Recently, mobile context inference becomes an important issue. Bayesian probabilistic model is one of the most popular probabilistic approaches for context inference. It efficiently represents and exploits the conditional independence of propositions. However, there are some limitations for probabilistic context inference in mobile devices. Mobile devices relatively lacks of sufficient memory. In this paper, we present a novel method for efficient Bayesian inference on a mobile phone. In order to overcome the constraints of the mobile environment, the method uses two-layered Bayesian networks with tree structure. In contrast to the conventional techniques, this method attempts to use probabilistic models with fixed tree structures and intermediate nodes. It can reduce the inference time by eliminating junction tree creation. To evaluate the performance of this method, an experiment is conducted with data collected over a month. The result shows the efficiency and effectiveness of the proposed method.
One of the important issues in recent mobile computing is context-awareness. In next-generation mobile systems, novel solutions for user-centric service are crucial to provide personalized views of only the services of potential interest (Bellavista, Corradi, Montanari, & Toninelli, 2006). The service personalization should be based on user context and environment conditions. It requires high-level context inference from raw sensor data. Some researchers have tried to gather raw data from a mobile device and infer high-level context (Gemmell et al., 2006, Korpipaa et al., 2003, Raento et al., 2005 and Siewiorek et al., 2003).
Models for context inference generate the desired target information (the high-level context) from existing information (low-level context). This leads to a kind of classification problem which determines “class” (state of the high-level context) from the combination of low-level information. Methods for context inference should be efficient, sound and complete according to Perttunen, Riekki, and Lassila (2009) and cope with imperfection of data (Bikakis, Patkos, Antoniou, & Plexousakis, 2008). Angermann, Robertson, and Strang (2005) specified the requirements further, suggesting means for fusion of several input sources and handling of possibly contradicting measurements, expressive modeling of situations, and adaptation to the needs of large-scale pervasive systems, i.e. distributed processing, personalization and adaptability to the system’s dynamics.
Probabilistic approach for context inference deals with the uncertainty in measurements and propositions about various situations in real world by using probabilities that are encoding degrees of belief with values between 0 and 1. Bayesian network is one of the most popular probabilistic approaches. It efficiently represents and exploits the conditional independence of propositions. This model can be learnt from existing data, but also set manually by human experts as they can be interpreted easily. Also combinations of learnt and set probability models are possible. Some researchers used Bayesian techniques to recognize activities of daily life (Korpipää, Koskinen, Peltola, Mäkelä, & Seppänen, 2003) from sensor data.
However, there are some limitations for probabilistic context inference in mobile devices. Mobile devices contain relatively insufficient memory capacity, lower CPU power (data-processing speed), and limited battery when compared to desktop PCs. In addition, they have to operate in the changeable real world, which means that they require more active and effective adaptation functions (Dourish, 2004).
This paper adopts Bayesian probabilistic models to efficiently manage various uncertainties that can occur when working with mobile environments, including real-world irregularities, like uncertain causal factors. The proposed method presents a method to infer high-level context using a two-layered Bayesian network using mobile data. It constructs Bayesian network models with tree-like structures from low-level context. The models work more efficiently in mobile environments than standard Bayesian network models. Moreover, the models can be linked with hierarchical structure to provide more accurate and broad-coverage context. In order to show the feasibility of the proposed model, it was applied to several experiments using mobile log data collected from a smartphone for a month in the real world.
The rest of the paper is organized as follows. Section 2 introduces related work regarding mobile context inference. Section 3 describes Bayesian network design using tree structure and intermediate nodes to reduce the cost of context inference. In Section 4 we introduce a two-layered Bayesian model to infer high-level context. Section 5 shows experiments and the results to evaluate the approach. Finally, Section 6 summarizes this paper and presents future works.
In this paper, we have proposed two-layered Bayesian network models for mobile context inference. To infer a user’s activity efficiently, the proposed method uses Bayesian probabilistic models with tree structures and eliminates junction tree algorithm for inference process. It also supports two-layered inference to cover broader activities. The proposed method provides suitable inference process in the mobile environment using a small amount of inference time. The performance of the inference is evaluated with the data for undergraduate students in a real life.
In future research, the comparison of performance will be extended to other classification methods such as neural network, decision trees and support vector machines. In addition, the types of context will be expanded to various kinds of sensory information such as acceleration, proximity, magnetic fields, and intensity of light. It is also necessary to consider more effective construction of layered structure for layered Bayesian networks.
