Labelling Topics in Weibo Using Word Embedding and Graph-Based Method
Zhipeng Jin, Qiudan Li, Can Wang, Daniel Dajun Zeng, Lei Wang · 2016
Nowadays, in China, Weibo is becoming an increasingly popular way for people to know what is happening in the world. Labelling topics is of much importance for better understanding the semantics of topics. Existing works mainly focus on deriving candidate labels by exploring the use of external knowledge, which may be more appropriate for well formatted and static documents. Recently, it has been a new trend to generate labels for sparse and dynamic microblogging environment using summarization method. The challenges of labelling topics are how to obtain coherent candidate labels and how to rank the labels. In this paper, based on the latest research work in deep learning, we propose a novel and unified model for labelling topics in Weibo, which firstly adopts word embedding and clustering method to learn dense semantic representation of topic words and mine the coherent candidate topic labels, then, generates interpretable labels using a graph-based model. Experimental results show that topics labels discovered by our model not only have high topic coherence, but also are meaningful and interpretable.