license: apache-2.0
This is a larger dataset of keyphrases that can be used for academic search. They were largely generated from over 1 billion ngrams, filtering out specific stop words, and each have count >= 5 in the original dataset. Additionally, the keyphrases you see here had a high score of tf-idf, where term frequency inverse document frequency has had "document" adjusted for academic subfields. The less subfields the phrase occured in, the higher the score. So, this dataset is a fairly large dataset of around the 3-4 million most important keyprases you'd encounter for searching for academic articles. It includes names of algorithms, niche subfields, scientific concepts, diseases, and so on.
Unigrams: 103.314 unique phrases
Bigrams: 378,041 unique phrases
Trigrams: 2,444,002 unique phrases
Fourgrams: 711,771 unique phrases