Natural language processing techniques have been confirmed to well classify the alarm messages based on the user intents. In this study, we propose an intelligent classification algorithm for burst alarm events. First, we analyze the features of user feedback text data. We then present a semi-automatic semantic mapping tool to calculate the probability, so as to form a valid alarm category. By application this method into the enterprise fault location, it takes advantages of (1) releasing alarms upon different user intents; (2) reducing the alarm miscalculation; and (3) helping game operation personnel accurately and quickly locating the reasons of the fault. The experimental results show that the intelligent classification algorithm based on semantic mapping can effectively improve the performance of the alarm distribution, and achieves better results than the simply algorithm for text categorization.