Productive opinions

Opinions about products and services abound on the modern internet thanks to social networking and social media apps. A new way to extract and analyse those opinions to help manufacturers and service providers tailor their products to consumer needs and wants is reported in International Journal of Data Mining, Modelling and Management.

Computer scientists Farek Lazhar and Tlili-Guiassa Yamina of Badji-Mokhtar Annaba University, in Annaba, Algeria, explain that there are huge amounts of subjective data available on the web. This data encompasses opinions, sentiments and beliefs and represents an important resource for companies and marketers with products and services to promote. Conversely, it is invaluable for individuals interested in the opinions of others to inform their purchases, the services they use and even the politicians they vote for.

The team has developed a computer algorithm that can mines reviews to extract features-opinion pairs. The tool then classifies these features into one of two main classes: positive or negative. “Our approach is articulated on the use of dependency grammar to extract explicit feature-opinion pairs and the use of domain ontology to extract implicit feature-opinion pairs by exploiting relations between concepts, individuals and attributes,” the team explains. The final output is guided by support vector machine (SVM) as a supervised learning technique so that human input can be used to validate and train the algorithm on test data so that subsequent runs work smoothly and automatically with few errors.

Given the vast quantity of opinion data now to be found on almost every e-commerce site, social media and networking apps and consumer review sites, homing in on a consensus when making purchasing choices, whether for a smart phone, kitchen refit, taxi ride, or holiday let or choosing a political leader, is no easy task. Moreover, in manually searching for opinions, one inevitably runs the risk of selecting opinions with which one agrees beforehand. Thus, an automated system than can objectively, rather than subjectively, garner opinions from a large data set and present the user with an opinion on a given choice based on the generalised opinion of other users would be much more useful. In other words, “In order to track people’s opinions and sentiments, a powerful tool is needed for both companies and individuals to classify automatically the features on which sentiments have been expressed,” the team reports.

The team has now defined and tested their algorithm on holiday accommodation. The next step will to be to enrich the ontology to allow more refined and subtle opinions, whether positive or negative or somewhere in between, to be extracted from a data set more precisely.

Lazhar, F. and Yamina, T.G. (2016) ‘Mining explicit and implicit opinions from reviews‘, Int. J. Data Mining, Modelling and Management, Vol. 8, No. 1, pp.75-92.

Author: David Bradley

Award-winning, freelance science writer based in Cambridge, England.