An Enhanced Method for Computation of Similarity between the ...

2 downloads 0 Views 177KB Size Report
An Enhanced Method for Computation of Similarity between the Contexts in Trust. Evaluation using Weighted Ontology. Mohammad Amin Morid. Amin Omidvar.
2011 International Joint Conference of IEEE TrustCom-11/IEEE ICESS-11/FCST-11

An Enhanced Method for Computation of Similarity between the Contexts in Trust Evaluation using Weighted Ontology Mohammad Amin Morid

Amin Omidvar

Hamid Reza Shahriari

Department of Computer Engineering and Information Technology Amirkabir University of Technology Tehran, Iran {morid, aminomidvar, shahriari}@aut.ac.ir

Evaluation of the trust valuing with regard to the context attribute is called context-aware trust. As it is clear, the context value for all the contexts may not be available. Thus, it is essential to have a mechanism for evaluating the trust value of certain context, which is not available, by using the trust value of another context, which is available. It can be done in many different ways such as multiplying the trust value of the trustee in the available context into the similarity between the available and the unavailable contexts [2]. As a result, computing the similarity between two contexts is crucial for trust evaluation in e-commerce systems. There are many researches that attempted to compute the value in different ways, which have their own advantages and disadvantages. Here, the most common methods for computation of the similarity are those use the ontology tree for representation of the contexts. Therefore, the performance of these methods is dependent on how good and balanced the ontology tree is constructed, which is a limitation for them. To overcome the mentioned limitation, we present a weighted ontology tree, which is independent of the tree’s structure. The weights of the ontology tree are computed using the WorldNet English lexical reference system. Finally we introduced an enhanced method for context similarity computation based on the weighted ontology tree. This paper is organized as follows. Section 2 attempts to define related work in context modeling and computing similarity between the contexts. In Section 3, our suggested model is described and at last our study is rounded off with a conclusion in Section 4.

Abstract—In the context-aware trust evaluation, most of the times ontology trees are employed to represent the relation among contexts. Then, the similarity between two contexts is computed according to the contexts’ distance in their ontology tree. Therefore, the performance of these methods is dependent on the tree’s structure and how balanced the ontology tree is constructed, which is a limitation for them. In an unbalanced ontology tree, one branch of a node is split generally while the other branch is split in more details. As a result, this unbalanced ontology tree negatively affects all computations of the mentioned methods. To overcome this limitation, we presented a weighted ontology tree, which is balanced and independent of the tree’s structure. In the proposed tree, each edge is labeled with the similarity distance between its corresponding nodes. To achieve this, we used an approach, which is based on the WorldNet English lexical reference system. Finally, the similarity between two arbitrary contexts in their weighted ontology tree is computed according to their weighted similarity distance in the tree. Having the contexts similarities, trust value for a new context is computed based on the previously experienced contexts. Keywords-Trust management; Context similarity; WordNet; Weighted ontology tree

I.

INTRODUCTION

Trust is a critical concept in the mutual collaboration in dynamic e-commerce systems. It is defined as particular level of the subjective probability with which an agent assesses that another agent will perform a particular action, before he observes such action [1]. In the context of ecommerce systems, the actions are the e-commerce transactions. The trusting agent is called the trustor entity, and the trusted agent is called the trustee entity. To evaluate the trustee’s trustworthiness for a certain trust scope, the trustor analyzes two different kinds of input: quality attributes and context attributes [2]. Quality attributes shows the essential data characterizing the Trustee. Without this attributes, a trustor has no a priori knowledge of the trustee, and cannot start transaction. Context attributes represent contextual information that the trustor requires in order to complete the evaluation of the trustee’s trustworthiness. As a formal definition for context, it is any information that can be used to characterize the situation of an entity [3].

978-0-7695-4600-1/11 $26.00 © 2011 IEEE DOI 10.1109/TrustCom.2011.93

II.

RELATED WORK

There are several previous works which attempted to compute the mentioned similarity between two contexts. To do so, in all of the researches first they used a model for context representation and then they introduced a method for computing similarity between the contexts. Therefore, we split this section according to these two steps. A. Context Modeling In order to compute the similarity between two contexts, the first step is to model the context, which is known as context representation or context modeling. Any approach is used for the context modeling results different types of the similarity computation. Three popular types of these approaches are: ontology remodeling, key word based 721

