Develop a computational trust prototype for the Semantic Web

1 downloads 0 Views 342KB Size Report
Develop a computational trust prototype for the Semantic Web. Yu Zhang. 1. , Huajun Chen. 1. , Zhaohui Wu. 1. , Xiaoqing Zheng. 1. 1Grid Computing Lab ...
Develop a computational trust prototype for the Semantic Web

1

Yu Zhang1, Huajun Chen1, Zhaohui Wu1, Xiaoqing Zheng1 Grid Computing Lab, College of Computer Science, Zhejiang University, Hangzhou, China E-mail: {yzh, huajunsir, wzh, zxqingcn}@zju.edu.cn Abstract

Nowadays, more and more research has focused on the trust management of the Semantic Web. Some mature technologies from many fields such as multiagent systems, social network analysis and artificial intelligence have been introduced to this field. However, some unique characteristics of the Semantic Web are more or less ignored. In this paper, we try to explore the semantic aspects of the trust management. We present DartTrust, a trust prototype for the semantic web in the application context of Traditional Chinese Medicine (TCM). We introduce RCSW — a Reputation-Chain trust model for the Semantic Web, which incorporates pairwise trust factor and reliable factor and constructs an edge-weighted graph to calculate trust ratings. We define a small trust ontology to illustrate the trust relationships. The trust ontology is described in OWL that can be understood and processed by machines. Therefore, DartTrust can provide reasoning functions and facilitate trust management with little human effort. As a demonstration of the prototype, we present its implementation on Traditional Chinese Medicine (TCM) at the end of this paper.

1. Introduction In the future, the Semantic Web (SW) should manage to support smooth interactions with a large variety of independent resources, objects and applications, running on heterogeneous platforms and distributed networks. In our common life, we place our trust in people and the services that those people provide. Without trust, our life would rapidly descend into chaos and we would be filled with all sorts of fearful possibilities. People will also face the same situation if there is no trust management on the SW. As a large and open system, the SW possesses the following features: (1) Information sources can freely join and leave at any time and anyone can be an information provider or consume anyone else’s information; (2) The quality of interactions may vary

dramatically from time to time; and (3) No central authority can control all the objects in the network. Therefore, without trust management, people are unable to winnow truth from falsehood so the utilization ratio of the SW is rather small. In this paper, we introduce a new trust prototype — DartTrust for the Semantic Web in the application context of Traditional Chinese Medicine (TCM). TCM is an efficient traditional medical therapy that embodies holistic knowledge with thousands of years’ clinical practice. Nowadays, there are many requirements in this area: (1) patients who are ignorant of medicine hope to have a personal online “family doctor”; (2) doctors and researchers need a convenient platform to learn from each other; (3), experts hope to have a place to publish their opinions, remarks and make a lead to other doctors especially in some urgent cases such as SARS in 2003; and (4), government wants to propagandize and maintain Chinese unique treasure — TCM. TCM is a highly professional field. DartTrust aims to provide medical information that is of vital importance to human’s health and life. Therefore, it is important and necessary to bring in trust management to this system. The rest of the paper is organized as follows. Section 2 presents the related work in this area. Section 3 shows the overview of the trust model. Section 4 introduces the basic definitions and mechanisms of RCSW. Section 5 illustrates the reasoning function of DartTrust. Section 6 presents the implementation of the prototype. Section 7 concludes this paper and outlines the future work.

2. Related Work There is a great deal of research in this field. SPORAS is a successful centralized trust model that provides a reputation mechanism for loosely connected online communities [8]. Amazon Auctions [2] and eBay [5] are also centralized reputation models that are widely used in our real life. Relying on a central agency to manage trust helps greatly reduce the workload of interactions. However, due to the sheer

