| Recently, due to the wide spread of the network technology and the rapid development of the network, speed of information dissemination and quantity has reached on an unprecedented scale. As a new online media, Microblog has gradually become an important source of information on the Internet. Since the release content of microblog is very simple and it can be posted on various terminals, resulting a lot of data in a short period of time on the microblog platform. While dealing with those huge&messy microblog data, it is a considerable workload. Besides, it is difficult to provide timely and accurate information. Hot topic detection can deal with the merge of the information. It can help users quickly understand the information they need.Traditional hot topic detection technology is based on the most widely used space vector model, which achieved good results, but it is still insufficient when dealing with microblog’s short texts. Traditional cluster methods mainly focus on semantic similarity of words only. Firstly this paper proposes an algorithm based on comprehensive similarity (semantic similarity and context similarity). The experiments on real world microblog data have shown the effectiveness and efficiency of the approach to detect hot topics.Secondly Latent Dirichlet Allocation(LDA) is extended to Microblog Latent Dirichlet Allocation (MLDA). An MLDA model is proposed in this paper, which takes microblog document relation, topic tag and authors’ relationship into consideration. The topic-word and document-topic distributions are inferenced by incremental Gibbs algorithm.Thirdly the paper focuses on the research of the topic evolution mode. The topic evolution in content and intensity are analyzed. The experiments on microblog data have shown the effectiveness and efficiency of the approach to depict topic evolution. Finally, the research works mentioned above are implemented. Experiments on real data are performed to validate the effectiveness of the works as well. |
PDF Full Text Request