LEBANONUPRISING: a thorough study of Lebanese tweets

TL;DR

This paper analyzes Lebanese tweets related to the October 2019 uprising, developing sentiment and emotion detection methods, creating a Lebanese-Arabic dictionary, and comparing datasets over time.

Contribution

It introduces a new Lebanese Arabic sentiment analysis framework, a Lebanese-Standard Arabic dictionary, and explores emotion detection using emojis in social media.

Findings

Sentiment analysis achieved satisfactory accuracy.

Emotion detection identified sarcastic and funny tweets.

Variation observed between datasets from different periods.

Abstract

Recent studies showed a huge interest in social networks sentiment analysis. Twitter, which is a microblogging service, can be a great source of information on how the users feel about a certain topic, or what their opinion is regarding a social, economic and even political matter. On October 17, Lebanon witnessed the start of a revolution; the LebanonUprising hashtag became viral on Twitter. A dataset consisting of a 100,0000 tweets was collected between 18 and 21 October. In this paper, we conducted a sentiment analysis study for the tweets in spoken Lebanese Arabic related to the LebanonUprising hashtag using different machine learning algorithms. The dataset was manually annotated to measure the precision and recall metrics and to compare between the different algorithms. Furthermore, the work completed in this paper provides two more contributions. The first is related to building a Lebanese to Modern Standard Arabic mapping dictionary that was used for the preprocessing of the tweets and the second is an attempt to move from sentiment analysis to emotion detection using emojis, and the two emotions we tried to predict were the "sarcastic" and "funny" emotions. We built a training set from the tweets collected in October 2019 and then we used this set to predict sentiments and emotions of the tweets we collected between May and August 2020. The analysis we conducted shows the variation in sentiments, emotions and users between the two datasets. The results we obtained seem satisfactory especially considering that there was no previous or similar work done involving Lebanese Arabic tweets, to our knowledge.