Sentiment Analysis of User Satisfaction Toward AI Medical Chatbot Services Using a Natural Language Processing Approach

The rapid development of artificial intelligence has led to the increasing adoption of AI-based medical chatbots to provide accessible health information and preliminary consultation. Understanding the emotional tone and communication characteristics of such systems is essential to evaluate their effectiveness and reliability. This study aims to analyze sentiment distribution, interaction dynamics, and communication depth within an AI medical chatbot dataset using a lexicon-based Natural Language Processing approach. Sentiment classification was performed on doctor responses using the VADER sentiment analyzer, categorizing responses into positive, negative, and neutral classes. Cross-tabulation analysis and a Chi-square test were conducted to examine the relationship between patient and doctor sentiment. Additionally, response length analysis was performed to assess communication depth, and word cloud visualization was used to identify dominant lexical patterns. The results indicate that the majority of doctor responses are classified as positive (65.17%), followed by negative (27.36%) and neutral (7.47%). A statistically significant association between patient and doctor sentiment was identified (χ² = 197.98, p < 0.001), suggesting systematic interaction patterns. Despite variations in patient sentiment, doctor responses consistently maintain a supportive tone. Furthermore, positive responses are associated with longer message lengths, indicating more elaborate and reassuring communication. Negative sentiment classifications were largely influenced by clinical terminology rather than interpersonal negativity. Overall, the findings demonstrate that AI-mediated medical communication prioritizes reassurance and structured guidance. This study contributes empirical insights into emotional and structural patterns in medical chatbot interactions and highlights the importance of contextual interpretation in domain-specific sentiment analysis.