Robust Sparse Weighted Classification For Crowdsourcing

Data collected from nature is usually unlabeled, and it difficult to be used directly. This issue well addressed by crowdsourcing, which provides a reasonable way for effectively using these unlabeled data. Generally, workers in crowdsourcing tasks are not professionals, so hard obtain high-quality labels. To address this issue, robust sparse weighted classification algorithm proposed, try adjust the samples that correctly classified original lables as much possible. Specifically, we evalute ability of different workers(indicator weight matrix) accurately label fitting real data matrix its reconstruction matrix. And then, $ l_{2,1}$-norm worker labeling similarity added, negative effects some bad eliminated through row sparsity property l_{2,1}$-norm. Finally, optimal indicator obtained optimizing two matrices objective function simultaneously. Therefore, takes into consideration, infers all predicted The results on synthetic sets demonstrate our superior other state-of-the-art methods.