Comparative analysis of ML POS on Arabic tweets

One of the challenges of natural language processing is social media text like tweets. Conversational text in contrast to genres that are highly edited (standard language) which traditional NLP tools have been developed for contains many syntactic patterns and non-standard lexical items. These are the outcomes of dialectal variation, diversity in topic, orthography, unintended errors, conversational errors and creative language use. The fact that twitter text is characterized by idiosyncratic style, noise and linguistic errors makes it difficult to part-of-speech tag. The aim of this paper is to design and implement models of speech tagging for Arabic tweets by investigating numerous models of machine learning like K-Nearest Neighbour, Naïve Bayes and Decision tree models. In this paper, a novel Arabic Twitter corpus is introduced while assessing various state-of-the-art POS taggers which retrained on the given corpus. A state-of-the-art accuracy of 87.97% is achieved when tagging twitter.