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE

magnitude and diversity of information sources, it is virtually impossible to control the whole Semantic Web in a centralized fashion. A Wide range of disciplines in the study of social science is also applied to the trust management [7][10][15]. In [1], the authors provide and discuss a trust model that is grounded in real-world social trust characteristics and based on a reputation mechanism. SchemaWeb [15] defines a trust ontology that extends the large and popular FOAF [6] (Friend Of A Friend) Vocabulary. SchemaWeb infers trust over web-based social networks and applications. The project also provides instrumental tools for users to enter the trust network and calculate trust ratings. THELLIS [19] presents an approach to assessing information sources based on individual feedback about the sources. Users can annotate their analysis of information sources and express credibility and reliability of the sources explicitly or implicitly. Later, those measures are averaged and presented to the viewers. The shortcomings of THELLIS are that the values of trustworthiness are not personalized for each individual and it requires the users to reach an agreement on the credibility of the sources. FIRE [4] manages to produce a trust measure and an associated reliability measure in most situations. However, it is a static parametric model and all its parameters are set by users to suit a particular application domain. Therefore, FIRE is limited and cannot adapt to the dynamic environment efficiently. In [17], a concrete computational trust model is developed, which takes into account both the direct and indirect interactions between agents. They adopt pessimistic strategies that restrict interactions with agents unless there is a reason to trust them. The system takes a high rank as evidence of reason for trust; hence in such a system many trustworthy agents may fail to be trusted. EigenTrust [16] algorithm computes global trust values as a function of local trust values in a peer-topeer network. It emphasizes on security problems such as betrayal and lie between peers. The algorithm computes a global trust value similar to PageRank without providing personalized trust information.

3. The Trust Model The current hypertext web is a set of pages clustered by hyperlinks that can only be understood by humans. The hyperlink between each pair of linked pages does not necessarily indicate a certain kind of relationship such as knows or trusts. Web sites owners construct their web pages according to their own interests and preferences, so they can link to every existing page as

they wish to. In other words, there may be no semantics between some web pages at all. In the future, the Semantic Web can be viewed as a collection of intelligent agents. Concepts on the SW are semantically and automatically linked. These agents should process the following features: 1. Agents can operate automatically according to predefined rules and respond timely to changes with little humans’ effort; 2. Agents can take advantage of the social trust network to exchange information and communicate with each other; 3. Agents should possess reasoning functions and learning abilities; 4. Agents are able to take the initiative to explore the environment and prepare for future interactions. With the introduction of ontology and RDF [14] (Resource Description Framework), meta-data of the distributed web are machine understandable, which allows agents to function automatically. Agents on the SW can exchange trust information and communicate with each other through the chain of the trust network. When an agent receives updated changes, it can automatically adjust corresponding trust ratings according to its individualized rules. Meanwhile, DartTrust can make good use of the reasoning and learning capabilities to take the initiative to find more reliable and suitable partners. Trust is not necessarily transitive. If A trusts B and B trusts C , you can’t certainly conclude that A trusts C . However, when we look deeper into trust phenomenon, we will find that the trust relationships between agents with similar preferences are partially transitive. For example, both A and B likes to listen to country music in the 20th century, therefore A ' s recommendations of good music are much more believable in B ' s opinion than a rock-and-roll lover. In this paper, we assume that given some predefined domain and context such as TCM, a user’s circle of acquaintances is more likely to share the same values as he or she does, at least they can understand them.

4. Basic definitions and mechanisms of RCSW The whole Semantic Web can be viewed as a directed Graph G . In this paper, we regard objects in OWL statement as vertices, and predicates that connect objects as edges. We use E to denote the set of edges and V to denote the set of vertices. Therefore, we can write G = (V , E ) . In order to present the model, we first introduce some basic definitions as follows.

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE

4.1 Trust

among v2 ' s ten sentences, only two of them are true,

Trust is used to define one’s relationships to others. In our daily life, we judge how to act based upon the trust we have in others. In this paper, the user chooses the most reliable service according to the trust rating of the service providers. Definition 1: Trust: If v1 , v2 ∈ V and (v1, v2 ) ∈ E ,

then Rv1 →v2 is 0.2. The range of R is [0,1] , where 0

then Tv1 →v2 denotes v1 ' s belief in v2 ' s competence to fulfill a task or provide a service. Trust is a complex and multidimensional concept, so in this paper we discuss trust with respect to a given domain — treatment in TCM field. For example, a patient trusts his doctor’s treatment and does exactly what the doctor says. In contrast, another doctor distrusts that doctor and will never adopts the same method in treating his own patients. The range of T is [0,1] which is divided into several subscales: Ɣ 0 : Absolute distrust; Ɣ (0, 0.2) : High distrust; Ɣ[0.2, 0.5) : Moderate distrust; Ɣ 0.5 : Neutral; Ɣ (0.5, 0.8] : Moderate trust; Ɣ (0.8,1) : High trust; Ɣ 1 : Absolute trust;

