Recommender systems are now a common occurrence when navigating popular websites. E-commerce sites suggest "related products" to our recent searches, media sites link to "similar pictures" and social networking platforms point out "people you may know". These systems have seen less integration into platforms hosting digital archives. As digitized archives grown in both size and complexity, some automated curation and dynamic guidance is needed for users to explore the breadth of potential offerings. In this talk, I will explore the potential benefits and pitfalls of using recommender systems within digital archives. Of specific interest are how extant machine learning techniques must be adapted in order to reach a wider public audience. I will also address the dual nature of privacy concerns in this process: collecting more user data has the potential to highlight better ways of reaching a more diverse audience but also forces one to make the learning process more closed and less public. The discussion will use several recent projects to illustrate the broad set of questions that must be answered, ranging from back-end architecture to the front-end user interface.
Copyright. All Rights Reserved.