Synonym ring construction as a means of enhancing the discoverability of the Digital Humanities subcollection

This thesis documents my Digital Humanities Commons internship at KU Leuven Libraries Artes. The main objective of my internship project was to build synonym rings that would make a subcollection of eBooks on the topic of Digital Humanities more discoverable in the KU Leuven Libraries’ online library catalog. Synonym rings are lists of words that should be considered equivalent during the process of retrieving information from a system. For example, a synonym ring might consist of the words “astronaut” and “cosmonaut.” If a user searches for the word “cosmonaut,” it is reasonable to assume that they are equally interested in retrieving documents that include the word “astronaut,” even though this word was not explicitly added to their query. This is why systems that make use of synonym rings “scan” users’ queries for words that are included in these synonym rings and thereby expand the original queries in order to retrieve a larger quantity of relevant information. The significance of using synonym rings in this way and specifically in the context of this project lies in its ability to help meet users’ information needs, which is one of the core missions of a library and its staff. I built my synonym rings mainly with the help of a Natural Language Processing technique called word2vec. To test the synonym rings that I created, I built a proof-of-concept prototype that mimicked the online library catalog environment, where I deployed the synonym rings that I created. After evaluating the prototype’s performance using two of the most popular measures for evaluating information retrieval systems (i.e., precision and recall), I determined that the rings had a positive effect on the prototype and subsequently I deployed them in KU Leuven Libraries’ online catalog. A second experiment was performed to evaluate the success of the rings in KU Leuven Libraries’ online catalog environment. Again, the precision and recall scores improved considerably after adding the synonyms. Although this method is proven useful for making a subcollection of books more discoverable, I additionally argue that metrics such as precision and recall can, at times, be misleading. Hence, reaching definite conclusions on its performance necessitates further research that evaluates how satisfied the users are with the system. Ultimately, what matters is whether the information needs of a library’s users are being satisfied. Finally, I argue that while it is important to ensure a search experience that is as effective as possible, providing ways of browsing the material of a collection should not be neglected. Browsing usually does not require extensive training and might lead patrons to serendipitous discoveries before they are able to express their information needs in the right words. In this project, an interactive visualization-based browsing experience was proposed. Hopefully, the results of this project will help the KU Leuven Libraries staff decide if constructing synonym rings would help make their collections more discoverable. The current way of adding synonyms rings to the online library cataloging software used by KU Leuven Libraries is laborious and library professionals should weigh whether the advantages exceed the disadvantages.