This system allows for the extraction of keywords from any sentence based on lexical centrality theory. This is the version with Lunaris 1.5 (why not 2.0, since we want to do word renking here, and version 2 isn't suitable for that). So here, the model is deterministic, meaning it doesn't depend on the context for its calculations, but it's local to the word level (like a TF-IDF or BM25). This approach allows it to be fast (O(n) versus O(n^2) for the version with attention). I find this compromise to be the best; admittedly, it might be slightly less qualitative than the version with attention, but the speed is well worth it. AmetrineOS - Lunaris Keyword Extractor Attention Head : https://scratch.mit.edu/projects/1378300424/
