We cannot see the specific situation with a single data. We need to combine category email list the previous data to find abnormal situations, such as the above data: Based on the previous data, we can find: not cyclical It suddenly dropped on June 13, and it has dropped for three days. It may continue to drop, and we need to be vigilant. From our data above, we can know that there is category email list indeed a problem with the daily activity. After completing the above steps, we proceed to data analysis. 3.
Divide the dimensions Now the problem is: the daily activity has category email list dropped by 10,000 or 20,000. So we need to split the daily active dimension: According to the split of new and old users; The split of the login platform, such as: IOS, Android; Split according category email list to APP version; According to the split of login channels, such as APP, applet; Split by region, such as: country, province;
First calculate the influence coefficient, and then compare category email list each item of data with the normal data in the past to calculate the influence coefficient. Formula: (Today’s Amount – Yesterday’s Amount) / The larger the coefficient, the greater the influence Now let's look at our data above, because the data suddenly dropped on June 13, so we use the category email list data of June 12 and June 13 to calculate the influence coefficient.