Modelling Evolving Voting Behaviour on Internet Platforms: Stochastic Modelling Approaches for Dynamic Voting Systems

摘要

Markov Decision Processes (MDPs) and their variants are standard models in various domains of Artificial Intelligence. However, each model captures a different aspect of real-world phenomena and results in different kinds of computational complexity. Also, MDPs are recently finding use in the scenarios involving aggregation of preferences (such as recommendation systems, e-commerce platforms, etc.). In this paper, we extend one such MDP variant to explore the effect of including observations made by stochastic agents, on the complexity of computing optimal outcomes for voting results. The resulting model captures phenomena of a greater complexity than current models, while being closer to a real world setting. The utility of the theoretical model is discussed by application to the real world setting of crowdsourcing. We address a key question in the crowdsourcing domain, namely, the Exploration Vs. Exploitation problem, and demonstrate the flexibility of adaptation of MDP-based models in Dynamic Voting scenarios.