Automatic Cause-Effect Relation Extraction from Dental Textbooks Using BERT

The ability to automatically identify causal relations from surgical textbooks could prove helpful in the automatic construction of ontologies for dentistry and building learning-assistant tools for dental students where questions about essential concepts can be auto-generated from the extracted ontologies. In this paper, we propose a neural network architecture to extract cause-effect relations from dental surgery textbooks. The architecture uses a transformer to capture complex causal sentences, specific semantics, and large-scale ontologies and solve sequence-to-sequence tasks while preserving long-range dependencies. Furthermore, we have also used BERT to learn word contextual relations. During pre-training, BERT is trained on enormous corpora of unannotated text on the web. These pre-trained models can be fine-tuned on custom tasks with specific datasets. We first detect sentences that contain cause-effect relations. Then, cause and effect clauses from each cause-effect sentence are identified and extracted. Both automatic and expert-rated evaluations are used to validate the efficacy of our proposed models. Finally, we discuss a prototype system that helps dental students learn important concepts from dental surgery textbooks, along with our future research directions.