CHSPAM: um modelo multi-domínio para acompanhamento de padrões em históricos de contextos

Abstract

Ubiquitous computing aims to make tasks that depend on computing, transparent to users, thus, providing resources and services anytime and anywhere. One of the key factors to the development this type of application is the matter of Context Awareness, which enables an application to adjust its operation as the situation in which the user is. Thus, several authors have presented formal definitions of what is a context and how to represent it. With the development of systems that use Context information previously stored, databases have emerged that store Historical Contexts captured over time. Many researchers have studied different ways to analyzes this data. This paper addresses a specific type of data analysis in historical contexts, which is the discovery and monitoring of patterns in Context Histories. For this purpose, a model called CHSPAM (Context History Pattern Monitoring) is proposed, that allows the discovery of patterns in Context History databases and keeps track of these patterns to monitor their evolution over the time. Ubiquitous computing aims aim to integrate information technology perfectly into people's daily lives, so that people are aided by technological resources in the real world, proactively, while performing daily activities. One of the fundamental aspects for the development of this type of application is the issue of Context Awareness, which allows an application to adapt its operation according to the context in which the user is. With the development of systems that use information from previously stored contexts, databases have emerged that store captured Context Histories over time. Many researchers have studied different ways to perform analyzes on these data. This work addresses a specific type of data analysis in context histories, which is the search for sequential patterns. With this purpose, a model called CHSPAM (Context History Pattern Monitoring) is proposed that allows the discovery of sequential patterns in Context Historical databases, making use of existing data mining techniques. The main contributions of this work are the use of a generic representation for the storage of contexts allowing its application in multiple domains. Another contribution is that the model monitors the patterns discovered over time, storing history pattern evolution. A prototype of the model was implemented, and from it three experiments were carried out for its validation. The first experiment was used to evaluate the functionalities and services offered by CHSPAM and was based on simulated data. In the second experiment, the model was used in a prediction application and the use of monitored sequential patterns provided accuracy improvement on predictions when compared to the use of common patterns. Finally, in the third experiment, CHSPAM was used as a component of a learning object recommendation application and the application was able to recommend objects related to students’ interests based on monitored sequential patterns extracted from users’ session history.