How many queries are sufficient for training our machine learning model? (a) The box plot of the amount of queries containing keyword “Earthquake” with respect to different time intervals after an earthquake occurred in our dataset. Indeed, the amount of queries in first 5 minutes increases sharply, and becomes stable afterwards. (b) The x-axis denotes the length of time window for accumulating search queries, and y-axis denotes the training/test error of our model with corresponding queries through a standard five-fold cross validation. Indeed, after 5 minutes, the errors become relatively stable and convergent, which indicates that using more queries will not significantly influence the learning results.