Name Entity Recognition and Binary Relation Detection for News Query

Abstract

This paper proposes an innovative slant on name entity recognition (NER) and binary relation detection (BRD) for current news search and retrieval. The main contribution is to apply MITIE tools to find relevant news according to the given news stories. BRD can retrieve relevant news according to the relationship between given query term and news dataset. Another salient contribution is to accomplish news story agency classification with Stanford classifier and automatic document classification tool WEKA. Given training data, many news stories with agency class labels, WEKA can predict and identify the test news stories’ class labels. The evaluation of relevant news is based on MAP@100 in 20 queries given 160000 news story dataset. The retrieved results of all the queries can demonstrate up to 100 relevant stories in terms of theirs IDs. The evaluation of news agency classification is based on accuracy with correct classified news stories. Scoring with NER yields the better performance.