Article Preview
TopIntroduction
The social media applications have become an important part of the daily life of millions of users. Twitter is an example of these famous social media platforms that is used by millions of users for different purposes. For instance, users in Twitter can share their interests, opinions and knowledge, as well as searching for latest news and reviews. This form of multi-usage by millions of users leads to generate massive amount of data in different types and forms. This makes the social media as a form of big data that is difficult to manage and utilize.
Big data concept has been introduced recently, and has been defined as the dataset that could not be handled by traditional software/hardware tools to process and manage within an acceptable time (Chen et al., 2014). Big data refers to the large complicated dataset that is difficult to process with regular data processing systems (Samuel et al., 2015). In addition, big data has been characterized by three properties, volume, velocity and variety (3V model). The volume is the size of the dataset that should be increasingly big; velocity is the speed of data generation, analysis and delivery that must be rapidly and timely conducted; and variety indicates the various types of data from different sources that include unstructured, semi-structured as well as structured data type (Chen et al., 2014; Tsai et al., 2015). Additional property of big data has been added to extend 3V model, in which 4V model has been introduced to include value property that indicates discovering values, e.g., meaningful information, from the dataset (Chen et al., 2014; Tsai et al., 2015).
Understanding the big data within certain context is a challenge in the social media. As a result, analysis process on this massive amount of big data is needed to better understand and utilize big data for specific purpose such as text classification. Text classification aims to assign pre-defined classes to text documents, such as labeling each news story with a topic like health, economy or sport (Hotho et al., 2005). There have been several different analysis techniques used for the purpose of text classification such as sentiment analysis (Waykar et al., 2016; Cai et al., 2010) which aims to analyze text to extract and classify user opinion either as positive or negative. In addition, classifying users’ opinions and interests is very important to understand the users’ concerns to provide them with better utilities and recommendations.
Many users in Twitter participate in writing tweets mainly for networking with others, and may not explicitly indicate their interests. Although some information such as name, age, location and short summary of interests may be available in the user’s profile; it can be incomplete, users may prefer not to share them, or deceptive users may choose to write fake information. Knowing users’ interests in social media is useful for different purposes, such as recommendation systems and marketing systems. Recommendation system may use the users’ interests to recommend friends for users that share the same interests, and marketing systems may use them for marketing purposes. In addition, it may be used for detecting abusive accounts.
There are several research studies such as Mangal et al. (2016), Michelson and Macskassy (2010), Lim and Datta (2013), Lee et al. (2011), and Magdy et al. (2015) that focused on Twitter classification for different purposes. Although these research studies have investigated the Twitter users’ opinions by understanding their tweets to classify users based on their interests such as Mangal et al. (2016), and Michelson and Macskassy (2010), most of these studies have been applied to English tweets. In spite of the fact that, millions of active Twitter users are Arabic speaking native, there is a lack of research that is conducted on Arabic language in comparison to English language due to the Arabic language’s morphological complexity and limited availability of software that is compatible with the Arabic language (Refaee and Rieser, 2014).