Text-mining based localisation of player-specific events from a game-log of cricket

Abstract

Analysis and visualisation of sports data pertaining to a particular player provides an avenue to study the game of a player (or group of players) in detail, to identify strengths and weaknesses to predict the outcome of a match or select players for a fantasy league. In this paper, we propose to detect and localise events (such as shots) specific to a player or a team and present visualisations that are easy to interpret. For this, we use a textual log of the match (or the detailed ball-by-ball commentary) of a game of cricket as input. We perform hierarchical n-gram matching to detect events of interest for the player(s) specified and match them with the location (on the field). We propose the disambiguation of action and intention verbs and resolve location through auxiliary information to improve the accuracy of detection and localisation of events. The results obtained on testing the proposed system on multiple game logs demonstrate that the accuracy of detection and localisation is comparable with that of manual detection and localisation.