Sentiment Analysis Techniques and Application-Survey and Taxonomy

Abstract

Nowadays, social media platforms, blogs, and e-commerce are commonly use to express opinion on politics, movies, products, education respectively; for election forecasting, business boosting and improvement of teaching and learning. As a result, data generation becomes easier; producing big data which requires appropriate techniques and tools to analyse easily, accurately and timely. Thus, making sentiment analysis very demanding research area. This study will investigate on what basis (sentiment classification level) or area of application (data source) do supervised machine learning approaches particularly Support Vector Machine (SVM), Naïve Bayes, and Maximum Entropy algorithms, and other technique-lexicon-based approach give the best result in sentiment analysis. Based on the review of the literature there is a contradiction on the point that SVM generated the best result in analyzing student sentiment on document level. This study also discovers that sentiment analysis differs from system to system based on polarity (types of the classes to predict: positive or negative, subjective or objective), different levels of classification (sentence, phrase, or document level) and language that is processed. This research produces a taxonomy which serves as a guide for the choice of techniques in sentiment analysis. The taxonomy explores the sentiment classification levels and data preprocessing stages. It also explores that sentiment analysis techniques were organised in to three (3) groups; Machine learning, Lexicon and hybrid or combination. The machine learning techniques were sub-grouped in to two (2) namely; supervised and unsupervised. The supervised were organized in to two (2): Classification and Regression. un-supervised machine learning techniques includes clustering and association. The clustering technique consist of k-means. Decision tree which is a classification based under supervised type of machine learning technique consist of random forest,(Akinkunmi, 2019) while the ruled-based classifiers consist of confidence criterion and support criterion. The commonly used tools are Weka, Python compiler, and R programming tool.

Sentiment Analysis in E-Commerce Platforms: A Review of Current Techniques and Future Directions

“…Anvar Shathik & Krishna Prasad et al [57]introduced the prevalent techniques used in SA from a machine learning perspective. Umar et al [60] investigated the sentiment clas-sification level or data source on which supervised machine learning techniques like SVM, NB, Maximum Entropy, and other technique such as lexicon-based which deliver optimal results in SA. Anvar Shathik, et al and Umar, et al solely encompassed a review focusing on machine learning techniques.…”

Sentiment Analysis Techniques –Survey

“…The machine learning requires too much of labelled data to give best accuracy result, and its characterized as unsupervised and supervised algorithms that divide into four part; decision tree classifiers, rule-based classifiers, linear classifiers (include SVM and Neural networks), and probabilistic classifiers (include Naïve Bayes, Bayesian network and maximum entropy). The lexicon-based approaches don't need labelled data because its unsupervised, and its split to two portion, dictionary and corpus based approaches (include statistical and semantic) [16].…”

“…This type is consists of neurons that linked to one another through weights that associate the neurons for easier transit of signals to revenue a single output [16]. The main advantages of neural network can execute tasks that a linear program unable to do it.…”

“…Essentially the algorithm supposed that any feature of class is separate of other feature in the same class [22][24]. This technique commonly used for simplicity through training and classification and it's based on Bayes theory, where the pre-processed data is the input to training set by classifier [16].…”

“…The Max Entropy classifier can be utilize to solve a large textual classification problems such as detection the language, subject classification, and SA. Features can be added to MaxEnt without any overloading like phrases and bigram, also, Its utilize to assess probability distribution utilization logistic regression [16].…”