Huo Yan is used to detect outliers from stock data. It constructs a dynamic graph based on the stock data and financial indices to detect the abnormal time interval. This system is able to help people discover the abnormal patterns of the stock.

Fig. 1. Example of time-series abnormality

The outlier is defined as a time interval in which relationships are significant different with other time intervals. In Fig. 1, we denote S_i, t_k as the i-th financial index and k-th time interval, respectively. In addition, S_f is a stock. The relationship graph at 7-th time interval are significant different with other time intervals. Therefore, the 7-th time interval will be detected as a outlier.

Our idea is constructing a dynamic relationship graph by discovering the spurious relationship among the stock and financial indices, which indicates that two time series are correlated but not causally related. Then, we use graph embedding method to map the dynamic relationship matrix into an embedding space. Thereby, outliers can be detected by using statistical methods. The process of our work are

- discovering spruious relationships among the stock and financial indices,
- constructing the dynamic relationship matrix,
- applying graph embedding method for the dynamic relationship matrix,
- and detecting abnormal time intervals by using statistical method.