Relation Extraction from Arabic Wikipedia

摘要

Objectives/methods: This study aims to extract relations between entities from the Arabic text. RelationExtraction is one of the most important tasks in text mining. Relation extraction is considered as themain step for many applications such as extracting triples from the text, Question Answering and Ontology building. However, extracting relations from the Arabic text is a difficult task compared to English due to a lack of annotated Arabic corpora. This study proposes a method for extracting relations from Arabic text based on ArabicWikipedia articles characteristics.The proposed system extracts sentences that contain principle entity, secondary entity and relation from Wikipedia article, then we use WordNet and DBpedia to build the training set. Finally,the Naive Bayes Classifier is used to train and test the datasets. Finding: There are few works to extract relations from Arabic text. These works depend on classification, clustering and rule-based. Application/improvement:The experiments show the effectiveness of the proposed approach which achieves high precision with 89% for classifying 19 types of semantic relations.