Mohammed N. Al-Kabi

An analytical study of Arabic sentiments: Maktoob case study

The problem of automatically extracting opinions and emotions from textual data have gained a lot of interest recently. Unfortunately, most studies on Sentiment Analysis (SA) focus mainly on the English language, whereas studies considering other important and wide-spread languages such as Arabic are few. Moreover, publicly-available Arabic datasets are seldom found on the Web. In this work, a labeled dataset of Arabic reviews/comments is collected from a social networking website (Yahoo!-Maktoob). A detailed analysis of different aspects of the collected dataset such as the reviews length, the numbers of likes/dislikes, the polarity distribution and the languages used is presented. Finally, the dataset is used to test popular classifiers commonly used for SA.

Automatic Lexicon Construction for Arabic Sentiment Analysis

Sentiment Analysis (SA) is the process of determining the sentiment of a text written in a natural language to be positive, negative or neutral. It is one of the most interesting subfields of natural language processing (NLP) and Web mining due to its diverse applications and the challenges associated with applying it on the massive amounts of textual data available online (especially, on social networks). Most of the current works on SA focus on the English language and follow one of two main approaches, (corpus-based and lexicon-based) or a hybrid of them. This work focuses on a less studied aspect of SA, which is lexicon-based SA for the Arabic language. In addition to experimenting and comparing three different lexicon construction techniques, an Arabic SA tool is designed and implemented to effectively take advantage of the constructed lexicons. The proposed SA tool possesses many novel features such as the way negation and intensification are handled. The experimental results show encouraging outcomes with 74.6% accuracy in addition to revealing new insights and guidelines that could direct the future research efforts.

Towards Improving the Lexicon-Based Approach for Arabic Sentiment Analysis

The emergence of the Web 2.0 technology generated a massive amount of raw data by enabling Internet users to post their opinions on the web. Processing this raw data to extract useful information can be a very challenging task. An example of important information that can be automatically extracted from the users posts is their opinions on different issues. This problem of Sentiment Analysis SA has been studied well on the English language and two main approaches have been devised: corpus-based and lexicon-based. This work focuses on the later approach due to its various challenges and high potential. The discussions in this paper take the reader through the detailed steps of building the main two components of the lexicon-based SA approach: the lexicon and the SA tool. The experiments show that significant efforts are still needed to reach a satisfactory level of accuracy for the lexicon-based Arabic SA. Nonetheless, they do provide an interesting guide for the researchers in their on-going efforts to improve lexicon-based SA.

Sentiment analysis of arabic social media content: a comparative study

The Internet became an indispensable part of peoples lives because of the significant role it plays in the ways individuals interact, communicate and collaborate with each other. Over recent years, social media sites succeed in attracting a large portion of online users where they become not only content readers but also content generators and publishers. Social media users generate daily a huge volume of comments and reviews related to different aspects of life including: political, scientific and social subjects. In general, sentiment analysis refers to the task of identifying positive and negative opinions, emotions and evaluations related to an article, news, products, services, etc. Arabic sentiment analysis is conducted in this study using a small dataset consisting of 1,000 Arabic reviews and comments collected from Facebook and Twitter social network websites. The collected dataset is used in order to conduct a comparison between two free online sentiment analysis tools: SocialMention and SentiStrength that support Arabic language. The results which based on based on the two of classifiers (Decision tree (J48) and SVM) showed that the SentiStrength is better than SocialMention tool.

Evaluating SentiStrength for Arabic Sentiment Analysis

Social networking websites are used today as platforms enabling their users to write down almost anything about everything. Social media users express their opinions and feelings about lots of events occurring in their daily lives. Lots of studies are conducted to study the sentiments presented by social media users regarding different topics. Sentiment Analysis (SA) is a new field that is concerned with measuring the sentiment presented in a given text. Due to their wide set of applications, several SA tools are available. Most of them are designed for English text. As for other languages such as Arabic, the case is different since only few tools are available. In fact, many of these tools were originally designed for English and were later adapted to deal with Arabic. SentiStrength is an example of tools that are successful for English and were later adapted to Arabic. However, the adaptation has been done in a crude manner and no deep studies are available to measure the effectiveness of such tools for Arabic text. In this paper, we perform a comprehensive evaluation of SentiStrength using 11 Arabic datasets consisting of tens of thousands of reviews/comments from different domains and in different dialects. We perform the evaluation in terms of positive and negative sentiments. The evaluation results show that overall SentiStrength achieves 62% accuracy, 83.7% precision, 64% recall (positive correct), 68% F1 measure and 55% negative correct.

