Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
535097 | Pattern Recognition Letters | 2016 | 6 Pages |
•We extended the Quantum Clustering to text analysis and authorship identification.•We performed the full-scale comparison between Quantum Clustering and Parzen-window.•The Quantum Clustering is better than Parzen-window on achieving optimal clustering.•Quantum Clustering depend much less sensitively upon the choice of parameter.•The Pattern Search could be an appropriate solution to the parameter estimation.
The article introduces Quantum Clustering, a novel pattern recognition algorithm inspired by quantum mechanics and extend it to text analysis. This novel method improves upon nonparametric density estimation (i.e. Parzen-window), and differentiates itself from it in a significant way, Quantum Clustering constructs the potential function to determine the cluster center instead of the Gaussian kernel function. Specifically, detailed comparative analysis shows that the potential function could clearly reveal the underlying structure of the data that the Gaussian kernel could not handle. Moreover, the problem of parameter estimation is solved successfully by the numerical optimization approach (i.e. Pattern Search). Afterwards, the results of detailed comparative experiments on three benchmark datasets confirms the advantage of Quantum Clustering over the Parzen-window, and the additional trial on authorship identification illustrates the wide application scope of this novel method.