Our work considers leveraging crowd signals for detecting fakenews and is motivated by tools recently introduced by Facebookthat enable users to flag fake news. By aggregating users’ flags, ourgoal is to select a small subset of news every day, send them toan expert (e.g., via a third-party fact-checking organization), andstop the spread of news identified as fake by an expert. The mainobjective of our work is to minimizethe spread of misinformationby stopping the propagation of fake news in the network. It isespecially challenging to achieve this objective as it requires de-tecting fake news with high-confidence as quickly as possible. Weshow that in order to leverage users’ flags efficiently, it is crucialto learn about users’ flagging accuracy. We develop a novel algo-rithm,Detective, that performs Bayesian inference for detectingfake news and jointly learns about users’ flagging accuracy overtime. Our algorithm employs posterior sampling to actively tradeoff exploitation (selecting news that maximize the objective valueat a given epoch) and exploration (selecting news that maximizethe value of information towards learning about users’ flaggingaccuracy). We demonstrate the effectiveness of our approach via ex-tensive experiments and show the power of leveraging communitysignals for fake news detection

Top