organizational theory (OT) to identify, quantify, and manage people-based phenomena that impact organizational operations and outcomes. OT research has long used computational methods and agent-based modeling to understand complex adaptive systems. Agent-based modeling methodologies within HRM, however, are still rare. Within the HRM and management science literature, Herzberg’s et al. (1959) Two-Factor Theory (TFT) is a framework that has been tested and used for decades. Its ability to capture the interaction between a work force’s motivation and their environment’s hygiene lends itself well to agent-based modeling as a method of study. Here, we present the development of the Human Resources Management-Parameter Experimentation Tool (HRM-PET) as the first explicit ABM instantiation of TFT, filling the gap between the study of HRM and computational OT tools like agent-based modeling. 

Keywords: Human Resources Management, Management Science, Workforce Dynamics, Agent-based Modeling.
HRM-PET graphical user interface.
Decision making process for the agents in HRM-PET.
Worker congregation to work units under three variations of work unit hygiene factor distributions and two variations of weighing worker satisfaction and dissatisfaction.



References:
Herzberg, F.I., Mausner, B. and Snyderman, B. (1959)The Motivation to Work (2nd ed.). New York: John Wiley.
Iasiello, C., Crooks, A.T. and Wittman, S. (2020), The Human Resource Management Parameter Experimentation Tool, 2020 International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation, Washington DC.

If you would like a pre-print of  paper, just let us know and we can email you one. 

Tuesday, September 01, 2020

Beyond Words: Comparing Structure, Emoji Use, and Consistency Across Social Media Posts

Continuing our work on Emojis, at the forthcoming International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation (or SBP-BRiMS for short) we (Melanie Swartz, Arie Croitoru and myself) have a paper entitled "Beyond Words: Comparing Structure, Emoji Use, and Consistency Across Social Media Posts." In the paper we introduce and demonstrate a language-agnostic methodology to characterize structures of content and emoji use within a document (in this case a tweet), measure consistency of structures across a set of documents, and cluster documents and users with similar patterns and behavior. Using a corpus of 44 million tweets collected in October and November 2018 related to the 2018 U.S. midterm elections based on keywords, hashtags, and user accounts associated with candidates or political parties we were able to gain insights into the unique or shared structures of communication styles and emoji use of over 3.3 million unique users and user roles such as journalists, bots and others. If this sounds of interest to you, below we provide the abstract to the paper, some tables and figures of the our findings along with the full reference and a link to the paper. Furthermore, if you are interested in extending this work to other areas, Melanie has made the code available at https://github.com/msemoji/.

Abstract
Social media content analysis often focuses on just the words used in documents or by users and often overlooks the structural components of document composition and linguistic style. We propose that document structure and emoji use are also important to consider as they are impacted by individual communication style preferences and social norms associated with user role and intent, topic domain, and dissemination platform. In this paper we introduce and demonstrate a novel methodology to conduct structural content analysis and measure user consistency of document structures and emoji use. Document structure is represented as the order of content types and number of features per document and emoji use is characterized by the attributes, position, order, and repetition of emojis within a document. With these structures we identified user signatures of behavior, clustered users based on consistency of structures utilized, and identified users with similar document structures and emoji use such as those associated with bots, news organizations, and other user types. This research compliments existing text mining and behavior modeling approaches by offering a language agnostic methodology with lower dimensionality than topic modeling, and focuses on three features often overlooked: document structure, emoji use, and consistency of behavior.
Keywords: Data Mining, Social Media, Emojis, User Behavior Modeling.
Emoji attributes

Most common content structures with emojis for non-retweets.

Clusters of users with similar behavior across two factors in non-retweets (left) and retweets (right) Colors indicate cluster assignments.

Full Reference: 
Swartz, M., Crooks, A.T. and Croitoru, A. (accepted), Beyond Words: Comparing Structure, Emoji Use, and Consistency Across Social Media Posts, 2020 International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation, Washington DC.

If you would like a pre-print of  paper, just let us know and we can email you one.

 

Monday, August 31, 2020

Beginning A New Chapter

Just a short post to say that over the summer saw us move from Northern Virginia to Buffalo to start a new chapter in our lives and research.  More specifically I have become a Professor in the Department of Geography and  a member of the RENEW (Research and Education in eNergy, Environment and Water) Institute at the University at Buffalo.

It does not seem that long ago I wrote a similar post about moving from CASA to George Mason University (but that was over 10 years ago!). I guess it does not feel that long as I really enjoyed my time at George Mason University and the people (faculty and students) who I worked with there.  I am also supper excited to see where this new chapter leads.


Wednesday, July 22, 2020

Diversity from Emojis and Keywords in Social Media

Building on our initial work on emojis  use and and how one can carry out a systematic comparison of emojis across individual user profiles and communication patterns within social media, we have a new paper entitled: "Diversity from Emojis and Keywords in Social Media" which was presented at the 11th International Conference on Social Media and Society

In the paper we present a novel method using a diversity language model to associate diversity related attributes to social media user accounts and content by analyzing the emojis and keywords used (in this case from Twitter). We used this diversity language model to shed light on the groups of social media users and content with similar diversity attributes related to American politics (specifically the 2018 U.S. midterm elections). Our results revealed topics of interest and patterns of social media engagement across political lines among the diverse populations that otherwise would not have been apparent if we only analyzed the key political campaign phrases and slogans (i.e. “Blue Wave” and “Make America Great Again”) without taking diversity into account.

For interested readers, below we provide the abstract to the paper along with some figures from the paper. These include our workflow for diversity analysis of social media content, a high level overview of our diversity language model. These are followed by some of our results. Specifically the presence of diversity keywords and emojis in user profiles, and the composition of users in our collection based on gender for two political campaigns. If this peaks you interest as the conferce was virtual we have also prepared a short movie of the paper. While at the bottom of the post you find the full reference to the paper along with a link to the paper itself.

Abstract:
Social media is a popular source for political communication and user engagement around social and political issues. While the diversity of the population participating in social and political events in person are often considered for social science research, measuring the diversity representation within online communities is not a common part of social media analysis. This paper attempts to fill that gap and presents a methodology for labeling and analyzing diversity in a social media sample based on emojis and keywords associated with gender, skin tone, sexual orientation, religion, and political ideology. We analyze the trends of diversity related themes and the diversity of users engaging in the online political community during the lead up to the 2018 U.S. midterm elections. Our results reveal patterns along diversity themes that otherwise would have been lost in the volume of content. Further, the diversity composition of our sample of online users rallying around political campaigns was similar to those measured in exit polls on election day. The diversity language model and methodology for diversity analysis presented in this paper can be adapted to other languages and applied to other research domains to provide social media researchers a valuable lens to identify the diversity of voices and topics of interest for the less-represented populations participating in an online social community.

Keywords: Social media, emoji, diversity, elections, political campaigns
Workflow for diversity analysis of social media content
Diversity Language Model
Presence of diversity keywords and emojis in user profiles
Composition of users in our collection based on gender for two political campaigns



Full Reference:

Swartz, M., Crooks, A.T. and Kennedy, W.G. (2020), Diversity from Emojis and Keywords in Social Media, in Gruzd, A., Mai, P., Recuero, R., Hernández-García, A., Lee, C.S., Cook, J., Hodson, J., McEwan, B and Hopke, J. (eds.), Proceedings of the 11th International Conference on Social Media & Society, Toronto, Canada, pp 92-100. (pdf)