A normalized score-based weighted PageRank algorithm on ranking prediction of basketball games

Abstract

Ranking of sports teams has always been significant to sponsors, coaches, as well as audiences. Prevailing prediction methods investigate probabilities by taking into account of different kinds of attributes (e.g. field goals, fields goal attempts) in order to establish a detail-based mechanism for analyzing the capability among competing teams. The different types of activation and inhibition actions between athletes provide a considerable challenge in the framework of network analysis. Moreover, these attributes interactions might add up the substantial redundancy to network frame as well. This paper proposes a weighted PageRank algorithm based on the normalized basketball match scores from a macroscopic point of view. Taking Chinese Basketball Association and Chinese University Basketball Association as examples, the developed approach takes into account the win/lose nature of interactions between each pair of competing teams in the framework of PageRank network. We also evolve a weighted network model for the network matrix which highlights the capability difference of teams whose PageRank probabilities are most sensitive with respect to the scores of the two competing teams. The chance of championship of teams can be better demonstrated by the PageRank probabilities. The results show that our method achieves more precise predicting result than that of original PageRank algorithm and Hypertext-Induced Topic Search algorithm.