Named Entity Recognition of Persons' Names in Arabic Tweets

The rise in Arabic usage within various social media platforms, and notably in Twitter, has led to a growing interest in building Arabic Natural Language Processing (NLP) applications capable of dealing with informal colloquial Arabic, as it is the most commonly used form of Arabic in social media. The unique characteristics of the Arabic language make the extraction of Arabic named entities a challenging task, to which, the nature of tweets adds new dimensions. The majority of previous research done on Arabic NER focused on extracting entities from the formal language, namely Modern Standard Arabic (MSA). However, the unstructured nature of the colloquial language used in tweets degrades the performance of NER systems developed to support formal MSA text. In this paper, we focus on the task of Arabic persons‟ names recognition. Specifically, we introduce an approach to extract Arabic persons‟ names from tweets without employing any morphological analysis or languagedependent features. The proposed approach adopts a rule-based model combined with a statistical one. This approach uses unsupervised learning of patterns and clustered dictionaries as constrains to identify a person‟s name and resolve its ambiguity. Our approach outperforms the best reported result in the literature on the same test set by an increase of 19.6% in the F-score.