Understanding and evaluating human behaviour: an application of psychology and machine learning

Abstract

In this thesis, we aim to understand human behavior and the associated humanistic properties through the use of computational methods. The work presented in this thesis is motivated by the fact that we as human beings do not completely understand the internal mental properties, e.g. Motivation, Interest, Altruism, of other people. It is indeed a challenge to have a mechanical machine do this. It could therefore very well be said that to look at such properties through the eyes of an artificial computational agent is non-trivial. In this thesis, we take on this challenge and try to show that there is a way through which we can handle the issue computationally. In doing so, our goal is to take one more step towards understanding the psychological properties of human beings through artificial agents. This is done by combining the core principles of two different fields of research: Psychology and Machine Learning. Further, to conduct a study of the humanistic properties, we perform the analysis of the human psychological properties in crowd based systems. These systems are chosen as they have a natural affinity towards both man and machine. Therefore, they present an excellent opportunity to focus on the technology of the machine and the psychology of the human simultaneously. In doing so, we address two major issues in the thesis.First, we aim to devise efficient techniques of addressing the challenge of enhancing user participation. The objective is to understand the psychological conditions that make people participate at the online platform and simulate them in a computational environments. The goal here is to make these systems more productive and labour & cost-effective. This is done via analyzing group interactions and collaborative processes at these online platforms. To do this, we divide the problem into two categories: 1) we borrow elements from Machine Learning and propose a recruitment strategy that selects an individual so that the probability of getting a response is maximized; 2) we dig deep into human psychology and try to find alternate means of promoting user participation. We look for new and otherwise overlooked patterns in the behavior of people to find a few interesting facts. Once we accomplish the previous objective(s), i.e. we are able to motivate users and generate their interest, we then move to the next significant challenge addressed in this thesis. We propose a framework that tries to quantify the interest of an individual towards any entity (sayFacebook, StackOverflow, Amazon Mechanical Turk and so on). Through the proposed framework, we make an attempt to model the long-term evolution of a person’s interest. Furthermore, we estimate try to interest at any given day, hour, minute, and so on. The problems addressed in this thesis are validated by performing simulations on one of the most mature crowdsourcing data repositories on the Internet: StackOverflow. The results show promise especially considering the fact that we have attempted to answer questions and explore some of the previously unexplored patterns in human behavior. We will show that the work carried out in the thesis complements existing literature, and at times, open a few new and previously unexplored dimensions of research in man-machine systems.