Framework of Reputation Aggregation Management for Service-Oriented Business Ecosystems

Size: px
Start display at page:

Download "Framework of Reputation Aggregation Management for Service-Oriented Business Ecosystems"

Transcription

1 Framework of Reputation Aggregation Management for Service-Oriented Business Ecosystems Le Xin Tsinghua Nationa Laboratory for Information Science and Technoogy, Department of Automation, Tsinghua University Being, China Yushun Fan Tsinghua Nationa Laboratory for Information Science and Technoogy, Department of Automation, Tsinghua University Being, China Abstract In Service-Oriented Business Ecosystems, services can be provided and consumed on demand dynamicay. However, seection ony based on function and QoS is not enough for business services. Reputation based trust is becoming more and more important in seection and recommendation of services. In current reputation management, the acks of customization and improper aggregation function coud ead to dissatisfactory resut. In this paper, a cross-patform feedback based reputation management method is proposed to support the customer s comprehensive requirement incuding preference, and to be abe to anayze the evauation of past consumers more exacty. This mode can sove the probems of data structure s difference and rating s unfairness. Finay the impementation framework is introduced to achieve these functions. Keywords- business services; trust and reputation; customer s preference; reputation aggregation I. INTRODUCTION With the support of IT technoogy, especiay internet and coud computing [1], enabed by the Web 2.0 and SOA(service-oriented architectures) [2], increasing numbers of business services have been virtuaized and depoyed on the internet. These business services can compose and cooperate with each other through business networks [3] or Service-Oriented Business Ecosystems (SOBE) [4, 5]. Internet of Service (IoS) then provides a concept, in which the interaction between service vendors and consumers, as we as orchestration of services can become easier and simper to appy via the Internet. Currenty the seection, composition, recommendation of services has been researched abundanty in Web service reated topic. These Web services with the same functionaities may have different performances, such as price, response time, avaiabiity, reputation, security eve and so on. These quaity-of-service (QoS) factors, as the measurements of non-functiona features, are important to the service seection and composition. Most composition and seection works of Web services have been done based on QoS [6-8]. However, business services have more important features different from Web services which may cause the deviation if we treat them as the same. Web services can be accessed via Simpe Obect Access Protoco (SOAP) and be understood by Web Services Description Language (WSDL) [9]. The interoperabe environment of Web Services is unified by standards and protocos. Thus the QoS of Web services are meaningfu that the invoker can reaize the quaity of the specific service. Aso some QoS attributes can be obtained automaticay from the service executing process on the Web. For business services, the environment is more compex because of more interaction between customers and service vendors, and the reiance among the services. From the aspects of persons, different customers vaue different characters of services, eg, some may vaue the response time most, whie other may want to save the cost first. The researches on service seection mosty hande different QoS factors equay [8, 10], which ead to absence of customization in service seection. Achieving to differ and determine the importance of each QoS attributes, dynamicay identify and satisfy customers demand is one of our research targets in this artice. From the aspects of services, a maority of business services are executed practicay offine. The dependence of each other may cause the quaity being mutabe, especiay in the composition of services. For exampe, the sharpy increasing shipment in saes promotion of Amazon may cause the ogistics deivery deay a ot. Other eements of services such as capacity, physica ocation wi affect the services seection resut. For the mentioned reasons, the QoS of business services cannot be acquired automaticay or assigned by vendors either. The most credibe and true information shoud be the feedback of consumers perception on services, that is the reputation. A service consumer can decide how he trust a service with the reputation data. Hence, reputation becomes an vita and essentia criteria in service seection and management. Varieties of researches have been done for managing the reputation of services. A number of modes for managing reputation have been introduced in the iterature [11, 12]. The approach of aggregating the feedback rating to deriving the service provider s reputation is estabished in [13]. The framework supporting the natura anguage in querying, evauating, and representing is introduced [14]. A dynamic service seection framework with the abiity of detecting and

