| News report is one of the most important information source from the enormous data coming from the internet due to its formality,objectivity and reliability.Automatically mining hot news topics by information technology and showing the content hot news topics from all kinds of perspective by visualization technology is a very important research area.Although,many approaches have been proposed to mine hot news topics.However,current works in hot news topics mining fail to discover three fundamental relevance among hot news topics,i.e.the structural relevance,the temporal relevance and semantic relevance.We propose a novel hot news topics mining algorithm that has the ability not only to mine the hot news topics but also to discover the three types of relevance among them.Our model is motivated by the topic models like Latent Dirichlet Allocation(LDA)are effective models for mining hot news topics without the need of large amount of labeled data.What's more,we propose a new visualization technique to visualize the structural,temporal and semantic relevance among hot news topics and provide a great tool to explain and analysis news data.The contributions of this dissertation are as follow:1)Propose a topic model to discover the structural relevance among hot news topics and a novel visualization layout for structural relevance;2)Propose a novel model to discover temporal relevance among hot news topics and a dynamic visualization technique to visualize the evolution of hot news topics;3)Propose a embedding model to discover the semantic relevance among the hot news topics and a projection based approach to visualize the semantic relevance.4)Integrate the proposed mining and visualization techniques as a hot news topic analysis system.This research has been applied in 973 national project "Cross-media Computing for Public Safety:Theory and Applications" and is used for mining news topics and their relevance as well as visualizing these relevance and the news corpus. |
PDF Full Text Request