The merits of Universal Language Model Fine-tuning for Small Datasets -- a case with Dutch book reviews (1910.00896v1)

Abstract: We evaluated the effectiveness of using LLMs, that were pre-trained in one domain, as the basis for a classification model in another domain: Dutch book reviews. Pre-trained LLMs have opened up new possibilities for classification tasks with limited labelled data, because representation can be learned in an unsupervised fashion. In our experiments we have studied the effects of training set size (100-1600 items) on the prediction accuracy of a ULMFiT classifier, based on a LLMs that we pre-trained on the Dutch Wikipedia. We also compared ULMFiT to Support Vector Machines, which is traditionally considered suitable for small collections. We found that ULMFiT outperforms SVM for all training set sizes and that satisfactory results (~90%) can be achieved using training sets that can be manually annotated within a few hours. We deliver both our new benchmark collection of Dutch book reviews for sentiment classification as well as the pre-trained Dutch LLM to the community.