Topic or Metadata Modeling for Cross-Disciplinary Scholarship: Challenges and Opportunities for Academic Libraries

At the University of Notre Dame, we have been exploring automatic classification of texts via topic modeling and user-generated metadata to support cross-disciplinary scholarship. This effort originated in 2015 from a collaboration between the libraries and the Center for Civil and Human Rights to create an online comparative research tool to explore documents of Catholic social teaching and international human rights law. The library built the infrastructure for indexing, retrieving, and visualizing records while the researchers provided the controlled vocabulary and initial classification scheme. From the onset, the project team realized there were limitations with current library classification standards and practices. To provide satisfactory discovery for cross-disciplinary content, the group "crowdsourced" the controlled vocabulary task to researchers and students of each respective discipline. Through the selection of controlled vocabulary, initial hand-tagging, and a more robust topic modeling, the researchers provided semantic linking of similar or different concepts at full-text and paragraph level. The modeling disambiguated terms (i.e., the use of child - biological vs. child of God ) and bridged the gap between different disciplines description of equivalent concepts. For example, users can select the topic “solidarity/cooperation” and explore meaningful search results from the two fields about working together to improve human lives. The modeling enables a user from one discipline to overcome the problem of nuanced vocabulary in the other domain and, hence, uncover relevant information that might otherwise remain hidden within the context of current classification schema. The project team is currently reconciling issues of transparency by providing detailed documentation on the application of the controlled vocabulary and in the process of implementing features for crowdsourcing data to enhance classification. The paper will present the updates of our topic modeling endeavor, and provide insights on considerations of the scalability and sustainability for academic libraries to support cross-disciplinary scholarship.