Article: Tracking the evolution of youth ideological public opinion based on multimodal transformer and SHAP attribution Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.50 pp.17 - 34 Abstract: Existing methods for tracking the evolution of ideological and political public opinion struggle to fully uncover intermodal correlations and exhibit low tracking accuracy. To address this, this paper first employs the Shapley additive explanations algorithm optimised by random forests to screen key influencing indicators. These selected indicators undergo a Gramian angular field transformation to generate a two-dimensional image. Subsequently, the Shapley additive explanations optimises the self-attention mechanism of the Transformer model while enhancing locally significant features that substantially influence tracking outcomes. The improved transformer model and bidirectional encoder representations from transformers model are employed to extract image and text features, respectively. Contrastive learning is introduced to align features across modalities. Multimodal fusion features undergo classification via the softmax function, enabling the tracking of public opinion evolution. Experimental results demonstrate that the proposed model achieves a tracking accuracy of 92.1%, exhibiting outstanding tracking efficiency. Inderscience Publishers - linking academia, business and industry through research

Title: Tracking the evolution of youth ideological public opinion based on multimodal transformer and SHAP attribution

Authors: Dan Yang

Addresses: School of Marxism, Suzhou Polytechnic University, Suzhou, 215104, China

Abstract: Existing methods for tracking the evolution of ideological and political public opinion struggle to fully uncover intermodal correlations and exhibit low tracking accuracy. To address this, this paper first employs the Shapley additive explanations algorithm optimised by random forests to screen key influencing indicators. These selected indicators undergo a Gramian angular field transformation to generate a two-dimensional image. Subsequently, the Shapley additive explanations optimises the self-attention mechanism of the Transformer model while enhancing locally significant features that substantially influence tracking outcomes. The improved transformer model and bidirectional encoder representations from transformers model are employed to extract image and text features, respectively. Contrastive learning is introduced to align features across modalities. Multimodal fusion features undergo classification via the softmax function, enabling the tracking of public opinion evolution. Experimental results demonstrate that the proposed model achieves a tracking accuracy of 92.1%, exhibiting outstanding tracking efficiency.

Keywords: ideological and political public opinion tracking; transformer model; SHAP algorithm; multimodal feature fusion; contrastive learning.

DOI: 10.1504/IJICT.2025.151073

International Journal of Information and Communication Technology, 2025 Vol.26 No.50, pp.17 - 34

Received: 25 Sep 2025
Accepted: 25 Oct 2025
Published online: 12 Jan 2026 *

Title: Tracking the evolution of youth ideological public opinion based on multimodal transformer and SHAP attribution

Keep up-to-date