Efficient processing of continuous queries utilizing F-relationship in stock databases

This paper analyzes the properties of user queries in a system for stock investment recommendation, and defines the F-relationship, which is a new type of a relationship between two queries. A query is composed of user defined conditions for an interesting stock item and is systematically invoked whenever the price of that stock item is changed. An F-relationship between two queries Q1 and Q2 means that, if the recommendation type of a preceding query Q1 is X, then its following query Q2 always has X as its recommendation type, where the recommendation type is one of SELL, HOLD, BUY, and NONE. If there is an F-relationship between Q1 and Q2, the recommendation type of Q2 is decided immediately by that of Q1, therefore we can keep Q2 from being actually processed. To exploit this fact, we suggest two methods in this paper. The former analyzes all the F-relationships among user queries in the system and represents them as a graph. The latter searches the graph and decides the order of queries to be processed, which makes the number of unexecuted queries maximized. With these methods, a large portion of user queries are not actually processed. As a result, the performance of processing all the queries is greatly improved. We examined the superiority of the suggested methods through a variety of experiments using real-world stock market data. According to the results of our experiments, the overall time of continuous query processing with our proposed methods has reduced to less than 10% of that with the traditional method.