Affiliations: [a] School of Computer Science and Engineering, South China University of Technology, Guangzhou, China. E-mails: [email protected], [email protected], [email protected] | [b] Winhong Corporation, Guangzhou, China
Abstract: In order to better demonstrate the evolution relationships between the events from newswires and to improve the readability of the event evolution graphs, we propose an improved news event evolution model from a view of users’ reading willingness. The model discusses two factors that affect the willingness of users’ reading, including the comprehensiveness of news information and reading cost. We define the cost function of user’s reading to determine the granularity of news events. After classifying the news stories by K-means clustering algorithm, this model takes the general structure of the news reports into consideration to calculate the TF-IDF weights and does some correction as well as model fusion. Finally, the parameters of the model are estimated by genetic algorithm based on Levy flight. By generating a more readable event evolution graph, our model is more capable of discovering the evolution relationships between the News events. We carried out experiments to evaluate the performance of our proposed model. The result shows that the proposed model outperformed the baseline and other comparable models in previous work by about 13% in the corpus we collected from the CNN & ABC News websites.
Keywords: Event evolution, term frequency–inverse document frequency (TF-IDF), reading willingness, K-means, information retrieval