2 deaing with fase or unfair feedback from cheating users is proposed in iterature [15]. However, these approaches are not enough for IoS. The muti-eve services on the internet are heterogeneous from the business ogic point of view. Some of them can be depoyed in Business to Customer (B2C) mode, which is usuay adopted in eectronic market of seing; others may be organized in Onine to Offine (O2O) mode, which supports the query and payment onine, and provide the actua service offine. There are miions of webs or patforms of services on the internet. The same service may be provided on different webs at the same time. To reaize the quaity of a service fuy, there shoud be a framework to support the aggregation of heterogeneous reputation data from various service webs. Considering the more and more attention to customization, this reputation management framework shoud be easy to appy to the individuay recommendation scheme. The rest of the artice is organized as foow. Section 2 presents a reputation aggregation process. Section 3 describes a fuzzy data mode of reputation based on QoS attributes, and then raised the web based muti-patform reputation anaysis mode. Section 4 proposed the impementation of a reputation management framework based on the proposed modes. Section 5 ends the paper with some concusion. II. WEB BASED REPUTATION AGGREGATION PROCESS In this section we describe the process of reputation aggregation for the Internet of Services in the whoe ife cyce of service seection. Figure 1. Life Cyce of Service Reputation Aggregation As iustrated in Figure 1, the whoe service seection and reputation aggregation process can be divided into three stages, service designing, service execution, and service evauation. The roes for service seection incude three types. Service provider is a kind of person who offers the service on the patform and gains from the services consumption. Service patform operator estabishes and maintains a patform, usuay on the internet, to integrate the services and buid a bridge between the vendors and customers. Service customer consumes the services, and pays for them, aso gives feedback of their experiences after using. After the service customer raises the request, the workfow or service composition wi be defined on the service patform on the internet automaticay or semiautomaticay. The function query wi be done through the service register center, and then the service seection scheme can be obtained. This is the designing stage before execution. At the second stage, the service execution is started from binding of composition scheme with the service instances. After this matching, the services can be impemented in accordance with the choreography scheme. During the carrying out of services, the effect and resut wi be monitored by the web patform automaticay. If there are any services in faiure, the composition wi be broke off and restarted with other avaiabe ones. The resut of this execution stage wi be stored and the feedback of customers wi be coected to do some pretreatment. The main aim of doing the pretreatment is to fiter out some incredibe evauation data or comments, and then the reputation can be more convincibe to be taken into account by the potentia customers. With the resut done by cross-patform aggregation, the service s reputation can be updated in rea time. Thus the service provider, service patform operator, and other service customers who care about this service can be aware of the change of its reputation data.

3 In the current reputation management system, there are severa probems waiting to be soved. The most significant issues are as foowing. A. Integration difficuties In the SOBE, the business services can be depoyed on different patforms on the internet. In order to coect the overa execution effects of a specific service, the best soution is to provide a search engine of service reputation data. For the different operation mode of these patforms, the customers feedback data structure wi differ from each other. The reputation data structures of severa main eectronic business patforms are introduced in Tabe 1. Some of the feedback are numera rates, some are inguistic comments; some of them are normaized in [0,1], and some are in. Even if the numera rates have the same range of vaues, their attributes may be extracted from various views to evauate the quaity of services. How to integrate these heterogeneous data into a unified assembe mode is one topic we discuss in the foowing section. TABLE I. Service Patform REPUTATION DATA STRUCTURE OF MAIN ELECTRONIC MARKETS Reputation Attributes Overa Item as described, Seer s attitude, Diivery speed. Item as described, Communication, Shipping time, Shipping and handing charges. Ease of Use, Quaity of Tech. (Shoe) Overa. Price/Vaue, Durabiity, Functionaity, Apperance. (Hote) Overa. Equipment, Price, Location, Restaurant, Traffic, Service. Range of Vaues {-1,0,1} [0,1] B. fairness of reputation In the evauation process, some improper mode may misead the reputation based on rating. For exampe, service A has 99 favorabe comments in 100 customers, and service B has ony 1 customer, he gave high opinion. If the patform uses the simpe mode to cacuate favorabe rate, service A(99/100) is worse than B(1/1), however, this is not the true description of their popuarity and reiabiity. In another case of hote reservation, the feedback rate expresses the experience of a specific customer s own opinion. Whie the rate of a customer who thinks price highy provides ess reference vaue for one who vaues service quaity most. How to take the preference of the rater and potentia customer into account and adust the evauation to eiminate the deviation is another topic we woud ike to sove in this artice. III. MULTI-PLATFORM REPUTATION ANALYSIS MODEL In this section, we first propose a customer s requirement mode, and then define the feedback based reputation, at ast introduce a method of reputation aggregation. A. Customer s requirement mode With the support of data storage and anaysis, the customization can be achieved better to subdivide the customer s requirement into three parts. CR { FunD, QosR, Pref } (1) FunD is a description of the customer s demand of function. It can be achieved through UDDI mechanism. Customers can express their own demand in keyword, or search the function ist to define the requirement. QosR stands for the restriction of QoS attributes. This part represents the acceptabe condition of QoS. It is usuay defined as the format of QoSi [min, max]. Pref is the information records of each QoS attributes importance to the customer. It is given by: Pref ( s, c ) [( QoS, w ),,( QoS, w )] (2) i 1 1 i i Where w [0,1] That is, for service s, a service customer c i vaue the attribute i of QoS at the degree of w. This degree can be assigned by the customer or obtained from his feedback history. B. Feedback based reputation mode The reputation presents the perception of the customers regarding a service vendor. A given service s reputation is a coective view of the customers who have interacted with it in the past [16]. The feedback of each consumer of the service is caed persona evauation (PerEva). This PerEva coud be a an overa rate or a vector vaue for each QoS attribute. For service s in the patform p k, a service customer c i gives a -eement vector PerEva representing c i s experience of s s behavior. Thus, the reputation of s can be defined as: Rep( p, s, c ) ( PerEva ) (3) k i pk, L Reputation ( s ) [ Rep( p, s, c )] (4) k i k K C where represents the aggregation function of services on the views of QoS attributes, and means the synthesis of a customer s feedback in the same patform, whie stands for integration of the reputation data from different service patforms.

