Combining human input with machine learning: A case study

Key points Publishers increasingly wish to create subject collections but need automated systems to carry this out in a cost‐effective fashion. The capabilities of unsupervised machine‐based tools to create subject collections have improved dramatically. Experts are still likely to find errors and exceptions in automated unsupervised content extractions and need ways to interact with semantic tools. Many semi‐automated content solutions require highly skilled and expensive technical staff to develop and maintain them. Karger's use of UNSILO's Package Manager appears to have improved efficiency and also fed back to UNSILO to help improve the product.