A Decision Support Approach to Modeling Trust in Networked Organizations

The main motivation for organizations to e-collaborate is to enable knowledge sharing and learning in order to effectively address a new business opportunity by forming a Virtual Organization (VO) for solving the given task. One of the difficulties in VO creation is appropriate partner selection with mutual trust, as well as the support for the management of trust in a broader Virtual organization Breeding Environment (VBE) – a cluster of organizations willing to collaborate when a new business opportunity appears. This paper proposes an approach to modeling trust in a network of collaborating organizations, aimed at improved trust management in VBEs and improved decision support in the process of VO creation. 1 A Decision Support Approach to Trust Modeling For trust modeling, the decision making problem of trust estimation can be decomposed into decision sub-problems. A mutual trust estimate can be performed by utility aggregation functions used in hierarchical multi-attribute decision support systems [1] in which values of top-level decision criteria are computed by aggregating values of decision criteria at lower levels of a hierarchical tree, which is used to decompose a decision making problem into sub-problems. Decision support system DEXi, used in our system for trust modeling, enables the development of qualitative hierarchical decision support models. DEXi is based on the DEX decision support system [1] which can be used to evaluate incompletely or inaccurately defined decision alternatives, by employing distributions of qualitative values, and evaluating them by methods based on probabilistic or fuzzy propagation of uncertainty. Knowledge about mutual trust can be acquired through a simple questionnaire that a partner of a networked organization can fill-in to describe the competencies of its own organization and the collaborating partner’s performance in previous joint collaborations (for organizations with which the partner has collaborated in past joint projects). For example, the relevant fields of a questionnaire could include: • a list of partner’s own competencies, • a list of competencies of the collaborating partner, and • collaborating partner’s trust estimate based on (a) estimated collaborating partner’s reputation (image, market share), (b) number of successful joint past collaborations, (c) estimate of the profit made in joint collaborations, (d) estimate of the partner’s timeliness in performing assigned tasks, (e) A Decision Support Approach to Modeling Trust in Networked Organizations 747 estimate of the partner’s quality of performance and products, and (f) estimate of the partner’s appropriateness of costs of performance and products. 2 Web-Based Trust Modeling: Estimating Research Reputation and Collaboration of Project Partners A questionnaire-based approach is a preferred means for the estimation of trust between organizations that have known each other based on their experiences in past collaborations. An alternative approach to trust modeling is through the analysis of publicly available Web resources. A Web-based trust modeling approach, similar to the one proposed by [2], is more adequate for roughly estimating the reputation and joint collaborations of partners (individuals or organizations) when a consortium is build of numerous new partners whose past performance is not known. It is also an interesting approach to trust modeling in professional virtual communities and communities of practice. This paper applies the proposed Web-based approach to modeling reputation and trust between partners of a large 6 FP integrated project ECOLEAD. The project has an ambitious goal of creating the foundations and mechanisms for establishing advanced collaborative and network-based industry society in Europe. There are 102 registered individuals from 20 organizations participating in the ECOLEAD project. The left-hand side of Figure 1 shows the Web-based estimates of research reputation and joint collaborations of individuals of the ECOLEAD consortium. To model trust between project members, the following procedure was used: 1. Collect the information about partners’ research reputation, based on the publications of each individual: WOS(Y) the number of publications of author Y in journals with SCI or SSCI factor (obtained through the Web of Science system), and CITESEER(Y) the number of citations of papers of author Y (obtained by the CiteSeer system). 2. Collect the information about past joint collaborations between each two individuals: CITESEER(X,Y) the number of jointly written documents of authors X and Y (obtained by the CiteSeer system), and GOOGLE(X,Y) the number of common appearances of individuals X and Y on the Web (obtained by Google search). 3. Finally, calculate research trust, estimated as weighted sum of reputation and joint collaborations estimates. The calculation of trust between two partners is performed using following function: )) , ( ) , ( ( )) ( ) ( ( ) , ( Y X GOOGLE w Y X CITESEER w w Y CITESEER w Y WOS w w Y X TRUST Google CiteDoc c CiteCit WOS p + + + = where wWOS, wCiteCit, wCiteDoc, wGoogle, wp, and wc are weights of WOS publications, CiteSeer citations, joint publications in CiteSeer, and collaborations found by Google, respectively. In the model used in our experiment, all the weights were set to 0.5, while the numbers of publications, citations, joint publications and collaborations were normalized to values on the [0,1] interval. Note that the functions used for trust estimation are not commutative, so trust of X to Y and trust of Y to X must both be calculated. Having calculated the trust estimates, one is able to rank individual 748 N. Lavrač et al. network partners according to their research reputation, joint collaborations and the overall trust estimate. The Web-based trust estimation model can be used also for other purposes: visualization of the entire trust network, as well as finding wellconnected sub-graphs with high trust utility value, representing ‘cliques’ of partners with strong mutual trust. Fig. 1. Two graphs showing Web-based estimates of research reputation and joint collaborations of individual ECOLEAD researchers (left-hand side), and organizations constituting the ECOLEAD consortium (right-hand side). For anonymity, actual names of individuals and organizations have been replaced by neutral member and institution labels In Figure 1, project and sub-project coordinators turn out to be in central positions according to collaborations. Some of the 102 individuals are not in the graph: those who have a few collaborations and/or low research reputation value. Some well collaborating individuals represent ‘cliques’ of individuals, e.g., researchers from the same organization (same color intensity of nodes) typically have more joint collaborations than researchers from different organizations. From the estimates of reputation and collaborations of individuals, research reputation and collaborations of organizations can be estimated.