Analysing the Noise Model Error for Realistic Noisy Label Data
This work addresses the challenge of improving model robustness for researchers and practitioners using automatically annotated data, though it is incremental as it builds on existing noise modeling techniques.
The authors tackled the problem of noisy label data from distant supervision by analyzing the error in estimated noise models, deriving theoretical error bounds and validating them on synthetic noise and a new realistic dataset, NoisyNER, which includes seven noisy label sets with parallel clean labels.
Distant and weak supervision allow to obtain large amounts of labeled training data quickly and cheaply, but these automatic annotations tend to contain a high amount of errors. A popular technique to overcome the negative effects of these noisy labels is noise modelling where the underlying noise process is modelled. In this work, we study the quality of these estimated noise models from the theoretical side by deriving the expected error of the noise model. Apart from evaluating the theoretical results on commonly used synthetic noise, we also publish NoisyNER, a new noisy label dataset from the NLP domain that was obtained through a realistic distant supervision technique. It provides seven sets of labels with differing noise patterns to evaluate different noise levels on the same instances. Parallel, clean labels are available making it possible to study scenarios where a small amount of gold-standard data can be leveraged. Our theoretical results and the corresponding experiments give insights into the factors that influence the noise model estimation like the noise distribution and the sampling technique.