Generating Hierarchical JSON Representations of Scientific Sentences Using LLMs

1. 1Collect scientific sentences from articles
2. 2Fine-tune a lightweight LLM using a novel structural loss function
3. 3Generate hierarchical JSON structures from the sentences
4. 4Use a generative model to reconstruct the original text from the JSON structures
5. 5Evaluate the similarity between the original and reconstructed sentences

NLP researchers and AI engineers on a team benefit from this approach as it enables the creation of structured representations of scientific text, which can be used for various downstream tasks

💡 Structured representations can preserve the meaning of scientific sentences

📚 Generate hierarchical JSON reps of scientific sentences using LLMs! 🤖