4.2 Reliable factor Agents cannot know all about the environment so it is very essential for them to communicate with each other and exchange trust information. Agents should take good advantage of the trust social network composed by their friends or neighbors to reduce the risks and losses of unsatisfied interactions. However, due to various reasons, such as different personal preferences, economic interests, deceit and betrayal, agents can’t fully trust others. Therefore, in this paper we use reliable factor to evaluate to which degree an agent should believe the information provided by its acquaintances. Definition 2: Reliable Factor: If v1 , v2 ∈ V and (v1, v2 ) ∈ E , then there is a Reliable Factor Rv1 →v2 that

denotes to which degree v1 believes in v2 ' s words or opinions. In other words, Rv1 →v2 means the probability that v1 thinks v2 tells the truth during the propagation of

trust information. For example, if in v1 ' s opinion,

means absolute unreliable and 1 means absolute reliable. Trust and Reliable factor have different meanings. Trust denotes to which degree that a consumer believes in a provider ’s competence to fulfill a task. It is the selection standard of service provider. Reliable factor denotes how credible an agent thinks of its neighbors or friends. The agent uses reliable factor to decide whose trust information to use and how to combine the information from various sources. The above trust information is gathered for the consumer agent to find the most suitable service provider.

4.3 Neighbor and Friends Within the community of agents on the SW, we define two kinds of acquaintances: Neighbor and Friends. Definition 3: Neighbor: If v1 , v2 ∈ V , (v1, v2 ) ∈ E and (v2, v1 ) ∉ E , v2 is v1 ' s neighbor, note as v1  v2 . Definition 4: Friends: If v1 , v2 ∈ V , (v1, v2 ) ∈ E , (v2, v1 ) ∈ E and Rv1 →v2 > 0.9  Rv2 → v1 > 0.9 , then v1 and v2 are friends, note as v1 ↔ v2 .

Figure 1. Neighbor and friends of RCSW On each edge between a pair of vertices, the number inside parenthesis denotes reliable factor and the number outside parenthesis denotes trust value. In Figure 1, two broken lines denote that A and G are friends to each other, and solid directed edges denote that vertices B , C , D and E are A ' s neighbors. M is not A ' s neighbor as the direction of the edge is the opposite . B , C , D and E are four channels that provide trust information to A . It is up to A to determine how to make use of the trust

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE

information. The trust value and reliable factor are labeled on each directed edge. For example, TA→ E = 0.31 and R A→ E = 0.73 .

4.4 Assign trust rating and reliable factor When a user first login the system, he can assign trust rating and reliable factor to others manually according to his preferences. His agent will record those values and analyze his preferences. In the future, the user’s agent can automatically assign trust values when it meets similar agents as before. Suppose that a consumer agent A wants to obtain a trusted service from a service provider. A may first consult its friends or neighbors about which is the most suitable one. If its friends or neighbors have had experiences with a trusted and good provider, they will reply to A directly, or else the trust network will expand out to gather more information. With the trust information gathered, A calculates referential trust value of potential provider using the algorithm in Section 4.5 through the trust network. After calculation, A can choose the provider with the highest trust rating.

number of paths between P and Q whose steps are less than MAX. Di denotes the number of edges from P to Q on the ith path. We call the set of Q’s friends or neighbors M. Therefore, Mi denotes Q’s friend or neighbor on the ith path. According to the above principles, we first calculate the weight of different paths:

1 wi =

Di n

1

1

i

(1)

∑D

The trust rating TP →Q is calculated as follows: n

TP →Q = ∑ 1

Tmi→Q ×



i  j ∪i ↔ j

Ri → j ×

n

1 ∑1 D i

1 Di

(2)

4.5 Calculate referential trust rating on the network

After calculation, P can decide whether to interact with Q or not according to the value of TP →Q . After