4 C. Method of reputation aggregation In order to integrate the feedback of different patform of services, taking the customer s persona preference into account, we propose a QoS attributes based distribution of reputation method for SOBE. The aggregation method is stated in detais as foowing. 1) Anayze the request customer s requirement. Get the customer s FunD and QosR to fiter out a ist of services that meet the requirement of this potentia customer. Aso obtain the Pref data for the customized recommendation. In this step, the customer coud define the aggregation QoS attributes that he cares about for this specific kind of service, denoted as { QoS C }. When customer views the reputation of service s, do the customized anaysis. 2) For each patform p k containing the service s, distribute the evauation to QoS attributes. In this step, acquire the average of ratings, denoted as m, and the average number of rates C. For each rate PerEva of s, if exists Q ( QoS ), et C r ( RC ) ri PerEva (5) Ese if for ( R C) r there is not a same QoS attributes in feedback data of p k, we coud distribute the evauation to it. R PerEva m (6) i ( RC ) ri m R wr ( ci ) (7) where wr is the importance degree of the past user c i in the equation (2). Thus for this patform, the reputation of s can be cacuated as, n Cm (R ) i 1 C ri Rep( pk, s ) [ ] (8) pk, r QoS C C n The dimension of this reputation is the same as QoS C, according to the requester s demand. And Equation (8) means use the Bayesian average method to eiminate the error cause by difference of rater numbers, in which n denotes the voter turnout of s. Each rate is normaized to. 3) Aggregate the reputation data from different patforms. After obtain each service s evauation from patform p k, the tota reputation can be achieved by assembe them together. In this step, for we have acquired the same dimension of vectors, as we as the same range of vaues in step 2), thereby the integration is easy to achieve. Reputation ( s ) Rep( p, s ) (9) k k K 4) Cacuate the recommendation factor. QoS C w 1 CR Reputation s RF( CR, s ) ( ( ) ( )) (10) Reputation ( s ) means the assembe reputation of s in the attribution QoS. And this factor is the degree that how this service fits the customer s requirement and it can be the basis for sorting the service ist to recommend services. Figure 2. Impementation of reputation management of framework

