RANER: RDI Framework for Arabic Named Entity Recognition

Abstract

Named Entity Recognition (NER) task allows NLP applications to extract proper names from text. In addition, a significant component affects the performance of other NLP tasks such as text summarization, topic detection and key phrase extraction. Although many researches are conducted to enhance the NER task, only limited researches in Arabic named Entity Recognition have been performed. Researches in this field use either rule-based approach or machine learning approach. This paper introduces a solution for the NER problem using a machine learning approach, which combines Conditional Random Fields (CRF) classifier and predefined gazetteers. Our system uses syntactic features and morphological lookup-table features to train the classifier. This features extraction approach saves the time of morphological features that depend on analyzers without affecting the precision of the system. We evaluate our system by way of experiments using different sets of features to extract three of main named-entity types; Person, Location, and Organization. Experimental results showed that our approach has achieved better performance that rule-based approaches. Our system achieved a performance of 70.7%, 90.0%, and 79.2% in term of recall, precision, and F-measure, respectively..