There are several basic principles when calculating the referential trust rating: (1) If two agents are neighbors or friends, then the trust ratings and reliable factors are obtained directly from the directed graph. (2) We do not enumerate all the paths between two vertices in the graph because it maybe impossible or requires a great deal of time. We adopt breadth-first strategy to find all the paths shorter than the maximum path length MAX. (3) The algorithm of the model gives more weight to shorter paths as the closer of the relationship between two agents, the more reliable of the information. Trust is a sociological concept and exists within interpersonal relationships. Agents can reason, learn from each other and make judgment to simulate humans’ behaviors of how to deal with trust issues. Trust relationships between agents on the SW in essence indicate the trust relationships between the agents’ owners — humans. Therefore, theories in social network field can be applied to the trust management on the SW [11][18]. Original studies of small world indicated that any two people in the world were separated by six acquaintances [18]. Therefore, in this paper, we let MAX equals to six. Suppose we want to calculate trust rating TP →Q . wi

direct interaction, P will get the practical trust rating of Q and use it to replace the computed referential trust rating.

denotes the weight of the ith path. n denotes the

4.6 The Trust Ontology of TCM In this paper, we define the TCM trust ontology that is implemented in OWL [13]. We define the following classes and properties to illustrate the trust relationships. DartTrust: Object has three basic subclasses and one abstract class shown in Figure 2. 1. DartTrust: Person represents all kinds of users in the system. All of them are identified by their email addresses as the email addresses can be viewed as their unique properties. 2. DartTrust: Document refers to TCM magazines, journals and periodicals. One main document database is the TCM bibliographic database built by Information Institute of China Academy of TCM, which contains about one and half million records from 900 biomedical journals and magazines published in China since 1984. The initial trust ratings are based on different ranks given by experts in TCM field. The trust evaluation of different data sources is not immutable. According to the feedbacks from different

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE

users, DartTrust adjusts the trust ratings to personal preferences accordingly.

Figure 2. TCM Trust Ontology 3. DartTrust: Record represents a single query result from database, namely an article in one journal or magazine. At the beginning, all the records from the same document database have the same default trust rating, but DartTrust regularly collects and updates trust information according to different users’ own situations and other users’ feedbacks. For example, initially, all the documents in “Information on Traditional Chinese Medicine” are ranked as high trust by default. When a doctor read about the articles in this journal, he has different opinions towards different articles. He may absolutely trust several articles while distrusts some others. Therefore, he can assign different trust ratings to these articles instead of using the default trust rating. After some practices in his work, he may have new changes about the trust ratings he previously assigned, and then he can update them accordingly. 4. Abstract Class DartTrust: Trusts In the system, there are three kinds of trust relationships: (1) A person trusts another person. For example, a patient trusts a doctor or a doctor trusts his colleagues. (2) A person trusts a document such as a magazine or a journal. For example, a doctor trusts the magazine “China Folk Treatment”, and then he may adopt new treatment methods in it. (3) A person trusts a single record such as an article from the database. The trust evaluations between persons are initially input by users of the system using TrustEditor Component (See Section 6). The other two kinds of trust relationships are initiated by experts and then undergo dynamic changes in accordance with users’ own trust network and preferences.

5. Reasoning function of DartTrust As mentioned in Section 4.6, the trust ontology of DartTrust is written in OWL. OWL is equipped with a formal semantics described in the OWL Web Ontology Semantics and Abstract Syntax [12]. By using these semantics, we can make inferences about ontologies and instances. The environment may vary from time to time therefore it is exhaustive for users to maintain a trust network correctly. Therefore, DartTrust provides a reasoning function to facilitate trust management. We present a simple illustrative example of the DartTrust reasoning function based on TCM trust ontology. In order to present the inference in a clearer way, the ontologies and reasoning functions are written with more details in the following abstract syntax format. Ontology ( ObjectProperty (a: person) ObjectProperty (a: has_friend range (a: person)) ObjectProperty (a: treats range (a: person)) ObjectProperty (a: trusts range (a: Object)) Class (a: friends partial annotation (rdfs: comment “Someone who you can rely on”)) Class (a: doctor complete annotation (rdfs: comment "Someone who treat patients") intersectionOf (restriction (a: treats someValuesFrom (a: person)) a: person))) We can make inference as follows: Class (a: Alan complete a: doctor) Class (a: Tom complete a: doctor) Class (a: Tracy complete a: doctor) Class (a: Alan partial intersectionof (restriction (a: has_friend allValuesFrom (a: doctor)) restriction (a: has_friend someValuesFrom (a: Tom)))) Class (a: Tracy partial intersectionof (restriction (a: has_friend allValuesFrom (a: doctor)) restriction (a: has_friend someValuesFrom (a: Alan)))) SubPropertyOf (a: has_friend a: trusts) Class (a: trust_person complete intersectionOf (a: doctor restriction (a: trusts someValuesFrom (a: friend)))) According to the above inferences, we conclude that: Tom is Alan's friend.

