Pre-screening training data for classifiers

摘要: A system and method provide recommendations for refining training data that includes a set of digital objects. submitter labels the objects in with labels, which may indicate whether object is considered positive, neutral, or negative respect to each predefined classes. Score vectors are computed by trained categorizer labeled set. From score vectors, various metrics computed, such as representative vector distances from label group, cluster, category categorizer. Based on metrics, heuristics applied evaluated be made submitter, proposing mislabeled relabeled. The include unlabeled objects, case, suggestions labeling