Article Preview
TopIntroduction
In recent years, the analysis of sentiment has developed explosively, a lot of companies use it to analyze the opinions of their customers for developing the quality of their products, and also it is used recently in e-learning platforms to analyze the motivation of learners (motivated, unmotivated). The analysis of sentiment has great development in scientific research and it is known as sentiment analysis.
Sentiment analysis (SA) is the process of extracting opinions, sentiments or feelings from documents or from social networks data. Sometimes it tries to classify text into classes such as positive, negative or neutral, or calculating a polarity (sentimental degree). The process of sentiment analysis uses a lot of other domains like natural language processing (NLP) through a combination of pre-processing steps, machine learning algorithms (supervised and unsupervised) and relevant statistical techniques (Lexicon-based approach, corpus-based approach, dictionary-based approach).
SA is used in many tasks of our daily life. For example, it is used by individuals for making a decision on what products to buy based on the opinions of others, or by companies to know the opinions and sentiments of costumers about their products. It is also used by teachers in an e-learning platform to define the motivation of learners based for example on their social network profiles (Madani, 2017).
Most recent SA approaches are based on social networks like Twitter, Facebook, or Google+. On these websites, people share a large amount of data and can express their opinions, attitudes, and feelings about products, movies, social events, etc., The big advantage of social networks is that anyone can express what he thinks in a freeway without hindrance.
Among variety of social networks, Twitter for example, is a popular microblogging website with over 328 million active users per month and about 500 million tweets per day in over 40 languages, messages are limited to 280 characters and are known under the name tweets and may include text, URLs, other user mentions and hashtag metadata to messages. These tweets represent the users' opinions and thoughts expressed in short and simple messages. Twitter gives everyone the power to create and share ideas and information instantly and without hindrance (https://about.twitter.com/fr/company). All that motivates researchers to use Twitter in the field of sentiment analysis.
Sentiment analysis over Twitter tries to classify the tweets into classes (positive, negative, neutral) or calculating a degree of importance (polarity) for each tweet (for example between 1 and 5). To classify a tweet, researchers propose a lot of methods namely the use of machine learning algorithms, the use of dictionaries such as AFINN, SentiWordNet or SenticNet, or some hybrid approaches based on both machine learning approaches and dictionaries (Ohana, 2009).
In this paper, we propose a new approach to classify tweets according to three classes (positive, Negative and Neutral) based on a dictionary-based approach, using SentiWordNet dictionary and the fuzzy logic with its three steps (Fuzzification, Fuzzy rules, and Defuzzification).
Most existing approaches in the literature treat the classification of tweets in a black and white manner (Rezwanul Huq, 2017), without considering fuzziness and vagueness of sentiments. Reality is far from optimistic. In fact, the sentiments are often fuzzy and contain a lot of vagueness. For example, the same word can explain more than one feeling at the same time. The fuzzy logic is a solution with which we can take into account the fuzziness when classifying tweets. From all that, our work consists of a hybridization between a dictionary-based approach using SentiWordNet and the fuzzy logic system.