DySem: Uncovering Dynamic Semantic Components of Large Language Models for Calculating Semantic Textual Similarity
DYSEM proposes a training-free method to calculate semantic textual similarity by dynamically selecting relevant internal components of large language models (LLMs). Instead of using fixed dimensions, it extracts sample-specific semantic dimensions across multiple languages using multilingual consensus. This allows the model to compute similarity over a shared, flexible subset of dimensions that captures robust semantic knowledge.