SentiArabic: A Sentiment Analyzer for Standard Arabic

SentiArabic: A Sentiment Analyzer for Standard Arabic Ramy Eskander Ramitechs New York, USA eskander@ramitechs.com Abstract Sentiment analysis has been receiving increasing interest as it conveys valuable information in regard to people’s preferences and opinions. In this work, we present a sentiment analyzer that identifies the overall contextual polarity for Standard Arabic text. The contribution of this work is threefold. First, we modify and extend SLSA; a large-scale Sentiment Lexicon for Standard Arabic. Second, we build a sentiment corpus of Standard Arabic text tagged for its contextual polarity. This corpus represents the training, development and test sets for the proposed system. Third, we build a lightweight lexicon-based sentiment analyzer for Standard Arabic (SentiArabic). The analyzer does not require running heavy computations, where the link to the lexicon is carried out through a morphological lookup as opposed to conducting a rich morphological analysis, while the assignment of the sentiment is based on a simple decision tree that uses polarity scores as opposed to a more complex machine learning approach that relies on lexical information, while negation receives special handling. The analyzer is highly efficient as it achieves an F-score of 76.5% when evaluated on a blind test set, which is the highest results reported for that set, and an absolute 3.0% increase over a state-of-the-art system that uses deep-learning models. Keywords: Sentiment Analysis, Sentiment Lexicon, Sentiment Corpus, Standard Arabic 1. Introduction Sentiment analysis is the process of applying computational approaches to identify attitudes, emotions and opinions in text, speech and visual data. While there is an enormous number of sentiment analysis tools for English and other common languages, Arabic has received less focus due to lack of resources, corpora and lexicons in particular, in ad- dition to the complexity of its morphological and syntacti- cal systems, which incurs ambiguity and requires extensive processing. In this work, we present SentiArabic, a lightweight lexicon- based sentiment analyzer for Standard Arabic. For a given text, the system identifies the contextual polarity (Positive or Negative) of the underlying sentiment. Our main contri- bution in this paper can be summarized as follow: 1. We modify and extend our previous sentiment re- source SLSA, a large-scale Sentiment Lexicon for Standard Arabic (Eskander and Rambow, 2015). SLSA includes 34,281 entries, where an entry con- sists of a lemma, a part-of-speech (POS) tag, the cor- responding English gloss, and three sentiment scores; positive, negative and objective, where the objective score is calculated as 1 - (positive score + negative score). The new extension follows the same structure but with a higher quality and better normalization of the polarity scores. 2. We create a new sentiment corpus of Standard Ara- bic text where each sentence is tagged for its polarity. The corpus is divided into three sets for the training, development and test of the sentiment analyzer. 3. We build SentiArabic, a lightweight sentiment ana- lyzer for Standard Arabic. SentiArabic is based on the lexicon in (1), and is trained and evaluated based on the corpus in (2). SentiArabic is “lightweight” …

Content truncated to 500 words