arXiv:2606.29571v1 Announce Type: new Abstract: The standard way to compare two text embeddings is cosine similarity. Scattered studies report that a different metric does better, but never pin down the geometric condition that decides when, or why. We settle both with a comprehensive empirical study: nineteen parameter-free similarity metrics on nineteen encoders, from compact sentence transformers up to seven-billion-parameter large language models, across seven datasets. The answer is geometric. When an encoder spreads its variance evenly across directions, cosine is the best parameter-free
The proliferation of various text embedding models and their applications necessitates a deeper understanding of optimal metric choices for performance and efficiency.
A refined understanding of text embedding comparison metrics can lead to more accurate AI systems and more efficient development cycles, impacting various applications of large language models.
The explicit identification of geometric conditions (anisotropy) dictating the choice between cosine similarity and rank metrics provides a clearer guideline for AI researchers and practitioners.
- · AI researchers
- · NLP developers
- · Large language model companies
- · Developers using suboptimal similarity metrics
- · Systems built on less accurate text comparisons
Improved performance and accuracy in AI applications relying on text embeddings.
Faster development and deployment of robust natural language processing (NLP) systems due to clearer metric selection guidance.
Potential for new embedding architectures or fine-tuning approaches optimized for specific geometric properties identified in this research.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL