The truth discovery problem involves identifying the truth associated with objects from observations, which is crucial in health recommendation systems. While many methods effectively calculate the truth from multiple data sources, few address the challenges posed by dynamically generated data. In dynamic environments, data continually arrive and require processing in time due to the limited memory of health monitoring devices. In this paper, we propose a novel approach to discovering truths for health recommendation systems that uses time-series analysis methods to mine the evolutionary patterns of truths and account for potential similarities between objects. Our approach efficiently manages dynamic data without compromising estimation accuracy. We evaluate the performance of our proposed framework on two real-world datasets and a synthetic dataset, demonstrating its superiority over state-of-the-art methods in terms of estimation accuracy and efficiency. This approach has great potential to advance the development of more accurate and personalized health recommendation systems.