Personalized task recommendation considering variant factors in internet of things environments

Abstract

In Internet of Things (IoT) environments, everyday consumer electronics become smart objects equipped with processors and memory. Because these smart objects are connected to each other, they are always accessible through the Internet. In these environments, when users want to perform their desired tasks according to their needs, it is difficult for the users to choose the most appropriate task variants that utilize the IoT environment in an efficient manner. Therefore, in this thesis, to deal with this problem, a collaborative filtering based task recommendation method is proposed to predict users’ needs and recommend most appropriate task variants that meet the users’ needs. Collaborative filtering is a method that utilizes users’ feedback information on items that they consumed or purchased to predict the users’ preferences and to recommend new items to them based on the predicted preferences. However, unlike traditional recommender systems, it is essential to consider task variants when recommending tasks in highly dynamic IoT environments. Specifically, rather than considering user ratings and/or purchase histories, taking into account variant factors such as the contextual factor, device-configurations, and environmental-effects is crucial to deal with the dynamic characteristics of IoT environments. Therefore, a task variant model is developed to represent various relationships between users and tasks, and the variant factors that affect the selection of task variants. Then, appropriate task variants are recommended to users by using the stochastic gradient descent algorithm which is one of the popular collaborative filtering methods. In addition, the genetic algorithm is used to find the optimal combination of variant factors that need to be considered to choose the task variants. To show the effectiveness of the proposed approach, an experiment has been conducted by using the datasets that are collected from practical IoT testbed environments, where there are various smart devices installed. The results show that the proposed task recommendation approach considering task variants is effective in terms of the accuracy and efficiency.