Investigating the Working of Text Classifiers
This work addresses the problem of overfitting to keywords in text classification for NLP researchers, offering an incremental improvement through regularization.
The study investigated whether text classifiers learn compositional meaning or rely on keywords by constructing datasets without lexicon overlap between training and test splits, finding a significant performance drop. It demonstrated that simple models with proposed regularization techniques to discourage keyword focus substantially improve classification accuracy.
Text classification is one of the most widely studied tasks in natural language processing. Motivated by the principle of compositionality, large multilayer neural network models have been employed for this task in an attempt to effectively utilize the constituent expressions. Almost all of the reported work train large networks using discriminative approaches, which come with a caveat of no proper capacity control, as they tend to latch on to any signal that may not generalize. Using various recent state-of-the-art approaches for text classification, we explore whether these models actually learn to compose the meaning of the sentences or still just focus on some keywords or lexicons for classifying the document. To test our hypothesis, we carefully construct datasets where the training and test splits have no direct overlap of such lexicons, but overall language structure would be similar. We study various text classifiers and observe that there is a big performance drop on these datasets. Finally, we show that even simple models with our proposed regularization techniques, which disincentivize focusing on key lexicons, can substantially improve classification accuracy.