5 IV. IMPLEMENTATION OF THE REPUTATION MANAGEMENT FRAMEWORK In this Section, the impementation framework of reputation management is introduced as Figure 2. The key activities can be supported in this frame. The function of every ayer is proposed as foows. User view ayer: In business service seection scheme, users are divided into three kinds of roes, customer, provider, and operator. A of them are abe to access their own customized pages. This is aso the human-computer interaction interface of the whoe system. Business component ayer: The main reputation management function modues are arranged in this ayer. Users requirement and preference are resoved in user management modue. The service modeing modue provides methodoogy of service defining for vendors. SOBE search engine enabe consumer to query services from different patform. Service monitor contros the executing effect in performance. In the reputation management component, the search and anaysis, reputation aggregation, and customized service recommendation based on reputation are supported for customers. Mode component ayer: Provide common access, encryption and decryption of data and other operation management. Storage ayer: Not ony the service data is stored, the customers preference is saved in the database. V. CONCLUSION In this paper, fist we introduce the business services in the SOBE, as we as the Internet of Services. To support the customized service recommendation, cross-patform service query and the personaized reputation anayze method is necessary. In order to sove the probems of integration difficuties and unfairness, we raise a new customer s requirement mode and a feedback based reputation mode. On these bases, a method for reputation aggregation of services on the Internet is proposed to assembe feedback evauation of past consumers from different patforms. Finay, the impement framework of feedback based customized reputation management is presented. Our work proposed in this artice is ony a basic research for the reputation management and the improved agorithm to reaize a the customization functions needs to be studied in the future work. ACKNOWLEDGEMENTS This work is supported by Nationa Key Technoogy R&D Program under Grant No. 2012BAF15G00. REFERENCES [1] M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. Katz, A. Konwinski, G. Lee, D. Patterson, A. Rabkin, I. Stoica, and M. Zaharia, "A view of coud computing," Communications of the ACM, vo. 53, pp , [2] C. Schroth and T. Janner, "Web 2.0 and soa: Converging concepts enabing the internet of services," IT Professiona, vo. 9, pp , [3] T. Kohborn, A. Korthaus, C. Ried, and H. Krcmar, "Service aggregators in business networks," in Enterprise Distributed Obect Computing Conference Workshops, EDOCW th, 2009, pp [4] J. L. Zhang and Y. S. Fan, "Service-Oriented Enterprise and Business Ecosystem," Computer Integrated Manufacturing Systems, vo. 16, pp , [5] S. F. Li and Y. S. Fan, "Research on the Serviceoriented business ecosystem," in Internationa Conference on Advanced Computer Contro (ICACC), 2011, pp [6] G. Canfora, M. Di Penta, R. Esposito, and M. L. Viani, "Qos-aware repanning of composite web services," in Web Services, ICWS Proceedings IEEE Internationa Conference on, 2005, pp [7] C. W. Zhang, S. Su and J. L. Chen, "Genetic Agorithm on Web Services Seection Supporting QoS," Chinese Journa of Computer, vo. 29, pp , [8] A. Huang, C. W. Lan and S. Yang, "An optima QoSbased Web service seection scheme," INFORMATION SCIENCES, vo. 179, pp , [9] M. Lun, W. K. Chan and T. H. Tse, "An Adaptive Service Seection Approach to Service Composition," in IEEE Internationa Conference on Web Services (ICWS '08), 2008, pp [10] X. Q. Fan, X. W. Fang and C. J. Jiang, "Research on Web service seection based on cooperative evoution," EXPERT SYSTEMS WITH APPLICATIONS, vo. 38, pp , [11] G. Chang, "A reputation mode of web services," in Communication Software and Networks (ICCSN), 2011 IEEE 3rd Internationa Conference on, 2011, pp [12] S. Ruohomaa, L. Kutvonen and E. Koutroui, "Reputation Management Survey," in Avaiabiity, Reiabiity and Security, ARES The Second Internationa Conference on, 2007, pp [13] Z. Maik and A. Bouguettaya, "RATEWeb: Reputation Assessment for Trust Estabishment among Web services," VLDB JOURNAL, vo. 18, pp , [14] S. Nepa, W. Sherchan, J. Hunkinger, and A. Bouguettaya, "A Fuzzy Trust Management Framework for Service Web," in Web Services (ICWS), 2010 IEEE Internationa Conference on, 2010, pp [15] S. Yan, X. Zheng and D. Chen, "Dynamic Service Seection with Reputation Management," in Service Sciences (ICSS), 2010 Internationa Conference on, 2010, pp [16] S. Nepa, Z. Maik and A. Bouguettaya, "Reputation Propagation in Composite Services," in Web Services, ICWS IEEE Internationa Conference on, 2009, pp