Ebook: Advances in Intelligent IT

In the great digital era, we are witnessing many rapid scientific and technological developments in human-centered, seamless computing environments, interfaces, devices, and systems with applications ranging from business and communication to entertainment and learning. These developments are collectively best characterized as Active Media Technology (AMT), a new area of intelligent information technology and computer science that emphasizes the proactive, seamless roles of interfaces and systems as well as new media in all aspects of digital life. An AMT based computer system offers services that enable the rapid design, implementation, deploying and support of customized solutions.

The first International Conference on Active Media Technology (AMT01) was held in Hong Kong in 2001, the second International Conference on Active Media Technology (AMT03) was held in Chongqing, China in May 29–31 of 2004, and the third International Conference on Active Media Technology (AMT05) was held in Kagawa, Japan in May 2005. The 4^th International Conference on Active Media Technology (AMT06) follows the success of AMT01, AMT03 and AMT05.

AMT06 is the leading International Conference focusing on Active Media Technology. It aims to bring together researchers from diverse areas, such as Web intelligence, data mining, intelligent agents, smart information use, networking and intelligent interface. It also encourages collaborative research in these areas to provide best services for enabling the rapid design, implementation, deploying and support of customized solutions.

The conference includes the following topics:

• Active Computer Systems and Intelligent Interfaces

• Adaptive Web Systems and Information Foraging Agents

• Web mining, Wisdom Web and Web Intelligence

• E-Commerce and Web Services

• Data Mining, Ontology Mining and Data Reasoning

• Network, Mobile and Wireless Security

• Entertainment and Social Applications of Active Media

• Agent-Based Software Engineering and Multi-Agent Systems

• Digital City and Digital Interactivity

• Machine Learning and Human-Centred Robotics

• Multi-Modal Processing, Detection, Recognition, and Expression Analysis

• Personalized, Pervasive, and Ubiquitous Systems and their Interfaces

• Smart Digital Media

• Evaluation of Active Media and AMT Based Systems

AMT06 is sponsored by the IEEE Systems, Man, and Cybernetics Society and Queensland University of Technology. It attracted 123 submissions from 19 countries and regions: Algeria, Australia, China, Canada, England, Finland, France, Hong Kong, India, Japan, Korea, New Zealand, Pakistan, Poland, Republic of Korea, Taiwan, United Arab Emirates, United Kingdom, and United States of America. The review process was rigorous. Each paper was reviewed by two reviewers at least, and most of them reviewed by three reviewers.

The Program Committee accepted 39 regular papers (the approximate acceptance rate is 32%), 33 short papers (the approximate acceptance rate is 39%) and 9 industry/demonstration papers.

We would like to thank the members of Program Committee and Organization Committee and reviewers who contributed to the success of this conference.

Yuefeng Li, Mark Looi and Ning Zhong, 17 March 2006

Many online applications of machine learning require fast classification and hence utilize efficient classifiers such as naïve Bayes. However, outside periods of peak computational load, additional computational resources will often be available. Anytime classification can use whatever computational resources may be available at classification time to improve the accuracy of the classifications made.

The aims and objectives of data mining is to discover actionable knowledge of main interest to real user needs, which is one of Grand Challenges in KDD. Most extant data mining is a data-driven trial-an-error process. Patterns discovered via predefined models in the above process are often of limited interest to constraint-based real business. In order to work out patterns really interesting and actionable to the real world, pattern discovery is more likely to be a domain-driven human-machine-cooperated process. This talk proposes a practical data mining methodology named “domain-driven data mining”. The main ideas include a Domain-Driven In-Depth Pattern Discovery framework (DDID-PD), constraint-based mining, in-depth mining, human-cooperated mining and loop-closed mining. Guided by this methodology, we demonstrate some of our work in identifying useful correlations in real stock markets, for instance, discovering optimal trading rules from the existing rule classes, and mining trading rule-stock correlations in stock exchange data. The results have attracted strong interest from both traders and researchers in stock markets. It has shown that the methodology is potential for guiding deep mining of patterns interesting to real business.

A tool designed to serve for the transformation of Web service description is described. The fundamentals of service –oriented architecture are heavily relied on the eXtensible Markup Language (XML) to describe the interfaces of web services and the access protocol. The automation process to transform the input of user requests into XML-based representation can facilitate the service composition. We proposed a solution tool to collect user requirements and transform them into XML-based document automatically. The tool is illustrated through a domain-specific task flow.

Technology is advancing at a rapid pace, automating many everyday chores in the process, changing the way we perform work and providing various forms of entertainment. Makers of technology, however, often do not consider the needs of the disabled in their design of products by, for example, providing some alternative means of interaction with their devices. The use of computers presents a challenge to many disabled users who are not able to see graphical user interfaces, use a mouse or keyboard or otherwise interact with standard computers. This paper introduces a head-tracker based on the use of a modified Lucas-Kanade optical-flow algorithm for tracking head movements, eliminating the need to locate and track specific facial features. The implementation presents an alternative to the traditional mouse input device.

