Project MUSE adopts UNSILO AI tools

Project MUSE (

Project MUSE is one of the best-known academic collections for higher education. Founded in 1995 at Johns Hopkins University, MUSE is a not-for-profit collection of peer-reviewed content. The first books were added in 2010, and today it contains the content of over 800 journals, and more than 80,000 books, of which more than 4,000 are open-access titles. Recent innovations on the MUSE platform have confirmed its value, not just as a repository of content, but as a discovery tool.

MUSE launched the “enhanced content feature”, on its platform in June 2021 – its first full use of AI tools. Using the UNSILO Recommender links, MUSE users can now open most of the articles or book chapters and see related links to other content on the platform. Given that there are over 1.5 million articles and chapters in the collection, that’s quite a wide range of content to link to!

One of the benefits of the Recommender links is that they are continuously updated. As new content is added every day, the index of related articles is rebuilt to incorporate links to new content – there is no need for manual subject tagging, and no delay in waiting for subject codes and taxonomies to be updated.

Recommender links are just one way in which MUSE is making use of AI tools. In a news brief at the MUSE Meet 2022, Jennifer D’Urso and Angelia Ormiston of Project MUSE outlined their vision for making use of AI. Stated Ormiston: “There is no doubt that AI, NLP, and Text Analytics, and Machine Learning, will all be an integral part of whatever the future brings for the scholarly publishing ecosystem.”  Jennifer D’Urso commented: “Our analytics shows an uptick in site usage by MUSE users now we have the recommendations in place.”

How the Recommender works

Example of the related content links on the MUSE platform (click here to see it on MUSE)

Articles on MUSE are linked via concepts. As each new content item is made available on the MUSE platform, every chapter or article is examined and key concepts identified – hundreds of concepts for each content item. To identify related content, the system matches other articles that have overlapping concepts with the article being examined. This is done using a cluster of concepts, rather than the more traditional three or four subject keywords per article. By using a cluster of overlapping concepts, the system is able to find very precise matches to other content on the platform; in contrast, with a keyword-based system, links are inevitably very wide-ranging, such as “physics” or “chemistry”. One great advantage of this system is that content can be linked more widely than a manual classification system: an article on gender studies, for example, might have related content that was tagged “history”, or “ethics”. Concepts are not limited to a standard taxonomy – the screenshot above includes the concept “vulnerability”, something not likely to be included in standard classifications for arts subjects.

Longer term, the same concept matching can be used in a personalized way: if a MUSE user looked at three articles in the last week, the system can deliver recommendations for further reading based on those specific documents. MUSE is planning to trial this service in the coming months. As Jennifer D’Urso points out, this functionality is commonplace on consumer and e-commerce sites such as Netflix.

Formerly, MUSE used Library of Congress subject headings for its journal content; but this was labour-intensive and time-consuming. Moreover, the subject headings are only updated at intervals and are slow to change, and some of the terms used by LoC were no longer in accepted use. Angelia Ormiston commented: “After experimenting with various classification systems, we decided that AI was going to be way we need to move forward into a world beyond taxonomy, a world where language evolves, and so does the technology that is around it.”

Future developments

Continued Angelia: “MUSE has begun looking into transforming our platform using AI. Starting with the small features like related content, we plan to continue to evaluate all of the features on our front end. We also plan to look at our metadata systems, our sales practices, and the overhaul of the production system, but most excitingly for me is the MUSE search. MUSE is ready to innovate our search once again. The combination of linked data and AI will allow the creation of semantic links without manual workflows to create them. This will offer a search experience that can provide information connected to content before the content has ever been read by the user.”

We look forward to seeing these new types of content collecting and discovery in use on the MUSE platform.

Receive an email every time we publish a new blog post