Combining humans and machines for the future: A novel procedure to predict human interest

摘要

This paper proposes a method to quantify interest. In common terminology, when we engage with an object, e.g. Online Games, Social Networking Websites, Mobile Apps, etc., there is a degree of interest between us and the object. But, owing to the lack of a procedure that can quantify interest, we are unable to tell by how 'much' of a factor are we interested in the object. In other words, can we find a number for someone's interest? In this article, we propose a method that uses the principle of Bayesian Inference to tackle this issue. We formulate the "interest estimation problem" as a state estimation problem to deduce interest (in any object) indirectly from user activity. Activity caused by interest is computed through a subjective-objective weighted approach, then using indirect inference rules, we provide numerical estimates of interest. To do that, we model the dynamics of interest through the Ornstein-Uhlenbeck process. To further enhance the base performance, we draw inspiration from Stochastic Volatility models from Finance. Subsequently, drawing upon a self-adapting transfer function, we provide an avant-garde statistical procedure to model the transformation of interest into activity. The individual contributions are then combined and a solution is provided via Particle filters. Validation of the method is done in two ways. (1) Experimentation is performed on real datasets. Through numerical investigation we have found that the method shows good performance. (2) We implement the framework as a Web application and deploy it on an Enterprise Service Bus. The framework has been successfully hosted on a Cloud based Virtualized testbed consisting of several Virtual Machines constructed over XENServer as the underlying hypervisor. Through this experimental setup, we show the efficacy of the proposed algorithm in estimating interest, at much the same time, we demonstrate the viability of the method in practical cloud based deployment scenarios. (C) 2018 Elsevier B.V. All rights reserved.