Web event evolution trend prediction based on its computational social context

Abstract

Predicting future trends of Web events can help significantly improve the quality of Web services, e.g., improving the user satisfaction of news websites. Existing approaches in this regard are based mainly on temporal patterns mined with the assumption that enough temporal data is available on hand. However, most Web events do not have a long lifecycle, but a burst property, which drastically reduces the performance of temporal patterns mining. Furthermore, these approaches overlook the influence of the social context surrounding the Web events. In this paper, we propose a novel method to predict future trends of Web events, based on their social contexts rather than temporal patterns. More specially, in the proposed method, a computational model for the social context is first built as a two-layer Association Linked Network considering its properties, such as the associative network property and the small world property. Then, the interaction between a Web event and the social context is simulated, based on the anchoring theory. Finally, an external force is defined and evaluated to quantify the influence of the social context on the evolution of Web events, which is used to predict future trends of Web events. Experiments show that the performance of the proposed method is better than that of the traditional time series-based approaches.