To facilitate automatic knowledge creation for the learners of intercultural communication, we propose a pipeline that combines AI analytics with human-in-the-loop evaluation. A case study was conducted on the English and Japanese versions of the film ‘Spirited Away’. Speech segments from both versions were extracted and processed using Valence-Arousal-Dominance (VAD) scores and discrete emotion classification, generated by pre-trained Wav2Vec 2.0 models. The segments with the highest discrepancies in emotions were identified for further analysis by a group of human raters. This approach not only improves the understanding of emotional differences in dubbed media but also benefits researchers studying cross-cultural communications by assisting knowledge discovery and the creation of knowledge pools. These insights may also contribute to the design of intelligent tutoring systems (ITS) for cross-cultural communication education, enhancing the overall learning experience for learners.