Aydoğan Reyhan - Modern Approaches to Agent-based Complex Automated Negotiation

Auckland University of Technology, Auckland, New Zealand

Nowadays, Multi-Agent Systems (MASs) have been perceived as a core technology for building diverse, heterogeneous, and distributed complex systems such as pervasive computing and peer-to-peer systems. However, the dynamism of MASs requires agents to have a mechanism to evaluate the trustworthiness other agents before each transaction. Trust was introduced to MASs as the expectation of an agent about the future performance of another agent. Thus, it is also considered as an effective tool to initiate interactions between agents. However, Sen [] stated that there are not enough research on the establishment, engagement, and use of trust. The establishment of trust can be seen as the flip side of the evaluation, which focuses on how trustees can gain trust from truster agents actively especially when the subjective opinions are ubiquitous. However, in many real-world circumstances, not only consumers concern about service providers reputation, but providers also care about who are making requests. For example, online auctions may fail due to the bad behaviours of bidders who refuse to pay after winning.

Obviously, since trust and reputation are crucial in the open and distributed environments, the classic single-sided trust evaluations are inadequate. It implies that truster agents also need to be evaluated for their credibility, i.e., identifying consumers behaviours. To address this, we introduce the BiTrust (Bijective Trust) model, which enables trustee agents (providers) to reason about truster agents (consumers) to improve the interaction satisfaction with a 2-layer evaluation filter. Specifically, in the first layer, providers evaluate the rating behaviour of the consumers. In the second layer, providers evaluate the utility gain of the transaction respected to their current expectation. The approach not only helps trustee agents to choose a suitable strategy for gaining trust from truster agents but also to protect their reputation actively. This paper steps toward comprehensive trust management in terms of establishing and using trust stated in []. The experimental results show several benefits from this model.

The remainder of this paper is structured as follows. Section by summarizing and highlighting the future work.

In [] models reputation as a congestion game and develop a strategy for decision making of trustee agents in resource constraint environment. The paper proposes a trust management model called DRAFT, which can reduce reputation self-damaging problem actively by investigating the capability of trustee itself to decide whether to accept or deny incoming requests. However, it is still weak against requests from truster agents who have biased rating intention after their requests are accepted.

The FIRE []. However, the opinion filter approaches perform poorly when the percentage of the unfair raters increase. In this situation, without a strategic establishment of trust, trustee agents may fail to interact with potential trusters.

SRAC [].

Single-sided evaluations could bring the incentives for trustee agents to behave honestly, but not for truster agents, e.g., giving fair ratings. With the BiTrust model introduced in this paper, trustee agents can gain reputation in the protective manner through analysing truster behaviour, estimating the interaction utility, and filtering out non-beneficial transactions.

In the BiTrust model, trust each other is considered as the key to make interaction. It means both the truster and trustee agents need to trust their interacting partners because any damaging reputation will result in the reducing of future interaction. In the BiTrust model, agents have their own preference utility over potential partners. The truster agents utility function is controlled by parameters related to agents preferences. The system is assumed to have no centralized database for trust management. Trustee agents evaluate request makers based on their previous behaviours to decide whether to interact with them or not. Thus, BiTrust does not follow prevailing accept-when-request assumption, and can be distinguished from the DRAFT approach [] in that trustee agents accept an interaction based on trust-aware utility gain rather than assessment of their own capability limitation. Below we will give detail definitions used in the model.