| Internet has greatly changed the people's habits of browsing news, now more and more people tends to browsing news on the web. This paper has investigated on how to find information in which users are interested from huge amount of news and news comments.First, news clustering technologies are used to better organize the news according to events contained by the news. In this paper, news clustering problems are defined as text clustering problems with different kinds of features together, and two widely used algorithms, K-means and agglomerative hierarchical clustering are used to solve the problems. A new clustering method based on Dirichlet Processes are also introduced to news clustering. Furthermore, the above three clustering methods are extended to handle more kinds of features. For clustering methods based on Dirichlet Process, a new method named Multimodal Dirichlet Processes Sets is proposed.The paper also defined the extraction of opinion target candidates from the news clusters as a named entity recognition problem, and proposed a fast algorithm for extracting named entities in dicourses. Named entities contained in news articles are also treated as a new kind of features for news clustering.Based on the extracted opinion target candidates, in this paper, a target-dependent sentiment analysis method based on the results of dependency parsing is proposed for news comments sentiment analysis. A rule-based and a statistical based target-dependent methods are proposed to help re-organize news comments according to the opinions the comments hold and the targets of the opinions. Experiments show that the clustering methods and target-dependent sentiment analysis methods can improve the performances of the tasks in this paper. |
PDF Full Text Request