Practical acceleration for computing the HITS expertrank vectors (Q442725)

The author aims at accelerating the computation of Hyperlink-Induced Topic Search (HITS), developed by Kleinberg in the 90s and used in the \texttt{ask.com} search engine. A practical acceleration scheme is proposed, exploiting a filtered power method by adaptive Chebyshev polynomials. Detailed algorithms are provided and carefully analysed. The proposed scheme appears efficient and can be extended for computing more principal eigenvectors, such as in truncated SVD calculations. Numerical examples enlighten the features of the presented method.