Truth discovery under resource constraints

Social computing initiatives that mark a shift from personal computing towards computations involving collective action, are driving a dramatic evolution in modern decision-making. Decisionmakers or stakeholders can now tap into the power of tremendous numbers and varieties of information sources (crowds), capable of providing information for decisions that could impact individual or collective well-being. More information sources does not necessarily translate to better information quality, however. Social influence in online environments, for example, may bias collective opinions. In addition, querying information sources may be costly, in terms of energy, bandwidth, delay overheads, etc., in real-world applications. In this research, we propose a general approach for truth discovery in resource constrained environments, where there is uncertainty regarding the trustworthiness of sources. First, we present a model of diversity, which allows a decision-maker to form groups, made up of sources likely to provide similar reports. We demonstrate that this mechanism is able to identify different forms of dependencies among information sources, and hence has the potential to mitigate the risk of double-counting evidence due to correlated biases among information sources. Secondly, we present a sampling decision-making model, which combines source diversification and reinforcement learning to drive sampling strategy. We demonstrate that this mechanism is effective in guiding sampling decisions given different task constraints or information needs. We evaluate our model by comparing it with algorithms representing classes of existing approaches reported in the literature.