Gradual Machine Learning for Entity Resolution

Usually considered as a classification problem, entity resolution (ER) can be very challenging on real data due to the prevalence of dirty values. The state-of-the-art solutions for ER were built variety learning models (most notably deep neural networks), which require lots accurately labeled training data. Unfortunately, high-quality usually expensive manual work, and are therefore not readily available in many scenarios. In this paper, we propose novel paradigm ER, called gradual machine learning , aims enable effective labeling without requirement effort. It begins with some easy instances task, automatically by high accuracy, then gradually labels more iterative factor graph inference. gradual learning, hard task small stages based estimated evidential certainty provided easier instances. Our extensive experiments have shown that performance proposed approach is considerably better than its unsupervised alternatives, highly competitive compared supervised techniques. Using test case, demonstrate promising potentially applicable other tasks requiring