Effectiveness of Text Representations in the Automatic Classification of Regional Game Design Trends in Video Game Reviews
The video game industry is one of the fastest growing segments of the global entertainment market, and thus represents the design decisions of a wide spectrum of developers. This paper seeks to show that text mining can be used to predict trends in game design by identifying the region and release date automatically from video game reviews. When framed as a multi-class classification problem, a Support Vector Machine (SVM) achieves an average predictive confidence of 30.27% for noun and verb text representations, or by individual text representation: text windowing 11.22%, noun phrases 32.15%, noun phrases without game titles 31.40%, noun phrases with verbs 29.66%, individual term 27.88%. The SVM achieved better performance of 52.93% when predicting the release date trained on nouns and verbs. By text representation, the classifier found: noun phrases 62.97%, noun phrases without game titles 55.13%, noun phrases with verbs 61.10%, individual term 36.51% features.
School:University of North Carolina at Chapel Hill
School Location:USA - North Carolina
Source Type:Master's Thesis
Keywords:text mining natural language processing representation video games review
Date of Publication:11/17/2008