modeling and task based modeling [4]. Of course, there are several other approaches which can be used in context modeling but they are not as popular as the above approaches which can be found in [11]. Ontology tree is referred to the approach where the contexts are represented in a context ontology tree in a hierarchical structure. Each node in this tree represents a context and is split into two lower level contexts, where the low level contexts are its sub-contexts. For example, Fig. 1 shows ontology tree for network context and its sub-contexts [5]. In [6] they make use of an ontology tree of services using DAML-S6, where each node in the tree representing a type of service. Using ontology tree for representing game application running on a gaming device is another work which is done in [2]. Here, a game application is composed by a game manager component (GM) and by one game scenario component (GS). In [7] they introduced a belieftheoretic reputation estimation model for multi-context communities. They employed an ontology tree to show consumer experience reports and beliefs about various products of a website (i.e. Epinion.com). In the keyword based modeling as the second common approach for context representation combination of keywords is used to show a context. In other words, each keyword set is referred to a certain context. For example, in all the articles there is a keyword section which introduces the main concepts, which shows the article objectives. For example, our paper keywords are: Context-aware trust evaluation, Context’s similarity, Context modeling and weighted ontology tree. In [8] they used this approach for context representation. They considered a file-server application which has three types of services (i.e. contexts): uploadPDFFile with keywords {write, pdf, file}, uploadDOCFile with keywords {write, doc, file}, login with keywords {LoginInfo, UserName, passWD}. Task based modeling is the more application based method for context modeling and is based on an environment’s tasks. Suppose that we are working in a certain environment with

certain tasks. Each task can be considered as a context and is composed of several sub-tasks which are known as task's aspect or task’s attribute. An aspect is the smallest element of a task which describes a special attribute of it. This approach is employed in several researches such as [4,9, 10]. B. Computing Similarity between the Contexts After identification of a model to represent a context, the next step is to specify a method for computing similarity between the contexts. In this section the goal is to show these methods, which have been used in previous researches. In [6] similarity between two contexts is computed by the distance between two node in the context’s ontology tree: ݈ܵ݅݉݅ܽ‫ݕݐ݅ݎ‬ሺܵͳǡ ܵʹሻ ൌ ‫݁ܿ݊ܽݐݏ݅ܦ‬ሺܵͳǡ ܵʹሻ

(1)

Here, the distance of two nodes is defined as the least number of intermediate nodes for one node to traverse to another node. For example, in Fig. 2 which shows services ontology tree, service S1 and S2 has a distance of 3.The method introduced in [2] is another context similarity computation, which uses ontology tree. In [8] they considered any context as a set of keywords and they computed the similarity between two contexts by using the set theory. Here, the similarity between two contexts, Si and Sj, with their individual keywords sets, K(Si) and K(Sj), is defined as the ratio between the set’s intersect and the set’s union: ݈ܵ݅݉݅ܽ‫ݕݐ݅ݎ‬൫ܵ௜ ǡ ܵ௝ ൯ ൌ

௄ሺௌ೔ ሻ‫ת‬௄ሺௌೕ ሻ ௄ሺௌ೔ ሻ‫׫‬௄ሺௌೕ ሻ

(2)

As it was elaborated, one approach for context representation is considering a context as task. In [10] the similarity D (S1, S2) between two tasks s1 and s2 is obtained from the comparison of the task attributes. ଵ

݈ܵ݅݉݅ܽ‫ݕݐ݅ݎ‬൫ܵ௜ ǡ ܵ௝ ൯ ൌ ͳ െ σ௡௟ୀଵหܵ௜ǡ௟ െ ܵ௝ǡ௟ ห ௡

(3)

Where n is the number of task attributes, Si,l is the l-th attribute of task Si, and Sj,l is the l-th attribute of task Sj. Other works in this context can be found in [4].

Figure 2. Figure 2. Services in a context ontology tree [6].

Figure 1. Figure 1. Example of an ontology tree.

722

III.

TRUST EVALUATION USING WEIGHTED ONTOLOGY

B. Proposed Approach In favor of overcome to the described limitation, we suggest to use a weighted ontology tree instead of the traditional tree. Weighted ontology tree is a tree with weighted edges, where each weight between couple of nodes shows the similarity between them. To specify these weights, we use a concept distance method based on the WordNet English lexical reference system. Then, the distance between any two arbitrary contexts can be specified more equitable by an enhanced method which is discussed later. For more clarification, an example is depicted in Fig. 4 to modify Fig. 3. In this figure, the total distance between hardware and VHDL is equal to the total distance between software and Java (i.e. 14). The reason is that, the software node has been split to more branches with low weight values while the hardware has been split to only one branch with high weight value. IV. Context Distance As mentioned before, in order to compute the similarity distance between two corresponding nodes of an edge in the context’s ontology tree, we use a WordNet based concept distance method. In this section, we elaborate this method in more details. WordNet [12,13] is an on-line English lexical reference system that was developed at Princeton University in the 1990s. As an ordinary online dictionary, WordNet lists alphabetically concepts important to a particular subject along with explanation. Additionally, as a machine-readable thesaurus having semantic network, it has a rich array of structures showing semantic relations among words [14]. In the context of text processing, there are variety of methods which aim to compute the similarity between two documents.

