Research of Alexander Hogenboom and Flavius Frasincar leads to P* publication

In this article, Hogenboom and Frasincar investigate an automated approach for discovering the sentiment in text by exploiting the rhetorical structure of text. With more than 2 billion prosumers, i.e., users that produce as well as consume data, the Web has become one of the most influential sources of information. Nowadays, one fifth of all tweets and one third of all blog posts discuss products or brands. Due to today’s abundance of user-generated content on products and brands, and the fact that Alexander Hogenboom en Flavius Frasincar are able to automatically analyze this content for people’s sentiment, can prove beneficial for both businesses and consumers. Knowing in real-time the sentiment behind text can help companies better define their marketing and public relations strategies, while at the same time it can help consumers have access to the up-to-date and honest opinion of other consumers.

Rhetorical Structure Theory 

Differently than most existing approaches, which employ a bag-of-words approach to determine the polarity of text, the researchers exploit the rhetorical structure of text in order to have a better understanding of the conveyed sentiment. For this purpose, they employ the Rhetorical Structure Theory to obtain a tree-like representation of text, with nuclei denoting important text information, and satellites representing less relevant information. Alexander Hogenboom and Flavius Frasincar use a particle swarm optimization algorithm in order to learn how to weight nuclei and satellites from distinct types of rhetorical relations for their importance in terms of conveying a text’s sentiment, while accounting for their position in the text’s hierarchical rhetorical structure. The experiments performed using a real-world English movie review corpus indicate that such an approach provides better accuracy for sentiment analysis compared to the traditional bag-of-words approach, as well as approaches that perform shallow, coarse-grained analyses of a text’s rhetorical structure. In addition, the researchers show that accounting for rhetorical structure at sentence-level provides better accuracy for sentiment analysis than using paragraph-level or document-level analysis.