Predicting social annotation by spreading activation

Abstract

Social bookmark services like del.icio.us enable easy annotation for users to organize their resources. Collaborative tagging provides useful index for information retrieval. However, lack of sufficient tags for the developing documents, in particular for new arrivals, hides important documents from being retrieved at the earlier stages. This paper proposes a spreading activation approach to predict social annotation based on document contents and users' tagging records. Total 28,792 mature documents selected from del.icio.us are taken as answer keys. The experimental results show that this approach predicts 71.28% of a 100 users' tag set with only 5 users' tagging records, and 84.76% of a 13-month tag set with only 1-month tagging record under the precision rates of 82.43% and 89.67%, respectively. © Springer-Verlag Berlin Heidelberg 2007.