Back in the day we called this the schema matching problem. I wrote a heuristic aided by learning model for epinions that tackled this in the space of product catalogs where you had to determine that a vendors “palm pilot with pen” was different than a “palm pilot pen.” It’s a very difficult problem space that could probably have been helped substantially by fine tuned LLMs. Instead we took our low confidence matches and shipped them to a company in India called Ugam that specialized in matching product catalogs. I was surprised there preexisted entire companies dedicated to the problem.