Maryam Boubekraoui – Mathematics and Decision

$\begin{quote} \begin{center} \textbf{Extrapolation methods for multilinear PageRank} \end{center} \medskip Multilinear PageRank stands as an innovative rendition of the Google PageRank algorithm, designed to consider a network's diverse web of connections. This algorithm promises heightened efficiency and precision in the ranking of web pages by simultaneously accounting for multiple relationship types. For the computation of the multilinear PageRank vector, the higher-order power method is commonly employed due to its practical implementation, minimal storage requirements, and its organic extension from the traditional power method employed in the classic Google PageRank algorithm. Nonetheless, the convergence of this method does not always come with a guarantee, and even when it does, its pace can be sluggish. In this presentation, we introduce an acceleration approach for computing fixed-point multilinear PageRank. We achieve this by harnessing vector extrapolation techniques, specifically the minimal polynomial extrapolation (MPE) and reduced rank extrapolation (RRE). These methods promise to enhance the efficiency of the computation process, a development that could have significant implications for web ranking systems. \end{quote}$