Detecting Interaction Period and Activity In End Host Devices

Abstract

The Internet has become an important part of human lives. Understanding how users spend their time and resources has become an important field of research for network management and sociology (study of users' online behaviours). For example, if a user is involved in a video call on Skype, while upgrading their operating system, the system could improve user experience by giving priority to the Skype call. In general, network management tools would benefit from identifying the traffic that corresponds to the application the user is interacting with. Furthermore, the identification of user activity, will shed light on user online behaviour, allowing us to build their profiles and use them later in applications such as advertising, content recommendation, and optimisation of application/online services. The first goal of this work is to characterise interaction period. By interaction period we refer to a period of time during which a user is actively interacting with one or more applications on a single device. To best study interaction period we need data from users end devices. To get this we have HostView data, which is good because it collects longitudinal application usage for 47 users. HostView data contains traffic and network performance statistics, application level context (for traffic flows), system performance and environmental data and user reports on network performance. But having this data, does not mean that from it we can obtain directly a report of a given interaction period. Thus our second goal is to develop a heuristic method in order to obtain such reports. Human interactions with computers are quite diverse. When they use their computer they have more than one application running and they use different peripheral devices (mouse, keyboard, touch-screen, light pen and so on). So, our main challenge is to infer behaviour while the user is interacting with the application. This knowledge will allow us to answer many questions like: (a) How many applications users have simultaneously running? (b) How many applications are generating networks traffic consuming resources? (c) How can we separate and determine online from general activity. The main idea of this work is to build a heuristic method to identify interaction period based on time series of keyboard, mouse click events and the knowledge of the foreground application. The study will establish different thresholds and period of times, which will allow the recognition and determination of the interaction period. Inside of this HostView data we have the foreground application and mouse-click and keyboard events. Our heuristic method use this information to identify and establish interaction periods. To define which is the application that user is interacting with, first we have to define different thresholds referring to when interaction period starts and finishes or how many applications are upon an interaction period running at the same time. After that we will able to determine the online applications used at the same time, their clicks-key intervals and the duration of their interaction. We apply our heuristics in six months of date of twelve users to identify interaction period over time. Our analysis shows that users normally have more than eight applications running at the same time while only three of them are over online interaction. Also we have found that when users interact with online applications, they only interact with one or at most two application, whereas running on background they might have up to extra six applications consuming resources at the same moment.