Dynamic-keyword extraction from social media

Abstract

Traditional keyword extraction methods make the assumption that corpora is static. However, in social media, information is highly dynamic, with individual words showing a dynamic behaviour. In this paper we propose an unsupervised approach that jointly models words’ temporal behaviour and keyword’s semantic affinity, to address the task of dynamic-keyword extraction. Experiments show the method effectiveness and confirm the importance of exploiting keyword dynamics.