Most of the existing recommender systems nowadays operate in a single organizational base, and very often they do not have sufficient resources to be used in order to generate quality recommendations. Therefore, it would be beneficial if recommender systems of different organizations can cooperate together sharing their resources and recommendations. In this paper, we propose a preliminary design of a distributed recommender system that consists of multiple recommender systems from different organizations. Moreover, a peer selection algorithm is also presented that allows a recommender system peer to select a set of other peers to cooperate with. The proposed selection mechanism not only ensures a high degree of user satisfaction to the generated recommendation, it also makes sure that every peer has been fairly treated and studied. The paper also further points out how the proposed distributed recommender system and the peer selection algorithm can provide a solution to the problem of resource lacking (e.g. cold start problem) and also enables recommender systems to provide recommendations with better novelty and quality to users.

There has been a move to acetate web context with explicit meaning so that machines will be able to make better use of it and thus better able to assist web users, leading to a Semantic Web. To date, the possibilities for delivering personalized digital media experiences have been limited however more and more companies are entering the marketplace to offer personal digital media delivery. These companies provide products and services that offer an instant solution for digital media deployments. This paper presents a prototype which demonstrates the power of the semantic web where users can connect to an existing database of media files (e.g. political interviews, new snippets, football manager pre-match interviews etc) and retrieve a selection of media clips. In the case of a football manager giving an interview before a game, it would be possible to also associate his interview with that of another premiership manager. Therefore whenever a user requests an interview clip, the other clip could also be offered to this user.

In traditional Information Retrieval (IR), user profiles are often represented by keyword/concepts space vectors or by some predefined categories. Unfortunately, this data is often inadequately or incompletely interpreted. Ontology-based user profile is another newer approach. This method is able to provide richer semantic information to facilitate information retrieval processes. It has become an important means for semantic-based information search and retrieval. Some ontology-based user profile models have been developed over the past few years. With the increasing usage of this method, it raises the issues of effective relevance measurement for the evaluation of ontologies. In practice, it is crucial to find a good relevance assessment algorithm for measuring the quality of ontologies. To represent user profile by relevant topic ontology, this paper presents a new method capable of measuring the user profile more objectively and hence has great potential to enhance the IR processes.

This paper focuses on information extraction from one site rather than from one page. A new directed-acyclic graph based representation method is introduced for representing link structures on the Web sites. A rule based language is developed for writing extraction rules. The approach is examined and tested on 42 Web sites in six different domains. The results suggest that the system can successfully extract most relevant information from Web sites in different domains.

E-finance is rapidly transforming and evolving toward more dynamic, flexible and automated solutions. This paper describes a conceptual model with dynamic multi-level workflows corresponding to a multi-layer Grid architecture, for multi-aspect analysis in building an e-finance portal on the Wisdom Web. The application and research demonstrate that mining-grid centric three-layer Grid architecture is effective for developing intelligent risk management and decision making financial systems.

XML has gained popularity for information representation, exchange and retrieval. As the XML material becomes more abundant, the ability to gain knowledge from XML sources decreases due to their heterogeneity and structural irregularity. The use of data mining techniques becomes essential to improve XML document handling. This paper discusses the capabilities and the process of applying data mining techniques in XML sources.

Although today's web search engines are very powerful, they still fail to provide intuitively relevant results for many types of queries, especially ones that are vaguely-formed in the user's own mind. We argue that associations between terms in a search query can reveal the underlying information needs in the users' mind and should be taken into account in search. We propose a multi-faceted approach to detect and exploit such associations. The CORDER method measures the association strength between query terms, and queries consisting of terms having low association strength with each other are seen as ‘vague queries’. For a vague query, we use WordNet to find related terms of the query terms to compose extended queries, relying especially on the role of least common subsumers (LCS). We use relation strength between terms calculated by the CORDER method to refine these extended queries. Finally, we use the Hyperspace Analogue to Language (HAL) model and information flow (IF) method to expand these refined queries. Our initial experimental results on a corpus of 500 books from Amazon shows that our approach can find the right books for users given authentic vague queries, even in those cases where Google and Amazon's own book search fail.

We present a general fault-tolerate multiparty contract signing protocol that uses verifiable signature sharing technique, the protocol works in the asynchronous network environment and realizes multiparty contract signing within a fixed expected number of computational rounds. The correctness of the protocol is proved in theory.

The proliferation of wireless network technologies has led to an explosion in the deployment of WiFi communications solutions. Consequently this propagation of wireless systems coupled with the frequency and familiarity of users to these systems has led to an increase in user expectation, whereby more and more mobile users are demanding the same Quality of Service (QoS) that they were accustomed to as fixed wired network users. As a result of the widespread deployment of Wi-Fi hotspots together with the expected increase in user demands for unlimited bandwidth and unrestrained network access, pressure has been placed on these wireless networks to respond to these user expectations. Most notable in this area are the complexities involved in providing the mobile user with seamless network connectivity during the traversal of a mobile cellular based network infrastructure. Fundamental to these issues is the latency and packet loss that occurs during the handoff process when a user moves between cells when traversing the wireless network as a handoff is required in order to maintain connectivity to the network, while at the same time ensuring minimum disruption to ongoing sessions. This paper presents a handoff algorithm for streaming media on mobile networks that will anticipate the handover procedure that occurs when a mobile device roams from one cell in the network to another. It proceeds to buffer packets from the anticipated ‘new’ access point and upon successful entry into the correctly predicted target cell will promptly ‘pick up’ the stream with minimal losses.

