ArSentD-LEV: A Multi-Topic Corpus for Target-based Sentiment Analysis in Arabic Levantine Tweets

Abstract

Sentiment analysis is a highly subjective and challenging task. Its complexity further increases when applied to the Arabic language, mainly because of the large variety of dialects that are unstandardized and widely used in the Web, especially in social media. While many datasets have been released to train sentiment classifiers in Arabic, most of these datasets contain shallow annotation, only marking the sentiment of the text unit, as a word, a sentence or a document. In this paper, we present the Arabic Sentiment Twitter Dataset for the Levantine dialect (ArSenTD-LEV). Based on findings from analyzing tweets from the Levant region, we created a dataset of 4,000 tweets with the following annotations: the overall sentiment of the tweet, the target to which the sentiment was expressed, how the sentiment was expressed, and the topic of the tweet. Results confirm the importance of these annotations at improving the performance of a baseline sentiment classifier. They also confirm the gap of training in a certain domain, and testing in another domain.

ARBERT & MARBERT: Deep Bidirectional Transformers for Arabic

“…evaluate AraBERT on 3 Arabic downstream tasks. These are (1) sentiment analysis from 6 different datasets: HARD (Elnagar et al, 2018), ASTD (Nabil et al, 2015), ArsenTD-Lev (Baly et al, 2019), LABR (Aly and Atiya, 2013), and ArSaS (Elmadany et al, 2018). (2) NER, with the ANERcorp (Columbia University, 2016), and (3) Arabic QA, on Arabic-SQuAD (Mozannar et al, 2019) and ARCD (Mozannar et al, 2019) datasets.…”

“…They fine-tune these two versions on 5 sentiment datasets. These are HARD (Elnagar et al, 2018), the balanced data for ASTD (which we will refer to as ASTD-B) (Nabil et al, 2015), ArSenTD-Lev (Baly et al, 2019), AJGT (Alomari et al, 2017), and the unbalanced positive and negative classes for LABR (Aly and Atiya, 2013). split the data into 80/20 for training/test, respectively and report results in accuracy using the best epoch identified on test data.…”

“…+ : "very negative"; pos + : "very positive".• ArSAS.The Arabic Speech Act and Sentiment (ArSAS) corpus(Elmadany et al, 2018) consists of 21, 064 tweets annotated with four labels: positive, negative, neutral, and mixed.• ArSenD-Lev. The Arabic Sentiment Twitter Dataset for LEVantine dialect (ArSenD-Lev)(Baly et al, 2019) has 4, 000 tweets retrieved from the Levant region (i.e., Jordan, Lebanon, Palestine, and Syria). Sentiment is labeled on a 5-point scale, with 835 positive, 397 very positive, 885 neutral, 1, 253 negative, and 630 very negative.• ASTD.…”

Subjective Sentiment Analysis for Arabic Newswire Comments

“…Baly et al ( 2019) [8] created: the Multi-Dialect Arabic Sentiment Twitter Dataset, and the Arabic Sentiment Twitter Dataset for the Levantine dialect. The authors experimented with SVM, logistic regression and random forest trees classifiers using POS tags, numbers of positive/negative emoticons and words from different lexicons, Twitter-specific features, etc., in addition to several deep learning models.…”

AraNet: A Deep Learning Toolkit for Arabic Social Media

“…We collect 15 datasets related to sentiment analysis of Arabic, including MSA and dialects (Abdul-Mageed and Diab, 2012;Abdulla et al, 2013;Abdul-Mageed et al, 2014b;Nabil et al, 2015;Kiritchenko et al, 2016;Aly and Atiya, 2013;Salameh et al, 2015;Rosenthal et al, 2017;Alomari et al, 2017;Mohammad et al, 2018;Baly et al, 2019).…”