Bullet screens are attracting increasing attention as a way to express emotions and interact on short video platforms. Prior studies have used natural language processing (NLP) to analyze bullet screen sentiment in order to evaluate public opinion trends regarding a specific topic, movie, or product. However, few studies have investigated the effectiveness of using bullet screen sentiment analysis to predict real-time emotional responses. Thus, this study examined whether and to what extent bullet screen sentiment analysis can be used to evaluate and predict real-time emotional responses to videos by employing physiological electrodermal activity (EDA) measurements. A behavioral experiment was conducted in which eight college students wore a set of wireless galvanic skin sensors while watching three music videos (MVs) in random or-der. The participants’ EDA data, including skin conductance responses and peak amplitudes, were then analyzed. Meanwhile, the sentiments expressed in the bullet screen comments on the three MVs were analyzed using three dictionary-based sentiment analysis algorithms: SnowNLP, BosonNLP, and Hel-loNLP. The bullet screen sentiment analysis and physiological measurement results were then compared using descriptive and correlation analyses. The bullet screen sentiment parameters were found to significantly correlate with the EDA measurements. This study confirms the effectiveness of using bullet screen sentiment analysis to predict participants’ real-time emotional responses, providing a convenient and flexible way for enterprises and governments to detect public opinion trends and take action accordingly.