Dr. Ashwin Bahulkar presented a demo titled "Modeling and predicting human social behavior" co-authored by Boleslaw K. Szymanski, Kevin Chan and Omar Lizardo at KDD'2019 in Anchorage, AK, on August 3, 2019

Dr. Ashwin Bahulkar presented a demo titled "Modeling and predicting human social behavior" co-authored by Boleslaw K. Szymanski, Kevin Chan and Omar Lizardo at the showcase session at the 25th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD'2019, in Anchorage, AK, August 4-8, 2019. The goal of the project is to predict how groups in social networks evolves over time. We created a model that predicts which of many potential changes to group membership or opinions on social issue each person in a group will make. Those changes include: changing opinion on an issue, joining a new group, or leaving a group to which the person currently belongs. We propose a benefit driven model in which the set of changes predicted to be made maximizes the benefits that a person derives from groups membership. We define benefit as a function of the interactions the person has with the group, the similarity the person has with other group members on various opinions and the number of groups the person is part of, where weights between these change options are defined by model parameters. The optimal values of model parameters are computed by minimizing penalty for changes predicted by the model but not made in reality and for changes not predicted as made by the model but made by the node in reality. There is no penalty for changes predicted correctly or changes not made and predicted as such. As usual, to find model parameters, we use past data on group dynamics. In our experiments, we use part of the NetSence dataset as the training set to learn optimal values of the model parameters. This is done by computing the penalty for every possible set of changes for each person and then choosing the one with the lowest penalty for prediction. We further propose a machine learning based model which uses many more parameters to predict if a set of changes will be made by a person or not. More details are provided in the attached abstract.