How are keyphrases calculated?

SciVal uses the Elsevier Fingerprint Engine to extract distinctive keyphrases within an entity.

The Elsevier Fingerprint Engine uses text mining and applies a variety of Natural Language Processing techniques to the titles, abstracts and author keywords of the documents in the Research Area, Publication Set, Topic or Topic Cluster in order to identify important keyphrases.

Keyphrases are obtained by annotating content with the OmniScience thesaurus – an in-house developed single unified thesaurus spanning all major disciplines to create a list of standardized concepts. OmniScience thesaurus consists of more than 50,000 manually curated core concepts and more than 700,000 semi-automated created extension concepts which are more fine-grained, both are updated quarterly.

For each document we take the list of standardized keyphrases and select which ones are important based on Inverse Document Frequency (IDF). This technique incorporates a factor that diminishes the weight of words that occur frequently in the set of documents and increases the importance of words that occur rarely. Each keyphrase is then given a relevance between 0 and 1 with 1 given to the most frequently occurring keyphrase and the ones occurring in key sections (Title and Author keywords). Remaining keyphrases are given a value based on their relative frequency. In SciVal we take a weighted list of keyphrases per publication – a fingerprint, and aggregate that up to different entity levels (i.e. a Research Area or Topic).