Towards improving Khoja rule-based Arabic stemmer

Stemming algorithms are used to remove irrelevant morphological variations from different words, and extract the stem or the root from which the inputted word is derived. Stemming can then help to standardize terms referring to the same concept. These algorithms are widely used in information retrieval systems and Web search engines, in addition to other systems such as: Machine translation, text clustering, text summarization, question answering, indexing, text mining, text classification... etc. Khoja stemmer is a standard Arabic stemmer, which has a number of flaws. Previous studies and this one show that Khoja stemmer is better than other two competitive ones evaluated in this study. The Khoja stemmer and the other two evaluated Arabic stemmers depend mainly in their work on (Patterns, Forms). Therefore the identification of the flaws leads to identification of missing Patterns not used by Khoja stemmer. So the enhancement to Khoja stemmer is restricted to adding missing patterns, and this leads to around 5% improvement to the accuracy of Khoja stemmer.

Are emoticons good enough to train emotion classifiers of Arabic tweets

Nowadays, the automatic detection of emotions is employed by many applications across different fields like security informatics, e-learning, humor detection, targeted advertising, etc. Many of these applications focus on social media. In this study, we address the problem of emotion detection in Arabic tweets. We focus on the supervised approach for this problem where a classifier is trained on an already labeled dataset. Typically, such a training set is manually annotated, which is expensive and time consuming. We propose to use an automatic approach to annotate the training data based on using emojis, which are a new generation of emoticons. We show that such an approach produces classifiers that are more accurate than the ones trained on a manually annotated dataset. To achieve our goal, a dataset of emotional Arabic tweets is constructed, where the emotion classes under consideration are: anger, disgust, joy and sadness. Moreover, we consider two classifiers: Support Vector Machine (SVM) and Multinomial Naive Bayes (MNB). The results of the tests show that the automatic labeling approaches using SVM and MNB outperform manual labeling approaches.

Measuring the controversy level of Arabic trending topics on Twitter

Social micro-blogging systems like Twitter are used today as a platform that enables its users to write down about different topics. One important aspect of such human interactions is the existence of debate and disagreement. The most heated debates are found on controversial topics. Detecting such topics can be very beneficial in understanding the behavior of online social networks users and the dynamics of their interactions. Such an understanding leads to better ways of handling and predicting how the online crowds will act. Several approaches have been proposed for detecting controversy in online communication. Some of them represent the interactions in the form of graphs and study their properties in order to determine whether the topic of interaction is controversial or not. Other approaches rely on the content of the exchanged messages. In this study, we focus on the former approach in identifying the controversy level of the trending topics on Twitter. Unlike many previous works, we do not limit ourselves to a certain domain. Moreover, we focus on social content written in Arabic about hot events occurring in the Middle East. To the best of our knowledge, ours is the first work to undertake this approach in studying controversy in general topics written in Arabic. We collect a large dataset of tweets on different trending topics from different domains. We apply several approaches for controversy detection and compare their outcomes to determine which one is the most consistent measure.

SPAR: A system to detect spam in Arabic opinions

The evaluation of the public opinion through websites, social networks, news feedback, etc. is currently getting an extensive research to discover public opinion regarding the current social and political changes in the Middle Eastern countries. However, the level of trust or confidentiality of such public opinion evaluations may have the risk of being spammed. This study aims to detect the spam opinions in the Yahoo!-Maktoob social network. The proposed system reads the opinions and classifies them into one of the following two classes: spam and non-spam opinions, based on a number of features. Each spam opinion categorizes into; high levels spam and low level spam, based on special metrics. While each non-spam opinion is labeled as; positive, negative, or neutral based on the language polarity dictionaries. Those dictionaries include words that can be classified as: positive, negative or neutral. The proposed system adopts machine learning classification technique to perform classification and prediction.

Polarity Classification of Arabic Sentiments

Sentiment Analysis/Opinion Mining is associated with social media and usually aims to automatically identify the polarities of different points of views of the users of the social media about different aspects of life. The polarity of a sentiment reflects the point view of its author about a certain issue. This study aims to present a new method to identify the polarity of Arabic reviews and comments whether they are written in Modern Standard Arabic MSA, or one of the Arabic Dialects, and/or include Emoticons. The proposed method is called Detection of Arabic Sentiment Analysis Polarity DASAP. A modest dataset of Arabic comments, posts, and reviews is collected from Online social network websites i.e. Facebook, Blogs, YouTube, and Twitter. This dataset is used to evaluate the effectiveness of the proposed method DASAP. Receiver Operating Characteristic ROC prediction quality measurements are used to evaluate the effectiveness of DASAP based on the collected dataset.