Logic-Based Online Complex Event Rule Learning with Weight Optimization

Complex event recognition system process continuous data stream in real time and use the defined event rules to infer the occurrence of the event in time. The rules used by complex event recognition systems are generally defined by experts manually. However, the manual definition of rules is often complicated and error-prone. Logic-based event recognition system has been studied in recent years. Event calculus is a kind of temporal logic that has become a logic-based expression of event rules. It combines with statistical relationship learning to conduct reasoning and learning of events, effectively solving the complicated and error-prone problems of artificial construction rules. Currently, logic-based event rule learning is mostly combined with ILP for top-down search rules, and the training time is often long. The existing logic-based online rule learning algorithm has a long training time and a fast learning rate. We proposed a method based on root mean square prop (RMSprop), a weight learning method to improve the original algorithm. Moreover, we optimize the weight of the rule structure, and combine the existing rule structure learning method and weight learning method to learn rules. The optimized method has been obtained good results. Finally, we design a set of experiments to verify the feasibility and effectiveness of our method.