4 ms·
Say I generate embeddings for a bunch of articles. Given the query "articles about San Francisco that don't mention cars" would cosine similarity uprank or down
by romanhn 2y ago
Say I generate embeddings for a bunch of articles. Given the query "articles about San Francisco that don't mention cars" would cosine similarity uprank or downrank the car mentions? Assuming exclusions aren't handled well, what techniques might I use to support them?
- mirekrusin 2y agoI think you have to separate it into negative query and run (negative) rank and combine results yourself.
- stared 2y agoIt is up for testing, but you likely get the effect of "don't think about a pink elephant." So I guess that for most embedding models, "articles about San Francisco that don't mention cars" are closest to articles about SF that mention cars. The fundamental issue here is comparing apples to oranges, questions, and answers.
- romanhn 2y agoSo is LLM pre/post-processing the best approach here?
- breadislove 2y agono this wont work. embedding models at the moment are pretty bad with negations