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE

Friends are trusted by each other. Alan is Tracy’s friend. Therefore, Tracy trusts Tom too. With the simple reasoning function, DartTrust manage to infer the friend relationship and expand the trust network to deal with more information sources. Friend is a shortcut for referring. When two agents are regarded as friends, the trust calculation can be simplified. For example, doctor Tracy wants to find an effective method of treating heart disease. It happened that doctor Tom trusts expert Peter absolutely in this field. Although Tracy doesn’t know Peter before, she will follow his method.

By default, their evaluations and annotations are ranked with different trust ratings by the system. Generally speaking, authoritative experts receive the highest trust ratings while the common users have the lowest.

6 The implementation of DartTrust In this section, we introduce the implementation of DartTrust. DartTrust is based on DartGrid [3][9][20] that manages to integrate heterogeneous database resources in an open, dynamic and wide-area environment. DartTrust is composed of four components: TrustEditor, TrustNavigator, TrustAgent and TrustMonitor. TrustEditor: With TrustEditor, users can configure trust values and reliable factors for information sources and other users with respect to a certain field. To generate trust values, DartTrust makes a call to web services, passing the identification of the user and his trust configuration. Users can add comments, annotations and trust ratings that include measures of credibility and reliability about the information sources and other users. Users’ trust network and trust preferences are stored in DartTrust Server, while the information sources store their own trust ratings and feedbacks given by different users in their local databases using the First-in First-out policy. DartTrust also supports sharing and collaboration between users. Users can search and view the evaluation and analysis given by other users as references. Figure 3 shows the trust network of a common user Karl. Broken lines indicate that the person at the center has a trust relationship with the persons at the perimeter but the persons at the perimeter have no trust relationship with the person at the center. The thick solid lines indicate that the person at the center and persons at the perimeter are friends. The thin solid lines indicate that the persons at the perimeter have a trust relationship with the person at the center but the person at the center has no trust relationship with the persons at the perimeter. There are four kinds of users in the system: authoritative experts, specialists, doctors and common users. All of them can express their opinions and rank trust ratings about documents, records and other users.

Figure 3. Trust network of DartTrust user Karl TrustNavigator: DartGrid has integrated more than 50 heterogeneous and distributed TCM databases throughout China [9][20]. Therefore it is essential to know the locations of the ontology definitions and instances. When users encounter some unfamiliar information or service, they can use TrustNavigator to acquire the general trust information about the sources and browse the trust ratings given by other users. TrustAgent: TrustAgent can make diagnosis and treatment recommendations to users base on personal trust networks built by users before. It provides reasoning functions to automatically calculate trust ratings and extend the trust network. TrustMonitor: With TrustMonitor, each object can acquire others’ evaluations about itself, subscription requests, updated broadcasting news and so on. TrustMonitor makes a call to web services to gather the above information from OGSA containers. Figure 4 is the screenshot of query results with trust ratings. The prototype focuses on providing the corresponding diagnosis, treatment and management information about severe diseases that have greatly affected public health. User Karl logs in the DartTrust system and query the information about the most common disease — headache. With his personal trust network and preferences, he obtained a personalized view of the results. The table shows the documents related to headache by a descending order of trust ratings obtained from Karl’s trust network.

