Open ccstan99 opened 1 year ago
Embedding the summary keys might help with this. We can generate synthetic summaries for those without human summaries. AllenAI's SPECTER embeddings were trained to match title & arxiv abstracts. To use it we'd need separate pinecone index & namespace since its a SentenceTransformer so embedding dimensions don't match OpenAI's.
Support paper/blog search answering questions like: What are some recent/new/popular papers/blogs? What was the paper/blog about X, by Y, from Z?