I'd go to the card catalog (index), turn my question into a bag of words (tokenize), fetch all the cards matching each token (posting lists), drop cards which d

I'd go to the card catalog (index), turn my question into a bag of words (tokenize), fetch all the cards matching each token (posting lists), drop cards which didn't include enough of the tokens (posting list intersection), ordering the cards by the number of tokens they matched (keyword match ranking), filter at some cutoff, and then reorder based on the h-index of the author (page rank). Then I would read each paper in order, following citations in a breadth-first manner. (the above is a joke comparing old school library work to search engines circa 2000; I didn't actually do all those steps. I'd usually just find the most recent review article and read the papers it cited).