Incentivized Information Fusion with Social Sensors

Abstract

This paper deals with the problem of incentivized information fusion, where a controller seeks to infer an unknown parameter by incentivizing a network of social sensors to reveal the information. The social sensors gather information on the parameter after interacting with other social sensors, to optimize a local utility function. We are interested in finding incentive rules that are easy to compute and implement. In particular, we give sufficient conditions on the model parameters under which the optimal rule for the controller is provably a threshold decision rule, i.e, don't incentivize when the estimate (of the parameter) is below a certain level and incentivize otherwise. We will further provide a complete sample path characterization of the optimal incentive rule, i.e, the nature (average trend) of the optimal incentive sequence resulting from the controller employing the optimal threshold rule. We show that the optimal incentive sequence is a sub-martingale, i.e, the optimal incentives increase on average over time.