Social learning refers to the process by which networked strategic agents learn an unknown state of the world by observing private state-related signals as well as other agents' actions. In their classic work, Bikhchandani, Hirshleifer and Welch showed that information cascades occur in social learning, in which agents blindly follow others' behavior, and consequently, the actions in a cascade reveal no further information about the state. In this paper, we consider the introduction of an information coordinator to mitigate information cascades. The coordinator commits to a mechanism, which is a contract that agents may choose to accept or not. If an agent enters the mechanism, she pays a fee, and sends a message to the coordinator indicating her private signal (not necessarily truthfully). The coordinator, in turn, suggests an action to the agents according to his knowledge and interest. We study a class of mechanisms that possess properties such as individual rationality for agents (i.e., agents are willing to enter), truth telling, and profit maximization for the coordinator. We prove that the coordinator, without loss of optimality, can adopt a summary-based mechanism that depends on the complete observation history through an appropriate sufficient statistic. Furthermore, we show the existence of a mechanism which strictly improves social welfare, and results in strictly positive profit, so that such a mechanism is acceptable for both agents and the coordinator, and is beneficial to the agent community. Finally, we analyze the performance of this mechanism and show significant gains on both aforementioned metrics.

翻译：社会学习是指网络中的策略性智能体通过观察私有状态相关信号以及其他智能体的行为来学习未知世界状态的过程。在经典研究中，Bikhchandani、Hirshleifer与Welch指出，社会学习中存在信息级联现象，即智能体盲目模仿他人行为，导致级联中的行为无法再提供关于状态的新信息。本文考虑引入信息协调者以缓解信息级联。该协调者承诺采用一种机制——即智能体可选择接受或拒绝的契约。若智能体加入该机制，需支付费用，并向协调者发送包含其私有信号（不必真实）的消息。随后，协调者根据自身知识与利益向智能体建议行动。我们研究了一类具有个体理性（即智能体自愿参与）、如实报告及协调者利润最大化等性质的机制。证明了协调者在不损失最优性的前提下，可采用基于摘要的机制，该机制通过适当的充分统计量依赖于完整的观测历史。此外，我们证明了存在一种能严格改善社会福利并带来严格正利润的机制，因此该机制对智能体与协调者双方均可接受，并有益于智能体群体。最后，我们分析了该机制的性能，表明其在上述两项指标上均有显著提升。