Empirical Evaluation for Cricket CommentaryDecoder

Abstract Cricket is not just a sport but a global phenomenon, attracting billions of fans and generating immense revenue through broadcasting rights, streaming platforms, and fantasy sports applications. With the advent of professional leagues like the Indian Premier League (IPL) and Big Bash League (BBL), cricket has become a data-rich and analytics-driven game. Real- time event classification from live commentary is crucial for powering these ecosystems, from enhancing viewer experiences to enabling fantasy leagues and predictive analytics. This project aims to develop a real-time cricket commentary decoder to classify events accurately from unstructured and informal textual commentary. Cricket commentary often includes colloquial lan- guage, ambiguous phrases, and context-dependent expressions, which traditional rule-based Natural Language Processing (NLP) techniques struggle to handle effectively. The solution incorpo- rates innovative strategies to address sense disambiguation by intentionally removing clear event-indicative keywords, thereby simulating real-world complexity. By tackling the limitations of rule-based systems such as their over reliance on explicit keywords and inability to handle ambiguous contexts, the system achieves robust event classification for runs, outs, extras, and dots game actions. Designed to seamlessly integrate with real- time sports tracking applications, the decoder enables dynamic updates and API calls. By balancing computational efficiency with classification accuracy, this project not only addresses the technical challenges of decoding live commentary but also supports the growing role of data and analysis in the modern cricket ecosystem. To overcome these challenges, the proposed system employs advanced machine learning methods, including LSTM, GRU, SVM, Logistic classifiers, and boosting techniques like LightGBM.