A quantitative study on user interaction and brand perception of visual elements in social media advertising driven by deep learning
- 1Hebei Vocational University of Technology and Engineering
Abstract
This paper focuses on the field of sentiment analysis for social media advertisements, specifically investigating the quantitative impact of visual design elements—such as color saturation, layout complexity, and image type—on user engagement metrics and brand perception metrics. Current challenges in this domain include insufficient quantification of visual features, inadequate multimodal data fusion, and lack of multi-objective collaborative optimization. Traditional approaches struggle to simultaneously meet demands for sentiment analysis accuracy, user engagement enhancement, and brand perception improvement. To address this, we propose VS-EmoNet (visual-sentiment emotion network)—a deep learning model integrating quantified visual design elements. By leveraging refined visual feature extraction, multimodal attention fusion, and multi-objective optimization mechanisms, VS-EmoNet achieves synergistic optimization of sentiment analysis with user behavior and brand perception. Experimental results demonstrate that this model achieves an average sentiment analysis accuracy of 92.5% across datasets from three major platforms: WeChat, Douyin, and Xiaohongshu. In high-volatility scenarios—where visual elements fluctuate by ± 30%—user click-through rates increased by 34.8%, Brand recognition enhancement reached 28.2%, with a multi-objective hypervolume indicator of 0.352 (normalized to [0,1]), indicating high-quality Pareto front approximation. This effectively quantifies the influence weight of core visual elements, providing data-driven support for social media ad design.
Related
- Common elements-based nurturing care intervention to improve mother-infant responsive interaction and maternal psychological wellbeing in rural Pakistan: findings of a single-blinded individual randomised controlled trial.
- Postural reeducation and oral device interaction in rhythmic gymnasts: effects of byte use on stiffness-related performance and low back pain perception
- Perception of Cardio-Kidney-Metabolic Risk and Health Literacy in Women: Quantitative and Qualitative Insights into Lifestyle and Prevention
- Prediction of Spinal Pain Intensity from Perception-Dependent Stimulation Thresholds and Patient Characteristics: A Regression and Machine Learning Approach
- OQ-FERNet: An occlusion-aware quantitative facial expression recognition network for real-time human-computer interaction
BibTeX
@article{wang2026quantitative,
title = {{A quantitative study on user interaction and brand perception of visual elements in social media advertising driven by deep learning}},
author = {Pei Wang and Xiaopeng Niu and Aiping Li},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-61139-9},
url = {https://doi.org/10.1038/s41598-026-61139-9},
note = {CC BY},
}
RIS
TY - JOUR
TI - A quantitative study on user interaction and brand perception of visual elements in social media advertising driven by deep learning
AU - Pei Wang
AU - Xiaopeng Niu
AU - Aiping Li
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-61139-9
UR - https://doi.org/10.1038/s41598-026-61139-9
AB - This paper focuses on the field of sentiment analysis for social media advertisements, specifically investigating the quantitative impact of visual design elements—such as color saturation, layout complexity, and image type—on user engagement metrics and brand perception metrics. Current challenges in this domain include insufficient quantification of visual features, inadequate multimodal data fusion, and lack of multi-objective collaborative optimization. Traditional approaches struggle to simultaneously meet demands for sentiment analysis accuracy, user engagement enhancement, and brand perception improvement. To address this, we propose VS-EmoNet (visual-sentiment emotion network)—a deep learning model integrating quantified visual design elements. By leveraging refined visual feature extraction, multimodal attention fusion, and multi-objective optimization mechanisms, VS-EmoNet achieves synergistic optimization of sentiment analysis with user behavior and brand perception. Experimental results demonstrate that this model achieves an average sentiment analysis accuracy of 92.5% across datasets from three major platforms: WeChat, Douyin, and Xiaohongshu. In high-volatility scenarios—where visual elements fluctuate by ± 30%—user click-through rates increased by 34.8%, Brand recognition enhancement reached 28.2%, with a multi-objective hypervolume indicator of 0.352 (normalized to [0,1]), indicating high-quality Pareto front approximation. This effectively quantifies the influence weight of core visual elements, providing data-driven support for social media ad design.
SN - 2045-2322
ER -