rjurney/iecompare repository overview

In this post dated January 31, 2025, I will compare and contrast three new models for information extraction to build a knowledge graph about GraphFrames. I’ll be evaluating each tool as components in a knowledge graph construction pipeline [my whole reason for being]. As we’ll see, the approaches of the three approaches vary. Two are custom transformer models and one is a fine-tuned LLM. I am particularly interested in how custom transformer models compare with NuExtract’s fine-tuned LLM for zero-shot prompting. Note that complete and runnable source-code for this post is available, along with a docker image to try all three models.