Abstract:
"The increasing popularity of gaming has led to a significant increase in the volume of
game reviews, making it challenging for game developers and players to monitor the
sentiment associated with a particular game. Sentiment analysis has grown in popularity
as a method for determining the sentiment connected with a piece of text, such as game
reviews. Even though sentiment analysis is widely used, issues such as use of negative
terminologies, and game context aware reviews are gaps that needed to be filled in this
domain and a valid approach is needed for this.
The proposed methodology involves utilizing BERT (Bidirectional Encoder
Representations from Transformers) with ensemble method to get polarity of reviews
using a contextual analysis approach. To further analyze the context of reviews a
BiLSTM layer and a CNN layer was added on top of the BERT. Creating an ensemble
method to improve the overall accuracy of the sentiment classification will be discussed
in this research. The approach is evaluated using a dataset of game reviews from Steam,
a popular online gaming platform.
This study advances the field of sentiment analysis and shows how transformer-based
hybrid models perform well in NLP applications. The results of the experiments show
that the suggested methodology works superior to traditional sentiment analysis
methods and achieves high accuracy in identifying the sentiment of game reviews by
achieving accuracy of 90% and f1-score 0f 0.9080. Using this companies can get more accurate information such as players satisfaction rate of the game which can be helpful in this challenging market."
