Recently, social media and specially Twitter has become a main source for news consumption and sharing
among millions of users. Those platforms enable users to author, publish and share content. Such
environments can be used to publish and spread rumors and fake news whether unintentionally or even
maliciously. That is why credibility of information in such platforms has been increasingly investigated in
many domains (i.e. information sciences, psychology, sociology...etc). This paper proposes a machine
learning - based model for Arabic news credibility assessment on Twitter. It uses hybrid set of features that
are topic and user related to evaluate news credibility. In addition to the traditional content-related features,
Content verifiability and users' replies polarity analysis used for a more accurate assessment. The proposed
model consists of four main modules: a) content parsing and features extraction module, b) content
verification module, c) users’ comments polarity evaluation and d) credibility classification module. A data
set of 800 Arabic news that are manually labeled is collected from Twitter. Three different classification
techniques were applied (Decision tree, support vector machine (SVM) and Naive Bayesian(NB). For
model training and testing, 5-fold cross validations were performed and performance diagnostics were
calculated. Results indicate that decision tree achieves TRP higher than SVM by around 2% and 7% than
NB, also FPR almost 9% lower than SVM and 10% lower than NB. For precision,recall, f-measure and
accuracy, decision tree achieves almost 2% higher than SVM and 7% higher than NB for the tested dataset.
Experiments also revealed that the proposed system achieves accuracy that outperforms the system
proposed by Hend.et.al [29] and TweetCred [2]. |
