A Prototype Model to Predict Human Interest: Data Based Design to Combine Humans and Machines

In this paper, the possibility of quantifying a person's interest using data-driven algorithms is investigated. In doing so, interest estimation problem is formulated as a latent state estimation problem, and an answer is deduced via Bayesian Inference. First, a Subjective-Objective approach is used to measure activity. Through this calculated activity, the method indirectly infers human latent state values. A formulation of interest is then presented by drawing inspiration from the Ornstein-Uhlenbeck (OU) process in Physics. Moreover, concepts of stochastic volatility are employed to vary the instantaneous volatility of the OU process. This is done to further improve the performance. Subsequently, the convergence speed of the OU process is varied with time. A novel statistical framework is discussed that dynamically transforms interest into activity. Each of these individual contributions is combined to present a solution via Monte Carlo Simulations. To demonstrate the efficacy of the proposed method, numerical simulations are performed on real datasets. Lastly, a prototype is engineered and the method is implemented as a RESTful Web service. The prototype is hosted as a Web service on several Virtual Machines to demonstrate the practical feasibility of the framework in cloud-based deployment scenarios.