Article Preview
TopIntroduction
Social media are considered as a huge corpus for extracting information of various types. One of the ‘hot topics’ in social media usage is the sentiment analysis. A group’s opinions and feelings about any mentioned topic, serve as a valuable source of marketing for companies and consumers alike. Companies may utilize this information to measure customers’ satisfaction about the product and it helps facilitate the decision-making process of the consumer. Consumers may seek both positive and/or negative feedback prior to make a decision, which is known as ‘opinion mining’. Twitter, which has over 600 million users and nearly 330 million active users worldwide (Statista, 2017), rapidly becomes a ‘golden astrologer’ in the corporate universe, as it allows companies to elicit and analyze the sentiment of users’ tweets and employ the result of the analysis to grow their own image and trademark. There are two key parameters that directing any research on sentiment analysis, these are: the target field and the study sample. In this study, the study sample will be the consumer reviews that are collected from Twitter and Cloud Service Providers will serve as the target field in this study.
Several approaches were developed to enhance the accuracy of the sentiment analysis in social networks. The most accurate results were obtained using a hybrid technique of machine learning approaches (e.g. support vector machine, Naïve-Bayes and K-mean Fuzzy) and lexicon-based approaches, like a dictionary-based and corpus-based approaches. A dictionary-based approach is based on utilizing dictionaries that include words and their sentiments, which is a robust and straight-forward approach as most dictionaries list synonyms and antonyms for each term. While there are various such dictionaries available online, according to Islam Zibran (2017) and Jhaveri et al. (2011) SentiWordNet and SentiStrength are the best among other online dictionaries. Different techniques addressed different challenges of sentiment analysis. These challenges have been addressed in the previous work, but separately. Some techniques focused on increasing the accuracy using a hybrid approach, but did not address the challenges of intensity, wsd and comparative words. Accordingly, a new technique that solved the expected forms of sentimental text, these are word sense, comparative word, compound words, negative, intensity, sarcasm and big data, is required.
The aim of this study is to measure user satisfaction about the Cloud services provided by Google, Microsoft and Amazon. Tracking the users’ replies will present a general view on the satisfaction level and customers’ opinions, which give the providers information about possible improvements. The proposed approach is built using these well-known dictionaries that are mentioned above and using fuzzy logic classification. Fuzzy logic is a computational model that mimics human decision-making process. Fuzzy logic represents the domain of interest using fuzzy terms, similar to how the human brain absorbs information (e.g. a temperature is hot, a speed is slow). The ability of the human brain to reason with uncertainties, inspired researchers to develop a fuzzy logic model (Liu & Cocea, 2017). Fuzzy allows to create general rules for decision making with uncertainty (Mary & Arockiam, 2017; Serguieva et al., 2017). Tweet classification, similar to other classification systems, is implemented based on a set of extracted features and using a classification procedure/algorithm. Fuzzy Inference Process (FIP) is used for the classification process because of its flexibility in addressing complex problems. Moreover, FIP allows for incorporating various feature forms, such as those extracted directly from the input or using a dictionary-based approach.
The proposed approach accurately classifies tweets based on their sentiment content. Accordingly, the tweets will be categorized into five categories: Very Positive, Positive, Neutral, Negative, Very Negative. The proposed model is built to achieve a set of goals: 1) To collect cloud services’ relevant tweets and analyzes them in real time. 2) To provide cloud services with reports about customers’ feedback. 3) To assist consumers in decision making within in a brief time span. 4) To gain analysis results that closely matched with human analysis. 5) To obtain accurate sentiment analysis results. The contribution of this paper is to propose a hybrid technique that uses the available resources, a set of rules, a fuzzy classification in order to accurately identify the sentiment of the input text while facing the challenges that are facing sentiment analysis.