I had no sense of how much research has been done on sentiment analysis, opinion mining, and machine learning. I suppose major progress had to wait until there were massive datasets of text content available to process. The explosion of data on social media has provided both the impetus and the fodder to develop increasingly sophisticated techniques that are beginning to actually work.
In this annotation I present an overview of the techniques and tools of comparative opinion mining. It’s great to know what people feel and think about one thing. It’s probably twice as great to know what they think about two things. Like two different cars, guitars, or political candidates. Here we go:
Varathan, Kasturi Dewi, et al. “Comparative Opinion Mining: A Review.” Journal of the Association for Information Science & Technology, vol. 68, no. 4, Apr. 2017, pp. 811–29.
In Comparative Opinion Mining: A Review, Varathan, Giachanou, and Crestani provide a review of recent research on a form of opinion mining known as comparative opinion mining. They define comparative opinion mining as “a subfield of opinion mining which deals with extracting information that is expressed in a comparative form (e.g. “paper X is better than the Y”).” Comparative opinion mining attempts to make use of the vast number of posts on social networks by billions of people where they express opinions and write reviews about movies, restaurants, music, cars, books, hotels, etc., often by comparing one product relative to another. The authors note that such comparative opinion information is especially useful in product marketing, and for “competitive intelligence” in identifying potential markets and risks.
The authors begin by describing opinion mining as an equivalent term with sentiment analysis, defined as methods for automatic detection of “opinionated information” and “polarity of opinion toward a specific target.” They cite Liu and Zhang in defining an opinion as “a subjective statement, view, attitude, emotion, or appraisal about an entity or an aspect of a entity from a opinion holder.” An entity is simply an abstract object “such as a product, person, event, organization” represented in some hierarchy of components and attributes. The authors argue that while opinion mining is useful in understanding what consumers and citizens think about products, services, or policies, it is often insufficient in revealing what people think about known alternatives. This is where comparative opinion mining can provide more actionable data.
After establishing this distinction, the authors provide an overview of comparative opinion mining techniques. They divide these into three classes: Machine Learning, Rule Mining, and Natural Language Processing. They describe each technique in great detail. For example within machine learning, they discuss support vector machines (SVM), naive Bayes, conditional random fields, supervised, and unsupervised learning. Their section on rule mining identified techniques for discovering meaningful signals in the patterns and structures of terms, such as comparative words and phrases. They describe NLP approaches to analyze language in two levels: syntactic analysis to parse the syntax of sentences, and semantic analysis to identify the meaning of the sentence content. In their discussion of semantic analysis they introduce the semantic networks WordNet and SenticNet, which provide accessible knowledge bases and lexical resources for detecting synonyms, comparative words, and sentiment. They also describe test collections of comparative language available to researchers seeking to develop better algorithms and tools, such the J&L, JDPA, and Kessler14 data sets.
The article does not cover specific software processing methodologies, but briefly discusses several open source preprocessing tools, such as Gate, Stanford CoreNLP, and CRF Toolbox. The authors conclude by suggesting further research in comparative opinion mining. And they observe that in the survey they conducted for this article, they found that more than 60 percent of the researchers currently working on comparative opinion mining are Chinese scholars. They suggest that sociological research might provide insight into why Chinese scholars are more interested in comparative opinion mining than other researchers.
Adventures in Information Science 