An increasing number of people search online for health information or suggestions when they are confronted with health problems. However, identifying good suggestions when inundated with a wide variety of responses is a big challenge. Most online forums don't provide an automated suggestion-rating procedure for users to locate the quality suggestions that meet their expectations. In this study, we focus on the problem of identifying good suggestions. We propose a novel framework that accounts for the dynamic nature of social media by modeling the evolution of features over time. We use a combination of LSTM time series prediction of temporal features and Adaptive Thresholding Normalization to address this problem. Our study discusses why evolving language features need to be considered to determine the quality of suggestions. Besides, our method can identify important language features that can boost the prediction ability of the best suggestions.