Article Preview
Top1. Introduction
With the advancement in technology and rapid increase in social media platforms, the distance between businesses and customers is reducing. Customer feedback plays an important role in business. It gives the customer satisfaction level with respect to their product and services which helps in improvisation of the products or services.
The data available on the web is huge and it is present either in a structured or unstructured format. This data includes reviews, opinions, blogs, etc. Customer feedback is usually unstructured. To interpret the unstructured data and understand the polarity of the opinions expressed, it is important to leverage sentiment analysis.
Sentiment refers to a human attitude, opinion and emotions that are qualitative objects. Sentiment analysis is the process of analyzing an individual’s opinions, sentiments, attitudes and emotions towards entities such as products, services, organizations, individuals and events (Liu, 2012). Sentiment analysis interprets and classifies the emotions within the text using text analysis techniques. The task of sentiment analysis identifies positive, negative and neutral opinions, emotions and feelings from written information.
Sentiment analysis can be performed at three levels i.e., document level, sentence level and aspect level (Joshi & Itkat, 2014). Document-level sentiment analysis determines the overall opinion of the document. An assumption is made that each document expresses opinions on a single entity. Sentence level sentiment analysis determines opinions expressed in a sentence. Aspect level sentiment analysis performs a granular analysis and requires the use of different linguistic and grammatical rules. In sentiment analysis, it is important to identify positive, negative and neutral feelings expressed about the entity under analysis.
Negation analysis is one of the important subtasks of sentiment analysis. It is a complex task of identifying the negative opinions expressed within a context, which depends on many linguistic or grammatical rules of the language and negation scope identification. If not handled correctly, the presence of negation words within a sentence changes the polarity of the complete sentence. This, in turn, affects the results of sentiment analysis. Handling negation efficiently gives correct feedback about the entity and provides scope for improvement.
Sentiment analysis is considered as one of the text classification tasks in which the review text is represented by a bag-of-words (BOW) model. Machine learning algorithms such as Naïve Bayes or Support Vector Machine (SVM) are used for classification. But BOW model discards the word order and hence results in the loss of semantic information of the text. It results in the polarity shift problem, which is a phenomenon that reverses the polarity of the sentiment by linguistic structures called polarity shifters.
In lexicon-based sentiment analysis, negation handling techniques exploit grammatical relations among the words in a sentence and produce a dependency-based parse tree. The dependency-based parse tree is used to get the scope of negation and thus invert the polarity of words that are affected by negation. But these techniques ignore the consideration of features such as the influence of conjunctions and punctuation marks.
There are negation forms beyond the prototypical negations “not” and “never”. These negation forms are called approximate negators. Approximate negators include words such as “rarely”, “hardly”, “seldom”, “few”, “little” etc. Approximate negators do not give the effect of absolute negators such as “not” and “never”. Hence, handling approximate negators in identifying negation scope is challenging.
There are adjectives or adverbs, that describe another word or change its meaning in some way. Such words are called modifiers. Modifiers include negations, amplifiers and downtoners. Though negations are widely studied, amplifiers and downtoners need extensive attention.
In this paper, different categories of negation words that can exist within a sentence are explained. Rules applicable to different types of negation words and negation scope detection have been described. An algorithm called SentiNeg is proposed to identify and perform negation analysis on both simple and compound sentences. The algorithm provides a clear aspect-based summary of negative opinions expressed on the entity under analysis.