Research on an automatic retrieval method for special topic news based on semantic frame

A novel negative news retrieving semantic frame(NNFrame) and its identification were presented. Different from traditional semantic FrameNets, which were defined based on word-sense disambiguation, NNFrames were defined on each subcategory of negative news with a single semantic context. By constructing the NNFrame knowledge base, domain ontology repository, and a collection of annotated example sentences for each NNFrame, a method was described for identifying NNFrame by a task-specific extended conditional log-likelihood model, that takes dependency-syntax structure representations, and the part of speech tags as input. This approach is practical, efficient, and can achieve state-of-the-art results on precision/recall metrics for identification and classification of negative news whose subcategories are pre-defined in the NNFrame knowledge base.