The Information Extraction Systems of PRIS at Temporal Summarization Track

This paper describes the information extraction systems of PRIS at Temporal Summarization Track. The Temporal Summarization Track includes two tasks: sequential update summarization and value tracking. For the first task, we focus attention on keywords mining and sentence scoring. The system utilizes hierarchical Latent Dirichlet Allocation (LDA) to do keywords mining and score sentences with keywords shooting method. For the second task, we define the value extracting as a sequence labeling problem and build a discriminative undirected graph model (CRF model) to extract attribute values of all topics.