MULTI-FACETED AUTOMATIC CLASSIFICATION OF INSTITUTIONAL REPOSITORY DIGITAL OBJECTS: A CASE STUDY OF THE UNIVERSITY OF ZAMBIA

Institutional Repositories (IRs) provide the ability to store, manage, and disseminate intellectual products created by an institution. They provide a complementary method to the traditional system of scholarly communication, making it easier to demonstrate the scientific, social, and financial value of an institution. The potential benefit of an IR goes beyond the desire to increase an institution’s profile. They also increase authors’ visibility and provide users with easy access to information. Despite the hasty pace at which organizations are creating IRs and with all the potential benefits they offer, recent studies have established that the two biggest existing problems are that of digital objects having missing important metadata elements and the wrong classification of digital objects into communities. This research outlines a case study conducted at the University of Zambia (UNZA). The aim of this study was to design, develop, and implement three classification models and a prototype tool that uses the models for effective ingestion of digital objects into an IR. To achieve this, firstly, a situational analysis was conducted to appreciate the challenges of the current system being experienced in the tagging and ingestion of digital objects into the IR. Furthermore, an exploratory study was conducted in order to assess the full extent of the problem. Finally, three classification models were implemented. Experiments on classification using the developed models were conducted, and the results demonstrated the possibility of automatically classifying digital objects into an IR with an accuracy of 77% for the collection classification model, 75% for the document type model, and 0.005% Hamming loss for the subject classification model. The results suggest that our proposed technique can help address the two biggest existing problems related to IRs.