| 讲座简介： | Abstract: As online ratings enable strangers to develop trust in interactions, users expect them to reflect a direct measure of quality, which allows for the assessment of the risk entailed in potential transactions. Using a massive dataset of nearly 1.5 million interactions between 2003 and 2012 since the inception of Couchsurfing.org, a popular online hospitality exchange service that disallows monetary exchanges, we show evidence that such evaluations in fact may be systematically distorted by many complex social-psychological processes. We find a major structural imbalance of mutual evaluations between interacting partners who possess different demographic and role-related characteristics. These distortions may lead to a number of undesirable outcomes, as they may signal inconsistent information about users and products, which creates biases and may affect decision making, inducing users to strategically seek to preserve their reputation at the expense of other more relevant utilities. We derive an evolutionary game theoretical model from empirical outcomes of interactions to reason about these situations. We treat rating imbalance as reputation payoffs and model user strategies in the reputation game as immutable (or evolutionary), capturing the fact that it is unrealistic to change one's demographic features for the sole purpose of playing optimally. Last, we seek evidence that users may try to counteract the inevitable effect of their personal features on ratings by strategically selecting others for interaction in ways that minimize their impact on reputation. We discover via a blockmodel that users often deviate from their optimal strategy. However, we find evidence that this sub-optimal behavior stems from a lack of awareness of the reputation game dynamics. We observe that many users learn and eventually switch to optimal strategies over time, while some abandon optimal strategies once they accrue enough reputation to achieve their goals. |
