The Popularity of Articles in PubMed

The PubMed search engine displays query results in reverse chronological order, which is appropriate for users interested in the latest publications. The purpose of this paper is to use machine learning to order documents by popular- ity, or the predicted frequency that an article is viewed by the average PubMed user. Other research on general search en- gine usability has applied machine learning to order documents by their relevance to a given query. The approach here takes a global view of popularity across all users in a given time period, independent of their information need. An effec- tive method for learning popularity from clickthrough data is identified, and a novel measure of success in this task is pro- posed. The resulting model shows that the topic of an article has the largest single influence on its popularity, and its pub- lication date has a strong secondary influence. Possible applications and extensions are discussed.