Aspect Based Sentiment Analysis Terhadap Ulasan Game Buatan Developer Indonesia di Steam Menggunakan Algoritma Logistic Regression

The Indonesian gaming industry is thriving on the Steam platform, but the large volume of reviews makes it difficult for developers to manually identify the product's strengths and weaknesses. This study applied Aspect Based Sentiment Analysis (ABSA) using the Logistic Regression algorithm to analyze five main aspects: Story, Gameplay, Graphics, Music, and Performance. The study collected 3,046 raw reviews from five popular local games which were then extracted into 7,175 opinion-aspect pairs through the ABSA process. After going through the labeling stage, 5,770 valid data were obtained consisting of 3,671 positive sentiments and 2,099 negative sentiments, while the other 1,405 data were categorized as nonlabeled (NaN) because they did not meet clear sentiment criteria. The test results of 1,731 test data showed that the Logistic Regression model without class balancing produced solid performance with an accuracy rate of 84.17%. This model is very effective in recognizing positive sentiment with a Positive Recall value of 92.73%, but faces challenges in the minority class with a Negative Recall of 69.21%. Sentiment analysis per aspect revealed the main strengths of local games in the aspects of Graphics (77.4% positive), Music (74.7% positive), and Story (71.7% positive). On the other hand, the Performance aspect was the main weak point with the dominance of negative sentiment at 63.4% due to technical obstacles such as bugs and optimization. These findings provide strategic insights for developers to prioritize improvements in technical performance to improve product competitiveness in the global market.