Article Preview
TopIntroduction
The increasing traction of social media avenues to verbalize personal notions & beliefs has created a need to put in place a paradigm which can analyse the humongous amount of data involved, the task is typically referred to as sentiment analysis (Kumar & Sharma, 2016). Formally, Sentiment Analysis is defined as the study, and subsequent categorization, of an individual’s feelings and opinions, communicated through text, with respect to a certain context (Kumar & Abraham, 2017; Kumar & Teeja, 2012). The categorization is carried out along the lines of polarities, such as positive and negative, etc. (Kumar & Sebastian, 2012; Kumar & Sharma, 2017).
Sentiment analysis, also known as opinion mining, is the means of recognizing and designating opinions communicated through a written piece to ascertain the author’s connotation (positive, objective or negative) of that piece using a combination of statistical and computational techniques (Kumar & Jaiswal, 2017).
The core module of the Sentiment Analysis process employs feature extraction, a process used to convert input data, consisting of text indicating opinions, into an array of features, which can represent the input data very well (Kumar & Khorwal, 2017). Feature Selection is a technique used to select a sub-set of relevant features, discarding nonessential attributes (Kumar & Rani, 2016). Effective and efficient feature selection affects the quality of sentiments extracted and hence the classifier performance. But it has been observed that many features exist which don’t contribute to accuracy, and thus can be removed without causing much loss. Fewer features reduce the complexity of the analysis, facilitating optimization.
Many researchers have adopted metaheuristic or stochastic methods for employing efficacious feature selection (Kumar, Khorwal, & Chaudhary, 2017). Metaheuristic methods exploit the trade-off which exists between a relatively robust solution and computational effort. Swarm intelligence-based stochastic methods are distinctly attractive for feature selection. Swarm Intelligence is the area of artificial intelligence that deals with systems composed of multiple entities called agents that correlate using self-organization and localized control. Agents are governed by simple rules and their behaviours are governed by their actual roles they play in their natural habitat. Movement of individual agents is decentralized, however, interaction between agents’ results in a universal intelligent behaviour.
Cuckoo Search (CS) algorithm is a nature inspired, metaheuristic optimization algorithm which belongs to a group of swarm intelligence algorithms (Yang & Deb, 2009). The algorithm takes its inspiration from the cuckoo birds’ parasitic practice of laying their eggs in the nests of hosts. The primary objective is to combine a set of binary coordinates for each solution, signifying if a particular feature belongs to the subsequent group of features or not. A classifier is trained with the selected features, encoded by the significance of the eggs. The solution’s quality is then determined by evaluating each nest (Yang & Deb, 2009).
Recent literature has shown that CS algorithm has been surveyed as being more computationally efficient than PSO (Adnan & Razzaque, 2013).
Pereira et al. (2014) have developed a binary adaptation of CS algorithm, named Binary Cuckoo Search (Bcs). BCS is designed specifically to achieve optimum feature selection. It is the modified variant of the generic Cuckoo Search (CS) algorithm, which outputs the subset of features that are most efficient in classification.