Vol. 29 · No. 4

Pervasive application of data mining technology is very important in analytical CRM software development when the distributed data warehouse is constructed. We propose a multi-factor customer classification evaluation model CLV/CL/CC which comprehensively considers customer lifetime value, customer loyalty and customer credit. It classifies clients with synthetic data mining algorithms. In this paper, we present an extended Bayes model which substitutes the primary attribute group with a new attribute group to improve the classification quality of naive Bayes.

IEEE 802.16 defines perfect QoS (Quality of Service) guarantee mechanism in PMP (Point to Multi-Point) mode and denotes mesh mode cannot provide this capability. To resolve this problem of WiMAX Mesh network, a novel QoS guarantee mechanism including protocol process and minislot allocation algorithm is proposed. This mechanism uses existing service classes in original standard. Protocol processes that manage dynamic service flow are defined. WiMAX MAC layer is re-designed to support service classification in mesh mode. Using extended distributed scheduling messages, the delivery method of dynamic service management messages in WiMAX mesh networks is implemented. Minislot allocation algorithm is given to support data scheduling of various services. Service establishment time is analyzed in this paper. The simulation result shows that the mechanism can provide QoS in WiMAX mesh networks effectively.

Today's information systems must operate effectively within open and dynamic environments. This challenge becomes a necessity for crisis management systems. In emergency contexts, in fact, a large number of actors need to collaborate and coordinate in the disaster scenes by exchanging and reporting information with each other and with the people in the control room. In such open settings, coordination technologies play a crucial role in supporting mobile agents located in areas prone to sudden changes with adaptive and flexible interaction patterns. Research efforts in different areas are converging to devise suitable mechanisms for process coordination: specifically, current results on service-oriented computing and multi-agent systems are being integrated to enable dynamic interaction among autonomous components in large, open systems. This work focuses on the exploitation and evaluation of the OpenKnowledge framework to support different information-gathering patterns in emergency contexts. The OpenKnowledge (OK) system has been adopted to model and simulate possible emergency plans. The Lightweight Coordination Calculus (LCC) is used to specify interaction models, which are published, discovered and executed by the OK distributed infrastructure in order to simulate peer interactions. A simulation environment fully integrated with the OK system has been developed to: (1) evaluate whether such infrastructure is able to support different models of information-sharing, e.g., centralized and decentralized patterns of interaction; (2) investigate under which conditions the OK paradigm, exploited in its decentralized nature, can improve the performance of more conventional centralized approaches. Preliminary results show the capability of the OK system in supporting the two afore-mentioned patterns and, under ideal assumptions, a comparable performance in both cases.

With the development of computer science, cognitive science and psychology, a new paradigm, affective learning, has emerged into e-learning domain. Although scientists and researchers have achieved fruitful outcomes in exploring the ways of detecting and understanding learners affect, e.g. eyes motion, facial expression etc., it sounds still necessary to deepen the recognition of learners affect in learning procedure with innovative methodologies. Our research focused on using bio-signals based methodology to explore learner's affect and the study was primarily made on Electroencephalography (EEG). After the EEG signals were collected from EEG equipment, we tidied the EEG data with signal processing algorithms and then extracted some features. We applied k-Nearest-Neighbor classifier and Naive Bayes classifier to these features to find out a combination, which may mostly contribute to reflect learners' affect, for example, Attention. In the classification algorithm, we presented a different way of using the Self-Assessment Manikin (SAM) model to classify and analyze learners attention, although the SAM was normally used for classifying emotions, for example, happiness etc. For the purpose of evaluating our findings, we also developed an affective learning prototype based on university e-learning web site. A real time EEG feedback window and an attention report were integrated into the system. The result of the experiment was encouraging and further discussion was also included in this paper.

This paper experimentally studies approaches to the problem of describing heterogeneous information sources in distributed environments. In particular, we consider a scenario where a large number of end users can share and retrieve text documents over a peer-to-peer network. Descriptions (or profiles) of peers are useful in a number of applications, such as query routing, overlay construction and expert search. The approach proposed in this paper introduces a new learning method that boosts the weight of query terms in a peer's profile when the peer provides useful documents w.r.t. a given query. Experimental results show high potential for this method. Therefore, various extensions are proposed that involve more user interaction.

