Deep Learning for Bibliographic Catalogue Assisting System

Abstract

Academic libraries play a major role in providing the information and resources to support formal and informal learning. In order to provide the circulation service, librarians have to deal with the cataloguing process after acquisition. Cataloguing has been a major workload process that requires the intellectuals of librarians. With different experiences of the librarians and the complexity of the content, the quality of cataloguing information and the time spending is out of control. This study developed a catalogue assisting model to reduce the bottleneck of assigning subject access fields in bibliographic records which presumed as the most difficult task in the cataloguing process. The Neural Network models were built by applying the words appearing in the title and table of contents of bibliographic records as the input and predict the list of suggested subjects. The performance of the models was evaluated through the value of precision, recall, and the percentage of bibliographic records that correctly assigned at least 1 subject. The experimental results suggested that combining the suggested subject list obtained from the title word and table of content word models provides better results than using only an individual model.