Establishing Trust in Multi-agent Systems and Developing an Adaptive Framework for Personalized, Persuasive Recommender Systems

本研究计划的目标是双重的：1）构建一个有效的信任建立模型的多代理系统，和2）开发一个自适应的框架，个性化的，有说服力的推荐系统。1)信任建模对于多Agent系统至关重要，在多Agent系统中，Agent需要找到可信任的合作伙伴进行交易，而环境中可能存在不诚实的Agent。到目前为止，信任建模的文献主要集中在提出信任评估模型，帮助代理评估其他代理的可信度。然而，很少考虑信任建立的方向，这使得代理人能够产生其他人的信任，以增加其被选中进行交易的机会。为了填补这一空白，本研究的第一个目标是构建一个有效的信任建立模型。我们将采用机器学习方法，允许代理从其他代理收集信息，学习和预测他们的行为和偏好，并相应地调整其行动过程以建立对这些代理的信任。此外，我们计划通过利用代理之间的关系的社会结构来增强这种方法。通过将研究方向从信任评估（帮助客户找到值得信赖的企业）转向信任建立（帮助企业建立对客户的信任），我们预见到我们提出的信任建立模型对工业界非常有用，具有很大的商业应用潜力。2)推荐系统是帮助用户找到信息、产品和服务的软件系统。几种推荐方法，例如，已经提出了协同过滤、基于知识等，所有这些都以提高推荐准确度为目标。然而，最近的文献已经证明，提供准确的建议是不够的，以增加用户的感知接受的建议。因此，我们的第二个目标是开发一个框架的推荐系统，有能力说服用户接受所提供的建议。此外，该框架必须能够适应任何推荐方法，并根据个人用户的具体特征进行个性化。我们将设计一个详细的框架体系结构和算法，这些算法控制框架的组件如何协同工作以实现预期的结果。我们计划使用强化学习来指导个人用户选择合适的说服策略。我们期望一个具有说服和个性化功能的适应性框架，可以与任何推荐系统一起使用，以提高其有效性。总的来说，我们的上述两个研究目标应提供理论贡献的信任建模和推荐系统的各个领域，并带来实际利益的许多应用领域，包括电子商务，移动商务，社交网络等。

The goal of this proposed research program is twofold: 1) to construct an effective trust establishment model for multi-agent systems, and 2) to develop an adaptive framework for personalized, persuasive recommender systems. 1) Modeling trust is of vital importance to multi-agent systems where agents need to find trustworthy partners for transactions while dishonest agents may exist in the environment. Until now the literature of trust modeling has mainly focused on proposing trust evaluation models that help an agent evaluate the trustworthiness of other agents. However, slight consideration has been given to the direction of trust establishment, which enables an agent to engender the trust of others to increase its chance to be chosen for transactions. To help fill this gap, the first objective of this proposed research is to construct an effective trust establishment model. We'll employ a machine learning approach that allows an agent to collect information from other agents, learn and predict their behaviors and preferences, and accordingly adjust its course of action to establish trust in those agents. Also, we plan to augment this approach by making use of the social structure of relations among agents. By changing the research direction from trust evaluation (helping customers find trustworthy businesses) to trust establishment (helping businesses build trust in their customers), we foresee that our proposed trust establishment model is very useful for industry and has a large commercial application potential. 2) Recommender systems are software systems that help users find information, products and services. Several recommendation methods, e.g., collaborative filtering, knowledge-based, etc. have been proposed, all with the goal of improving the recommendation accuracy. However, the literature has recently witnessed that providing accurate recommendations is not enough to increase the users' perceived acceptance of the recommendations. Therefore, our second objective is to develop a framework for recommender systems that has the ability of persuading users to accept the recommendations provided. Moreover, the framework must be adaptive to work with any recommendation methods, and personalized to the specific characteristics of individual users. We'll design a detailed architecture of the framework and the algorithms that govern how the framework's components work together to achieve the desired results. We plan to use reinforcement learning to guide the selection of appropriate persuasion strategies for individual users. We expect an adaptable framework with persuasion and personalization capabilities that works with any recommender systems to increase their effectiveness. Overall, our above two research objectives should offer theoretical contributions to the respective areas of trust modeling and recommender systems, and bring practical benefits to many applications domains including e-commerce, m-commerce, social networks, etc.