Freehand drawing is an easy and intuitive method for thinking input and output. In sketch based interface, there lack support for natural sketching with drawing cues, like overlapping, overlooping, hatching, etc. which happen frequently in physical pen and paper. In this paper, we analyze some characters of drawing cues in sketch based interface and describe the different types of sketching primitives. An improved sketch information model is given and the idea is to present and record design thinking during freehand drawing process with individuality and diversification. The interaction model based on context is developed which can guide and help new sketch-based interface development. New applications with different context contents can be easily derived from it and developed further. Our approach can support the tasks that are common across applications, requiring the designer to only provide support for the application-specific tasks. It is capable of and applicable for modeling various sketching interfaces and applications. Finally, we illustrate the general operations of the system by examples in different applications.

The increasing availability of game based technologies together with advances in Human-Computer Interaction (HCI) and usability engineering provides new challenges and opportunities to virtual environments in the context of e-Teaching. Consequently, an evident trend is to offer learners with the equivalent of practical learning experiences, whilst supporting creativity for both teachers and learners. Current market surveys showed surprisingly that the Wii remote controller (Wiimote) is more widely spread than standard PCs and is the most used computer input device worldwide, which given its collection of sensors, accelerometers and bluetooth technology, makes it of great interest for HCI experiments in e-Learning/e-Teaching. In this paper we discuss the importance of gestures for teaching and describe the design and development of a low-cost demonstrator kit based on Wiimote enhancing the quality of the lecturing with gestures.

To learn a Web browsing behavior model, a large amount of labelled data must be available beforehand. However, very often the labelled data is limited and expensive to generate, since labelling typically requires human expertise. It could be even worse when we want to train personalized model. This paper proposes to train a personalized Web browsing behavior model by semi-supervised learning. The preliminary result based on the data from our user study shows that semi-supervised learning performs fairly well even though there are very few labelled data we can obtain from the specific user.

Ubiquitous services have gained increasing attention in the area of mobile communication aiming to allow service access anywhere, anytime and anyhow while keeping complexity to a minimum for both users and service providers. Ubiquitous environment features a wide range and an increasing number of access devices and network technologies. Context-aware content/service adaptation is deemed necessary to ensure best user experience. We developed an Adaptation Management Framework (AMF) Web Service which manages the complexity of dynamic and autonomous content adaptation and serves as an invisible enabler for ubiquitous service delivery. It remains challenging to manage the tasks involved in the communication between the AMF Web Service and the user's environment, typically represented by various types of intelligent agents. This work presents a middleware which manages those tasks and serves not only as a protocol gateway, but also as a message translator, a service broker, a complexity shield etc., between AMF Web Services and User Agents.

Target tracking is an important cooperative sensing application of wireless sensor networks. In these networks energy, computing power and communication bandwidth are scarce. In this paper, we consider a randomly deployed sensor network with sensors acting as a set of distributed datasets. Each dataset is assumed to have its local temporal dataset, along with spatial data and the geographical coordinates of a given object. An approach towards mines global temporal patterns from these datasets and to discovers nonlinear trajectories of a moving object is proposed. It is tested in a simulation environment and compared with straightforward method. The results of the experiments clearly show the benefits of the new approach in terms of energy consumption.

Accurate pupil features extraction is a key step for iris recognition. In this paper, we propose a new algorithm to extract pupil features precisely within gray level iris images. The angular integral projection function (AIPF) is developed as a general function to perform integral projection along angular directions, both the well known vertical and horizontal integral projection functions can be viewed as special cases of AIPF. Another implementation for AIPF based on localized Radon transform is also presented. First, the approximate position of pupil center is detected. Then, a set of pupil's radial boundary points are detected using AIPF. Finally, a circle to the detected boundary points is fitted. Experimental results on 2 655 iris images from CASIA V3.0 show high accuracy with rapid execution time.

In this paper we study a modified work function algorithm (WFA) for solving the on-line k-server problem. Our modification is based on a moving window, i.e. on an approximate work function that takes into account only a fixed number of most recent on-line requests. We give a precise specification of the modified WFA, investigate its competitiveness, and explain how it can be implemented efficiently by network flows. We also present experiments that measure the performance and computational complexity of the implemented algorithm. The results of the paper can be summarized as follows: the modified WFA is not competitive, but according to the experiments it still provides almost the same quality of serving as the original WFA while running much faster.