Sentiment Analysis and Rating Video Game Dimensions via NLP
Liwei Yuan, Maoxiang Ding, Feiyang Meng, Yumin Tian · Applied and Computational Engineering · 2025
Online gaming platforms generate vast amounts of user comments, which serve as valuable information for potential users and purchasers. Extracting meaningful guiding information from this data is crucial. As a Natural Language Processing technique (referred to as NLP), sentiment analysis technology demonstrates a high efficiency in accurately discerning the emotional tones within the comments, thereby objectively evaluating the advantages and disadvantages of gaming products based on user comments. In this paper, the comments of two games are analyzed, Grand Theft Auto V (referred to as GTAV) and Cyberpunk 2077 (referred to as 2077) on the Steam platform. Based on the self-built sentiment dictionary, keywords representing different sentiments were extracted by applying two different sentiment analysis models which are VADER and TEXTBLOB. Then we apply two models to classify the sentiments that the keywords express. Finally, the ratings of five dimensions (community, gameplay, storyline, sound and graphic) were obtained by transforming the sentiment analyze score, and the differences in the training results of different models were compared. The result analysis shows the performance difference between the VADER model and TEXTBLOB model through analyzing the standardized scores and the Pearson correlation coefficient. And it was particularly noted that the VADER model is better at capturing emotional changes in game reviews, while the TEXTBLOB model may not fully represent the user’s emotional inclination towards various aspects related to the game. The results reveal significant differences in model performance, providing insights into their effectiveness for gaming comment analysis.