SemSim (Semantic Textual Similarity) is an important evaluation metric for text rewriting / paraphrasing tasks in natural language processing. Semantic similarity directly measures how close the rewritten sentence is to the original sentence in the semantic space, and is a core requirement for rewriting quality. It is typically reported as a percentage or a score from 0–1. This metric is widely used in relevant benchmarks and papers, and effectively reflects model performance on this dimension. Higher values are better. It helps researchers and engineers quantify differences in actual model performance in text rewriting / paraphrasing scenarios, facilitating side-by-side comparison of different methods or model versions.