Event Detection from Web Data in Chinese Based on Bi-LSTM with Attention.

Events are important activities people are involved in real life, and the information about events may be fascinating and important for people to understand and keep abreast with the key developments of some important social and individual subjects. In the big data era, event detection methods can help people efficiently and quickly extract specific information from massiveWeb information. However, the existing methods usually load the entire Web page information as the input into the models, and the rich noise and irrelevant information on Web pages will seriously impact the event detection performance of these methods. Also, the existingmethods mostly used staticmodels, which fail to consider the dynamics of information on theWeb. To improve the performance of event detection and classification, we propose in this paper a new method that partitions the Web pages into multiple text blocks and utilizes Bi-LSTM with the attention mechanism for fine-grained event detection from Chinese Web pages. We also propose a dynamicmethod that updates the data as well as themodel regularly and incrementally, making our modelmore adaptive to the ongoing changes of theWebpage data. The experimental results show that our model outperforms existing methods in event detection in terms of detection performance, the associated computational overhead, and the ability to deal with evolvingWebpage information.