Article: Emotion computing for narrative salience in children's stage performances based on adaptive attention flows Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.89 pp.20 - 38 Abstract: In the emotional understanding of children's stage narratives, children's emotional expressions are often embedded within the narrative flow of the performance. However, existing methods focus on isolated facial expressions or speech features, ignoring the dynamic evolution of emotions over time and their significant migration within the narrative context. This paper proposes the adaptive attention flow network, which introduces an attention flow field based on conservative flux constraints to dynamically track key emotional nodes across time, space, and narrative. The attention weights are regarded as information flows that can be transferred across frames, ensuring smooth evolution and lossless transmission of emotional representations along the narrative timeline. Experiments show that the classification accuracy of the adaptive attention flow network reaches 84.6% with an area under the curve of 0.91, outperforming the timesformer baseline (74.2% accuracy and 0.83 area under the curve) by 10.4 percentage points and 0.08, respectively. Inderscience Publishers - linking academia, business and industry through research

Title: Emotion computing for narrative salience in children's stage performances based on adaptive attention flows

Authors: Ruiyi Yang

Addresses: School of Media and Communication, Weinan Normal University, Weinan, 714099, China

Abstract: In the emotional understanding of children's stage narratives, children's emotional expressions are often embedded within the narrative flow of the performance. However, existing methods focus on isolated facial expressions or speech features, ignoring the dynamic evolution of emotions over time and their significant migration within the narrative context. This paper proposes the adaptive attention flow network, which introduces an attention flow field based on conservative flux constraints to dynamically track key emotional nodes across time, space, and narrative. The attention weights are regarded as information flows that can be transferred across frames, ensuring smooth evolution and lossless transmission of emotional representations along the narrative timeline. Experiments show that the classification accuracy of the adaptive attention flow network reaches 84.6% with an area under the curve of 0.91, outperforming the timesformer baseline (74.2% accuracy and 0.83 area under the curve) by 10.4 percentage points and 0.08, respectively.

Keywords: emotional computing; children's stage narrative; attention flow; multimodal fusion.

DOI: 10.1504/IJICT.2026.155694

International Journal of Information and Communication Technology, 2026 Vol.27 No.89, pp.20 - 38

Received: 07 Apr 2026
Accepted: 26 May 2026
Published online: 10 Aug 2026 *

Title: Emotion computing for narrative salience in children's stage performances based on adaptive attention flows

Keep up-to-date