Best Match in Pubmed

The default sorting order in Pubmed is with the most recent articles first, but you may have noticed that there is also an option to sort by relevance, Best Match?

Best Match is useful when you quickly want to find a few relevant references. It can also be a great way to start a more exhaustive search. In most cases, the sorting will make relevant references appear in top of the list. But keep in mind that you are always responsible for you own selection of articles. Do not rely solely on the sorting of Pubmed, but make your own choice.

But Best Match is actually not only a sorting option, but also affects how the search is performed. If you repeated the same search but got a different number of hits the reason might be that you used different sorting options. Best Match can give either more hits or fewer hits than Most Recent. There may be unique references in both searches. This means that you can’t be sure that all references in the search with the lower number of hits are included in the search with more hits.

A Best Match search is based on these parameters:

  • how many times the search words appears in the documents;
  • how frequently the search words appear in Pubmed’s references in general (an uncommon word is given more weight than a common word that appears in many references);
  • in which fields the search words appear (for example a word that occurs in the title is given more weight than if it appears in the abstract);
  • how much of the field the search words uses (a search word in a short title or abstract is given more weight than a search word in a longer title or abstract since it is a bigger share of the field).

Different search words will be given different weight in the search based on these factors. In addition, a machine learning algorithm is applied to the search. The algorithm is based on a large number of logged searches in Pubmed. All this means that the standard boolean search logic like AND, OR, and NOT is not fully applied. At the top of the list of result, the document will contain all the keywords you combined with AND, but at the end of the list, the search words that Pubmed assigned a low weight might be missing.

If you are doing a structured search where you plan to review all references in the search result, the sorting option Most Recent is preferable. With his option you know how the search was performed in Pubmed, what search words you used and how they were combined. In a structured search, you are expected to have this control over the search. In a Best Match search, you do not know exactly how the search is performed.

Another situation when Most Recent is the best option is if you are going to present your search strategy, for example in a degree project or a thesis. The machine learning in Best Match means that the search and sorting may change, which means that the search is not fully reproducible. You can read more on the Best Match sorting and searching here. On the same page, you can also comment on how Best Match works.

Magdalena Svanberg (on leave)

Librarian who teaches and tutors students at all levels. Also part of the search group at KIB.