An Evaluation Research of Topic Extraction Results Based on External Metadata

With the increasing scale of literature publication and the rapid growth in the number of publications, topic extraction methods have also made great progress, yet the extent to which the topic extraction results reflect the true subject structure of the scientific literature remains an open question. In the development of topic extraction, the validity of topic extraction results has been an issue of concern to scholars. This paper contributes to validating the topic extraction method by using an external validation method to evaluate the topic extraction results. Considering that each piece of literature may involve more than one research topic, this thesis completes topic extraction using four overlapping community detection algorithms, COPRA, DEMON, SLPA, and GCE, based on direct citation network construction. The topic extraction results were then investigated for validation with the help of external metadata as the validation dataset.