Dimensions uses article-level classification instead of journal classification due to the difficulties and inaccuracies that journal level classifications bring. For example, many broad-scope or multidisciplinary journals are almost impossible to logically categorize into a small number of relevant categories, and although many other journals may only have a small variation in subject between publications, an article level categorisation system will always be preferable if the categories are to be accurate.
Dimensions uses document-level categorisation based on automatic classification. This automatic classification is carried out based on noun phrase extraction and machine learning based emulations of standardised classification systems. All of this means that single documents can be rapidly classified, without compromising on accuracy.
Fields of Research classification and article/journal classification
The Fields of Research (FoR) classification system includes categories for all areas of research, and is the base level classification system in Dimensions present in all Dimensions versions. Categorising documents at the single article level is the best approach for accurate search results, however classification requires a minimum amount of text from each document in order to assign a category, which can lead to many publications not receiving a classification if not enough information exists. Due to this, Dimensions no longer only uses automatic classification of individual documents, but in cases where documents cannot be classified individually due to lack of information, we will now assign an FoR code based on journal level classification of the journal from which the document came (if available). This system therefore includes the best possible situation - using both article level and journal level information to decide on the categorisation of documents used.