Can reindex search results by lemmatizing (with NLP) all the text of a result and then intersecting the embedding cosine distances (with an embeddings model) between the tokens. I am planning to do that for news article keyword distillation, one day, when I'm not swamped.