Approaching Sentiment Analysis by using semi-supervised learning of multi-dimensional classifiers

Abstract

Sentiment Analysis is defined as the computational study of opinions, sentiments and emotions expressed in text. Within this broad field, most of the work has been focused on either Sentiment Polarity classification, where a text is classified as having positive or negative sentiment, or Subjectivity classification, in which a text is classified as being subjective or objective. However, in this paper, we consider instead a real-world problem in which the attitude of the author is characterised by three different (but related) target variables: Subjectivity, Sentiment Polarity, Will to Influence, unlike the two previously stated problems, where there is only a single variable to be predicted. For that reason, the (uni-dimensional) common approaches used in this area yield suboptimal solutions to this problem. In order to bridge this gap, we propose, for the first time, the use of the novel multi-dimensional classification paradigm in the Sentiment Analysis domain. This methodology is able to join the different target variables in the same classification task so as to take advantage of the potential statistical relations between them. In addition, and in order to take advantage of the huge amount of unlabelled information available nowadays in this context, we propose the extension of the multi-dimensional classification framework to the semi-supervised domain. Experimental results for this problem show that our semi-supervised multi-dimensional approach outperforms the most common Sentiment Analysis approaches, concluding that our approach is beneficial to improve the recognition rates for this problem, and in extension, could be considered to solve future Sentiment Analysis problems.

Semi-supervised learning for ordinal Kernel Discriminant Analysis

“…The problem of semi-supervised learning (i.e. learning from both labelled and unlabelled observations) is an example that has been the focus of many machine learning researchers in the past 10 years. In many real-world applications, obtaining labelled patterns could be a challenging task, however, unlabelled examples might be available with little or no cost.…”

“…These learning approaches have been empirically and theoretically studied in the literature and represent a suitable solution for such circumstances, where the use of unlabelled data has been seen to improve the performance of the model and stabilise it. Semi-supervised learning has being mainly studied for binary classification [5,6] and regression [2], al- 20 though recently the main focus has shifted to multi-class problems [7,8,9] (and even multi-dimensional ones [10]). This paper tackles the use of unlabelled data in the context of ordinal classification [11], a learning paradigm which shares properties of both classification and regression.…”

Tax Complaints Classification on Twitter Using Text Mining

“…The Naïve Bayes classifier is a probabilistic classifier that assumes the statistical independence of each feature (or word) and is a conditional model based on Bayes' formula [27] [28]. This classifier estimates the probabilities that an object from each class falls in each possible discrete value of vector variable x [29].…”

Understanding Natural Disasters as Risks in Supply Chain Management through Web Data Analysis

“…Machine leaning classification methods such as naï ve Bayes [10], [11], maximum entropy [12] and support vector machines [13], [14] have been widely used in classifying text data. However, their heavy reliance on training datasets makes this approach prohibitively difficult as datasets are not always readily available.…”