A quantitative study on user interaction and brand perception of visual elements in social media advertising driven by deep learning.
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.
Paper
Full text
A quantitative study on user interaction and brand perception of visual elements in social media advertising driven by deep learning.
Semantic Scholar · Computer Science · 2026
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.