Keywords: political debate, automatic speech recognition, speaker diarization, automatic topic identification, artificial intelligence, sentiment analysis, emotional impact

Understanding communication dynamics is key to analyzing messages effects, especially in polarized contexts. Automatically identifying speakers, topics, and sentiments during interactions offers powerful insights into how messages shape perceptions and audience reactions. When combined with psychophysiological or declarative data on viewers' emotional responses, this approach enhances the assessment of the discourse's impact. This approach enhances communication analysis and supports applications in political communication, data science, and social psychology. This study focuses on the political debate between Kamala Harris and Donald Trump.
This study follows a structured process to analyze speeches and assess emotional impact. It begins with word-level automatic speech recognition. Next, speaker diarization is applied, identifying timestamps corresponding to each speaker. Subsequently, the transcribed texts are matched to their respective speakers. Texts are segmented into sentences for sentiment analysis, classifying the emotional tone as positive, negative, or neutral. Individuals' emotional impact data is then integrated for a comprehensive analysis of the debate and audience reactions. Psychophysiological data was gathered using the James Four sensor, which measures electrodermal activity and photoplethysmography, paired with a mobile app recording declarative responses ("Like" or "Dislike"). Recruitment was conducted through the MindProber Labs community of panelists. The sample consisted of 466 participants, with a mean age of 44.80 ± 12.50 years, comprising 44.4% Democrats, 29.4% Independents, and 23.6% Republicans.
The sentiment analysis revealed that Donald Trump uses a negative tone in 70% of his speech, compared to Kamala Harris with 51%. Positive sentiment accounts for 19% of Trump’s speech versus 26% for Harris, while neutral sentiment is 12% for Trump and 22% for Harris.
Donald Trump triggers a higher overall Emotional Impact Score (EIS=641.05) and more negative valence (Dial=-0.0049) compared to Kamala Harris's (EIS=623.77; Dial=0.0038). This shows that Trump has a greater emotional impact but with a more negative valence in his interventions.
Additionally, the analysis identified economy (EIS=715.18; Dial=-0.0014), immigration (EIS=697.48; Dial=-0.0065), and civil rights (EIS=689.79; Dial=-0.0010) as the topics with the greatest impact during the debate. Furthermore, statistical analysis indicates that the topic, speaker, and their interaction have significant effects on EIS. However, sentiment, whether considered independently or in interaction with other factors, does not significantly influence the results.
These findings highlight the importance of integrating emotional and psychophysiological data to understand audience reactions, offering valuable insights into political message perception and impact in polarized contexts. By advancing methods to assess emotional engagement and topic-specific influences, this research strengthens political communication studies and lays a foundation for future exploration of message delivery, audience perception, and emotional impact.