A hybrid approach for extracting arabic persons’ names and resolving their ambiguity from twitter

Abstract

Tweets offer a novel way of communication that enables users all over the world to share real-time news and ideas. The massive amount of tweets, generated regularly by Arabic speakers, has resulted in a growing interest in building Arabic named entity recognition (NER) systems that deal with the informal colloquial Arabic. The unique characteristics of the Arabic language make Arabic NER a challenging task, which, the informal nature of tweets further complicates. The majority of previous works addressing Arabic NER were concerned with formal modern standard Arabic (MSA). Moreover, taggers and parsers were often utilized to solve the ambiguity problem of Arabic persons’ names. Although, previously developed approaches perform well on MSA text, they are not suited for colloquial Arabic. This paper introduces a hybrid approach to extract Arabic persons’ names from tweets in addition to a way to resolve their ambiguity using context bigram patterns. The introduced approach attempts not to use any language-dependent resources. Evaluation of the presented approach shows a 7% improvement in the F-score over the best reported result in the literature.