Documents that express an opinion abound, especially in the so-called web 2.0 era of social media and social networking. Jae-Young Chang of the Department of Computer Engineering at Hansung University, in Seoul, South Korea, suggests that there is a need to find ways to summarise their contents for a wide range of applications.
Writing in the International Journal of Computational Vision and Robotics, he points out that conventional text summarization methods do not work well with multiple documents authored by different writers. He has now proposed an algorithm that can identify and extract the representative documents from a large number of documents. Applying the process might be the first step towards a new approach to “opinion mining”, which could be useful in politics, marketing, education, and many other areas of human endeavour.
The approach involves detecting the sentiment of the most important – judging – document in a corpus and then ranking the relevance of others from this central point to allow a summary of the opinions expressed to be constructed. A successful proof of principle was carried out on movie reviews. The same approach should work well with product reviews and other kinds of opinion.
Chang, J-Y. (2020) ‘Multi-document summarisation using feature distribution analysis’, Int. J. Computational Vision and Robotics, Vol. 10, No. 2, pp.111–121.