Weakly Supervised Concept Map Generation through Task-Guided Graph Translation

Recent years have witnessed the rapid development of concept map generation techniques due to their advantages in providing well-structured summarization knowledge from free texts. Traditional unsupervised methods do not generate task-oriented maps, whereas deep generative models require large amounts training data. In this work, we present GT-D2G (Graph Translation-based Document To Graph), an automatic framework that leverages generalized NLP pipelines derive semantic-rich initial graphs, and translates them into more concise structures under weak supervision downstream task labels. The maps generated by can provide interpretable structured for input texts, which are demonstrated through human evaluation case studies on three real-world corpora. Further experiments document classification show beats other methods. Moreover, specifically validate labeling efficiency label-efficient learning setting flexibility graph sizes controlled hyper-parameter studies.