Is there a way to automate the extraction of value from the references in a scientific research paper? Yi Zhao, Keqing He, and Junfei Guo of Wuhan University and Zhao Li and Bitao Li of China Three Gorges University are working on this problem and discuss details of their findings so far in the International Journal of Computational Science and Engineering.
One of the secondary, but important features of a scientific paper, are the references. They are the underpinnings of the research being discussed on which the new discovery is built. They also have another function within the scientific community, they can act as recommendations, suggested reading, that might lead others to novel work they may otherwise not have encountered.
Unfortunately, as scientific projects expand and develop and the literature around the Sciencebase accumulates, the reference sections of many research papers themselves grow more and more unwieldy. Finding the hidden gems, the essential reading, the deepest foundations, becomes increasingly difficult, especially in the digital age where access to such large numbers of papers and all of their references is available with a few touches of a screen or a mouseclick or two.
The team has now developed a new type of recommendation research approach – a collective intelligence network approach – using classification, clustering, and recommendation models integrated into the system. When they compare the output from their algorithm against other approaches they see a 10% higher accuracy and 8% higher F-score for the recommendations compared with keyword matching approaches.
Zhao, Y., Li, Z., Li, B., He, K. and Guo, J. (2019) ‘Collective intelligence value discovery based on citation of science article‘, Int. J. Computational Science and Engineering, Vol. 19, No. 4, pp.527-537.