This paper attempts to show an advanced ontology tree for context representation in order to overcome limitation of the previous trees. To do so, we show an enhanced method for computing the similarity between two contexts based on the proposed tree. In this section, first we show the limitation of the previous methods and then the proposed enhanced solution will be presented. A. Limitation of Previous Approaches for Context Modeling In section 2, we elaborated three approaches for context modeling. Among these approaches the most popular one is the context modeling by using the ontology tree. The most important limitation of this approach is that the tree may be unbalanced. In particular, on branch of a node may be split generally while the other branch is split in more details. This limitation is elaborated by an example. As shown in Fig. 3, computer science is split to software and hardware. Afterward, the hardware node is split to VHDL programming language while the software is split to programming language, afterward object oriented language and finally the java programming language. As it is seen in Fig. 3, VHDL and java are both a programming language in hardware and software context but their distribution is not equitable. In particular, the distance between hardware and VHDL is an edge while the distance between software and java is three edges and so it is not an equitable distribution. Therefore, the VHDL node should be split into more nodes in order to have a balanced ontology tree. As it is clear, this unbalanced ontology tree causes several problem in the context’s similarity computation methods which are based on these ontology trees.

Figure 3. Figure 3. Unbalanced ontology tree for the computer science concepts.

Figure 4. Balanced ontology tree for the computer science concepts.

723

Where the Si and Sj are any two arbitrary contexts. By using the above method, the weighted distance between two nodes shows their similarity. Here, two contexts which have a large distance value are less similar than those have a small distance value. As it is seen, this modified similarity value is balanced and independent to the structure of the ontology tree. Finally, the last step, is computing the trust value, which is needed for any application. The trust value of any trustor to any trustee in a new context, Cn, can be computed by an experienced context, Ce, according to their similarity. This can be done using different equations such one introduced in [8]:

In the context of the WordNet, if we consider that documents correspond to concepts and words for a document corresponds to words for a concept, then we can use the document’s similarity computation method for the concept’s similarity computation. One of these methods is introduced by Salton [15]. By using this model, words that come out very frequently in the concept’s explanation in WordNet, are unlikely to discriminate concepts sufficiently, words that are used only for few times should have more weights [14]. Let the word frequency wfij be defined as the number of occurrence in explanation of a concept. Also, the concept frequency cfi, is defined as the number of concepts in which the word occurs in a collection of N concepts. Here, the inverse concept frequency (icf) factor is given by log N/cfi. A combined word importance indicator is the product (wf * icf) where the importance, or weight, Wij, of a word wi in the concept cj is defined as the word frequency multiplied by the inverse concept frequency [15]. That is ܹ௜௝ ൌ ‫݂ݓ‬௜௝ Ǥ ݈‫݃݋‬

ே ௖௙೔

.

ܶ‫ݐݏݑݎ‬ሺ‫ܣ‬ǡ ܷǡ ‫ܥ‬௡ ሻ  ൌ ܶ‫ݐݏݑݎ‬ሺ‫ܣ‬ǡ ܷǡ ‫ܥ‬௘ ሻ ‫ݕݐ݅ݎ݈ܽ݅݉݅ܵ כ‬ሺ‫ܥ‬௡ ǡ ‫ܥ‬௘ ሻ (8)

Where Trust(A,U,Cn) determines the trust value of user A to user U in the context Cn.

(4)

VI.

In the context-aware trust evaluation, the context value for all the contexts may not be available. So, it is essential to have a mechanism for indicating the trust value of certain context which is not available. To do so, the available trust value of another experienced context is employed by using many different approaches. In all these approaches, it is needed to compute similarity between the available context and the non-available one. Therefore, computation of this similarity is essential in trust evaluation and there are several methods which attempted to do so. Among these methods, some of them which use ontology tree to represent contexts are more popular. However, the performance of these methods is dependent on how balanced the ontology tree is. It is possible that, one branch of a node is split generally while the other branch is split in more details. To overcome the above limitation, we presented a weighted ontology tree, where an edge of the tree is labeled with the similarity distance among its corresponding nodes. To do so, we employed an approach which uses the WordNet English lexical reference system. By using this approach, words that come out very frequently in the concept’s explanation in the WordNet, have no significant weights, while words that are used only for few times have more weights. The reason is that, the frequent words are unlikely to discriminate concepts sufficiently. Finally, the similarity between two arbitrary contexts of an ontology tree, which are two different nodes, is computed according to the reverse value of their weighted similarity distance. Having the contexts similarities, the trust value for a new context based on previous experienced contexts can be computed.