Peer-to-Peer (P2P) architecture is one of the most interesting topics in distributed systems and other related areas (e.g., AI, Database, etc). Basic P2P applications have only implemented limited aspects of a real P2P environment. Meanwhile the fast growing technology, autonomous agents, appear to be a good candidate for most of the complex and dynamic problems. This paper proposes a flexible design for the creation of agent auctions in a distributed environment. Interactions between agents occur in a P2P communication protocol, reducing the role of the centralized auction process to an auction initiator, and to inform agents when a general equilibrium is reached. Agents and auctions are designed using the object-process methodology (OPM) which represents a comprehensive approach to system evolution that incorporates the static-structural and dynamic-procedural aspects of a system into a single unifying model. OPM includes a clear and concise set of symbols that form a language enabling the expression of the system's building blocks and how they relate to each other.

This paper presents a rough set model for constraint-based multi-dimensional association rule mining. It first overviews the progress in constraint-based multi-dimensional association rule mining. It then applies the constraints on the rough set model. To set up a decision table, it adopts the user voting and the thresholds on condition granules and decision granules. Finally it employs the extended random sets to generate interesting rules. It shows that this rough set model will effectively improve the quality of association rule mining by reducing the attributes greatly in the vertical direction and clustering the records clearly in the horizontal direction. To describe the association among the attributes, it constructs an ontology and presents a new concept of an association table. The construction of a tuple in an association table indicates the relationship among different levels on the ontology towards decision support.

A Kernel-Based Nonparametric Multiple imputation method is proposed under MAR (Missing at Random) and MCAR (Missing Completely at Random) missing mechanisms in nonparametric regression settings. We experimentally evaluate our approach, and demonstrate that our imputation performs better than the well-known NORM algorithm.

As the first stage for discovering association rules, frequent itemsets mining is an important challenging task for large databases. Sampling provides an efficient way to get approximating answers in much shorter time. Based on the characteristics of frequent itemsets counting, a new bound for sampling is proposed, with which less samples are necessary to achieve the required accuracy and the efficiency is much improved over traditional Chernoff bounds.

One common source of error in data is the existence of missing value fields. Imputation method has been a widely used technique in preprocessing phase of data mining, in which missing values are replaced by some estimated values. Previous work is trying to seek the “original” values according to specific criteria, such as statistics measure. However, in domain of cost-sensitive learning, minimal overall cost is the most important issue, i.e. a value which can minimize total cost is prefer than the “best” value upon common sense. For example, in medical domains, some data fields usually are left as absent and known information is enough for a decision. In this paper, we proposed a new method to study the problem of “missing or absent values?” in the domain cost-sensitive learning. Experiment results show some improvements with distinguished missing and absent data in cost-sensitive decision tree.

Conventional information displays provide us some images. However these images are generally two-dimensional(2D). If the images are displayed in the three-dimensional (3D) space, these images are in the air at a distance from a desk and users can directly touch and interact them such that kids can make play clay. In addition, many applications for 3D imaging can be proposed. The authors have researched the 3D displays and applications. We developed new applications and a prototype interactive tabletop holographic display system. These systems consist of the object recognition system and the spatial imaging system. In this paper, we describe the recognition system using RFID tags and a tabletop display system using a conventional CRT display and a holographic technology.

We present an approach using story scripts and action descriptions in a form similar to the content description of storyboards to predict specific personality and emotional states. By constructing a hierarchical fuzzy rule-based system we facilitate the personality and emotion control of the body language of dynamic story characters. Our ultimate goal is to facilitate the high-level control of synthetic characters.

Electrical stimulation of the human visual system can result in the perception of blobs of light, known as phosphenes. Artificial Human Vision (AHV or visual prosthesis) systems use this method to provide a visual substitute for the blind. This paper reports on our experiments involving normally sighted participants using a portable AHV simulation. A Virtual Reality Head Mounted Display is used to display the phosphene simulation. Custom software converts captured images from a head mounted USB camera to a DirectX based phosphene simulation. The effects of frame rate (1, 2 and 4 FPS) and phosphene spatial resolution (16x12 and 32x24) on participant Percentage of Preferred Walking Speed (PPWS) and mobility errors were assessed during repeated trials on an artificial indoor mobility course. Results indicate that spatial resolution is a significant factor in reducing contact with obstacles and following a path without veering, however the phosphene display frame rate is a better predictor of a person's preferred walking speed. These findings support the development of an adaptive display which could provide a faster display with reduced spatial resolution when a person is walking comfortably and a slower display with higher resolution when a person has stopped moving.