KeyBERT is an algorithm used in semantic analysis to efficiently extract the core, semantically most important N-gram (single word) and Bigram (two words) from a keyword or query. It provides a key mechanism for succinctly labeling keywords within a universe. By grouping keywords into themes based on the core terms identified by KeyBERT, marketers can gain insights into high-level categories and user intent surrounding those topics. This process is useful for analyzing core entity and sub-entity relationships across the dataset.