Through the wf * icf values, the word-concept matrix which is the basis for the similarity measurement of a pairs of concepts can be made. The pair wise comparison of a matrix gives N*(N -1)/2 different pairs of similarity coefficients for the concepts. Finally, the similarity between concepts is measured by the following cosine coefficient measure. That is, ‫݉݅ݏ‬ሺܺǡ ܻሻ ൌ 

σಿ ೔సభ ௫೔ ௬೔

(5)

మ ಿ మ ටσಿ ೔సభ ௫೔ Ǥσ೔సభ ௬೔

where x and y are vectors of wf * icf values and N is the dimension of the vector space. The cosine coefficient can have a value between 0 and 1. As it seen, the more the concepts have overlapping words, the higher the cosine coefficient will be [14]. If there are no overlapping words between the concepts, the cosine coefficient is zero. In order to compute the weighted distance between two contexts of an edge in their ontology tree, we use the equation (5). ܹ݄݁݅݃‫݁ܿ݊ܽݐݏ݅ܦ݀݁ݐ‬൫ܵ௜ ǡ ܵ௝ ൯ ൌ ‫݉݅ݏ‬ሺܵ௜ ǡ ܵ௝ ሻ

CONLUSION

(6)

where Si and Sj are corresponding nodes of the edge. Context Similarity Computation In this section, the whole process for computation the similarity between two arbitrary contexts is presented. First, it is needed to construct a weighted ontology tree. Therefore, we need to have a method for computing the weights of the edges in the ontology tree. As discussed in the previous section, this will be done by equation (6). After specifying the weights in ontology tree, the similarity between any two arbitrary contexts can be computed by using the reverse value of total weighted distance between them: V.

REFERENCES [1] D. Gambetta, “Can we trust? In Gambetta, Diego (ed.) Trust: Making and Breaking Cooperative Relations, 1990, pp.213–237. [2] S. Toivonen, G. Lenzini, I. Uusitalo, “Context-aware Trust Evaluation Functions for Dynamic Reconfigurable Systems.” Workshop on Models of Trust for the Web (MTW'06), May 2006, Edinburgh, Scotland.

݈ܵ݅݉݅ܽ‫ݕݐ݅ݎ‬ሺܵ௜ ǡ ܵ௝ ሻ  ൌ  ͳΤܹ݄݁݅݃‫݁ܿ݊ܽݐݏ݅ܦ݀݁ݐ‬ሺܵ௜ ǡ ܵ௝ ሻ (7)

724

[3] A. Dey, “Understanding and Using Context.Personaland Ubiquitous Computing”, 5, 1, 2001, pp.4-7.

[9] A. Caballero, J. Blaya, and A. Gomez-Skarmeta, “A new model for trust andreputation management with an ontology basedapproach for similarity between tasks.” LECTURE NOTES IN COMPUTER SCIENCE, 2006, pp.172-183.

[4] M. Tavakolifard, S. Knapskog, and P. Herrmann, “Trust Transferability Among Similar Contexts” In Proceedings of the 4th ACM International Workshop on QoS and Security for Wireless and Mobile Networks (Q2SWinet 2009). ACM, 2009.

[10] A. Caballero, J. Botia, and A. Gomez-Skarmeta, “On the Behavior of the TRSIM Model for Trust and Reputation”. LECTURE NOTES IN COMPUTER SCIENCE, 2007, pp.182-193.

[5] S. Toivonen, G. Lenzini, and I. Uusitalo, “Context-aware trust evaluation functions for dynamicrecon_gurable systems”. In Proceedings of the Models of Trust for the Web Workshop (MTW_S06), held in conjunction with the 15th International World Wide Web Conference (WWW2006) May, volume 22, 2006.

[11] T. Strang and C. LinnhoPopien, “A context model in survey”. Workshop on Advanced Context Modeling,Reasoning and Management as part of UbiComp, 2004. [12] G. Miller, R. Beckwith, C. Felbaum, D. Gross and K. Miller, “Introduction to WordNet: An on-line lexical database”. International Journal of Lexicography, 3, 4, 1990, pp. 235–244.

[6] J. Liu, V. Issarny, “Enhanced reputation mechanism for mobile ad hoc networks”. Second International Conference, iTrust 2004, Oxford, UK, March 29–April 1, 2004, pp.48-62.

[13] F. Christiane, “WordNet–An electronic lexical database”. MITPress, 1998.

[7] E. Bagheri, M. Barouni-Ebrahimi, R. Zafarani, and A. Ghorbani, “A belief-theoretic reputation estimation model for multi-context communities”. In Canadian Conference on AI, 2008, pp. 48–59.

[14] S. Lee, S. Huh and R.D. McNiel. “Automatic generation of concept hierarchies using WordNet”. Expert Systems with Applications, 35, 3, 2008, pp.1132–1144.

[8] M. Uddin, M. Zulkernine, and S. Ahamed, “CAT: a context-aware trust model for open and dynamic systems”. In Proceedings of the 2008 ACM Symposium on Applied Computing, NY, USA, 2008, pp.2024-2029.

[15] G. Salton, “Automatic text processing: The transformation, analysis, and retrieval of information by computer”. Addison-Wesley, 1990.

725