Adaptive Visualization Framework for Real-Time User Engagement in Big Data

With the growth of data and the diversification of users' needs, traditional dashboards cannot support real-time interaction and multiple opinions. In this paper, a double closed-loop framework of “user perception-component scheduling” is proposed, which combines semantic-behavior fusion and Bandit-Mermaid algorithm to dynamically optimize the visualization of big data. An 8-week experiment with 3.8 TB logs and 1250 surveys shows that the delay is reduced by 97 milliseconds, the attention is increased by 28.4%, and the task completion rate is improved by 16.2%. There is a linear positive correlation between scheduling frequency and satisfaction (r = 0.41, P < 0.01). Bandit parameter ε = 0.06 optimizes the stability and click rate. The research results show that the framework can be applied to financial monitoring, e-commerce, and smart cities and provides a replicable model for adaptive visualization.