Reproduce “Weakly Supervised Concept Map Generation through Task-Guided Graph Translation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiaying Lu, Xiangjue Dong, Carl Yang
The Graph Translation-based Document To Graph (GT-D2G) framework presents a weakly supervised approach for generating concept maps from text, integrating linguistic analysis with a task-guided graph translation neural network. This method creates semantic-rich and concise concept maps that are adaptable in size and require less labeled data, leading to improved performance in downstream tasks like document classification.
Key insights
- Integrating robust linguistic preprocessing from traditional NLP pipelines is effective for creating semantically rich initial graphs, mitigating incompleteness in generated concepts.
- Weak supervision, derived from downstream task labels, can successfully guide a graph translation model to produce concise and task-relevant concept maps without requiring ground-truth concept map examples.
- Iterative, differentiable node selection combined with deeply coupled node and link generation, supported by techniques like Gumbel-Softmax, enables the flexible and high-quality construction of concept maps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.