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE

8. References

Figure 4. A screenshot of query results of headache

7 Conclusion and Future work In this paper, we have presented a trust prototype of the Semantic Web based on Traditional Chinese Medicine. We illustrate a novel Reputation-Chain trust model — RCSW and introduce the basic concepts and mechanisms of it. RCSW incorporates pairwise trust factor and reliable factor and constructs an edgeweighted graph to calculate trust ratings. We also define a new trust ontology in the TCM domain and present the reasoning function of DartTrust. Our proposed model can provide users with individualized trust management and quickly adapt to the changes. Although we have chosen TCM as our target application, the prototype presented in this paper is also applicable to other Semantic Web projects. For simplicity, we regard trust rating as a single value, which should be extended as tuples in order to meet more requirements of users. At present, DartTrust doesn’t have a strong reasoning ability to deal with complicated situations such as friends’ lying or betrayal. In the future, we will focus on the above aspects and improve them.

Acknowledgement The work is funded by China NSF program (NO. NSFC60503018), and subprogram of China 973 project (NO. 2003CB316906), also a grant from the Science and Technique Foundation Programs (NO. NCET-04-0545).

1. A. Abdul-Rahman and S. Hailes, "Supporting Trust in Virtual Communities", In Proc. of the Hawaii International Conference on System Sciences, Jan, Hawaii, 2000. 2. Amazon Web Site, http://www.amazon.com. 3. DartGrid Project Website, http://ccnt.zju.edu.cn/projects/dartgrid/. 4. D.Huynh, N.R.Jennings, and N.R.Shadbolt, “Developing an Integrated Trust and Reputation Model for Open MultiAgent Systems”, In Proc.3rd Int. Joint Conf. on Autonomous Agents and Multi-agent Systems (AAMAS), New York City, 2004. 5. eBay Web Site, http://www.ebay.com. 6. FOAF Vocabulary Specification, http://xmlns.com/foaf/0.1/. 7. G. Jennifer, B. Parsia, and J. Hendler, “Trust Networks on the Semantic Web”, In Proc of Cooperative Intelligent Agents, Helsinki, Finland, Aug, 2003. 8. G. Zacharia and P. Maes, “Trust management through reputation mechanisms”, Applied Artificial Intelligence, 14(9): 881–908, 2000. 9. H. Chen, Z. Wu, G. Zheng, and Y. Mao, “RDF-Based Schema Mediation for Database Grid”, 5th IEEE/ACM International Workshop on Grid Computing (Grid Computing 2004), Pittsburgh, USA, 2004. 10. J. Sabater and C. Sierra, “Regret: a reputation model for gregarious societies”, In Proc. of the 1st Int. Joint Conference on Autonomous Agents and Multi-agent Systems (AAMAS), 2002, pp. 475-482. 11. L. Adamic, "The Small World Web", In Proc. of ECDL, 1999, pp. 443-452. 12. OWL Web Ontology Language Reference, http://www.w3.org/TR/owl-ref/. 13. OWL Web Ontology Language Overview, http://www.w3.org/TR/owl-features/. 14. RDF Primer, http://www.w3.org/TR/rdf-primer/. 15. SchemaWeb Web Site, http://www.schemaweb.info/. 16. S.D. Kamvar, M.T. Schlosser, and H. Garcia-Molina, “The EigenTrust algorithm for reputation management in P2P network”, In Proc. of the Twelfth International World Wide Web Conference, 2003. 17. S.D. Ramchurn, N.R. Jennings, C.Sierra, and L.Godo, “A computational trust model for multi-agent interactions based on confidence and reputation”, In Proc.2nd Int. Joint Conf. on Autonomous Agents and Multi-agent Systems (AAMAS), Melbourne, Australia, 2003. 18. S. Milgram, “The small world problem”, Psychology Today, 1967, pp. 61-67. 19. Y. Gil, and V. Ratnakar, “Trusting information sources one citizen at a time”, First International Semantic Web Conference, Sardinia, Italy, 2002, pp. 162-176. 20. Z. Wu, H. Chen, C. Huang, and J. Xu, “Towards a Gridbased Architecture for Traditional Chinese Medicine”, International Workshop on Challenges of Large Applications in Distributed Environments, 2003.

Proceedings of the 22nd International Conference on Data Engineering Workshops (ICDEW'06) 0-7695-2571-7/06 $20.00 © 2006

IEEE