Search

Text classification is a very common and important task in Natural Language Processing. In many domains and real-world settings, a few labeled instances are the only resource available to train classifiers. Models trained on small datasets tend to overfit and produce inaccurate results – Data augmentation (DA) techniques come as an alternative to minimize this problem. DA generates synthetic instances that can be fed to the classification algorithm during training. In this article, we explore a variety of DA methods, including back translation, paraphrasing, and text generation. We assess the impact of the DA methods over simulated low-data scenarios using well-known public datasets in English with classifiers built fine-tuning BERT models. We describe the means to adapt these DA methods to augment a small Portuguese dataset containing tweets labeled with smart city dimensions (e.g., transportation, energy, water, etc.). Our experiments showed that some classes were noticeably improved by DA – with an improvement of 43% in terms of F1 compared to the baseline with no augmentation. In a qualitative analysis, we observed that the DA methods were able to preserve the label but failed to preserve the semantics in some cases and that generative models were able to produce high-quality synthetic instances. [ABSTRACT FROM AUTHOR]

Optical character recognition (OCR) is typically used to extract the textual contents of scanned texts. The output of OCR can be noisy, especially when the quality of the scanned image is poor, which in turn can impact downstream tasks such as information retrieval (IR). Post-processing OCR-ed documents is an alternative to fix digitization errors and, intuitively, improve the results of downstream tasks. This work evaluates the impact of OCR digitization and correction on IR. We compared different digitization and correction methods on real OCR-ed data from an IR test collection with 22k documents and 34 query topics on the geoscientific domain in Portuguese. Our results have shown significant differences in IR metrics for the different digitization methods (up to 5 percentage points in terms of mean average precision). Regarding the impact of error correction, our results showed that on the average for the complete set of query topics, retrieval quality metrics change very little. However, a more detailed analysis revealed it improved 19 out of 34 query topics. Our findings indicate that, contrary to previous work, long documents are impacted by OCR errors. [ABSTRACT FROM AUTHOR]