Tan Hongye, Zhao Tiejun, Zheng Jiaheng
Jul 8, 2008
2008 IEEE 8th International Conference on Computer and Information Technology Workshops
Event detection and recognition is a major task in ACE evaluation plan. In this paper, we focus on solving the two subtasks: (1) event detection and classification, (2) their argument role identification. For the first subtask, the strategy of local feature selection and explicit discrimination of positive and negative features is used in order to ensure the performance of each type. For the second subtask, the approach based on multi-level patterns is presented in order to improve the coverage of patterns and to use various language information. Experiments on the ACE2005 corpus show that performance of the first subtask is satisfying with the 83.5% macro-average F1-measure. And experiments of the second subtask show that the method based on multi-level patterns is very promising.