Studies of stock market prediction show that stock movements are related to the sentiment of social media. However, few studies have investigated the role of online social relations in predicting stock movements. This paper aims at constructing features that capture users’ online social status and incorporating these into stock prediction models. Online opinions are often developed through interactions and are weaker in their early stages. We developed a feature-enhancing procedure motivated by statistical surveillance approaches to strengthen the ability to capture emerging trends. We evaluated our feature-enhancing procedure by developing models to predict stock returns in the following 20-minute period. A comparison of experimental results with baseline models shows that our feature-enhancing design helped to predict stock movements. The model (SE_CUSUM) that adopted features enhanced by cumulative sum (CUSUM), a statistical surveillance approach, performed better than baseline models in terms of directional accuracy, balanced error rate, root mean square error, and mean absolute error. Our simulated trading also showed that SE_CUSUM realized a higher profit than the baseline approaches. These results suggest that incorporating online social status and our feature-enhancing procedure improve high frequency stock prediction performance.