Collaboration through computation: incorporating trust model into service-based software systems

Abstract

The open and dynamic nature of service-based software systems necessitates spontaneous and trustworthy interactions between collaborating entities. Service providers are exposed to users spanned across multiple organizational domains, so can be exploited by potentially untrustworthy service requestors. Given that, service providers need to trust requestors before granting them with services. Trust encompasses a number of quality attributes (e.g., security, competence, honesty) and helps in dynamic decision making. In this paper, we present a trust-based service collaboration approach, facilitated by the analysis of service-based interactions between service providers and requestors, and recommendations between service providers. Service providers exchange recommendations to convey their trust on requestors. This collaboration is quantified using our proposed trust model, called CAT, a Context-Aware Trust model based on service-based interactions by considering services as contexts. We identify a number of collaboration-based trust properties including risk and context-awareness and incorporate them in CAT. A context-similarity parameter is introduced to decide on similar services. A time-based ageing parameter is proposed to decrease trust values over time without any further interactions. Direct and indirect recommendations from other service providers are included in total trust calculation, with a path-based ageing parameter applying over indirect recommendations. A mechanism to calculate the accuracy of recommendations is proposed to differentiate between reliable and unreliable recommendations. These calculation schemes are employed in a trust-based service collaboration algorithm to automatically decide on granting services to requestors. The approach is elaborated using examples from file sharing applications, and successfully evaluated by implementing a prototype service-based file sharing grid.

Trustworthy interaction balancing in mixed service-oriented systems

“…This trust model relies on mechanisms of SOA, such as SOAPbased interaction monitoring, or Web services based trust provisioning. Other works in that domain, such as [9,16,20] disregard the human factor in large-scale networks.…”

Dynamic Service Selection with Reputation Management

“…They claim the proposed reputation management mechanism is robust against various malicious behaviors, but they presume that there exist a few trusted third parties that always produce credible QoS reports and are used as trustworthy information sources to evaluate the behaviors of the other users. M. G. Uddin et al [10] purpose a CAT (Context-Aware Trust) model to make service granting decision on behalf of a service provider entity. But [10] does not consider defense against the Sybil attacks, where a malicious entity creates a large number of artificial entities for the purpose of providing false recommendations to influence trust-based decision making.…”

“…M. G. Uddin et al [10] purpose a CAT (Context-Aware Trust) model to make service granting decision on behalf of a service provider entity. But [10] does not consider defense against the Sybil attacks, where a malicious entity creates a large number of artificial entities for the purpose of providing false recommendations to influence trust-based decision making.…”

“…The semantic matching algorithm is based on [4], but our semantic matching algorithm extends the original approach and allows for expressing of the user perspective satisfaction level. The trust and reputation management of our work is similar to that of [7,9,10] but we employ an idea of median filter and subdivision of time to restrict malicious reputation manipulation behaviors and reputation abuse. We also provide exhaustive means for credibility evaluation of feedback resource to handle collaboration attacks of various collusive malicious behaviors.…”

“…For the sack of easy illustration, we group fifteen service users with similar behaviors as one.According to ε =0.1, μ=0.9, we can get Qmin=150 and λ =1 based on (9) and(10). The credibility of a user takes into account only his local reputation as the λ equals 1, i.e.…”