Data exception attribution analysis method and apparatus

Abstract

The invention discloses a data exception attribution analysis method and apparatus. The method comprises the steps of obtaining first data of a to-be-monitored index; calculating an exception probability of data exception of the first data by utilizing a preset data model; if the exception probability is greater than or equal to a preset threshold, obtaining attribution index data; and based on the attribution index data, performing attribution analysis calculation by utilizing a preset attribution algorithm, and sending an attribution analysis result to a processing end. The manpower cost ofa company or an enterprise for data exception monitoring can be greatly reduced; the accuracy can be improved; and the method has very high universality. In addition, the dependency on business experience is greatly reduced; by quantizing the exception degree, a large amount of indexes can be automatically and accurately monitored, so that the data monitoring efficiency is improved, and the workload of analysis personnel is reduced; and by sending the attribution analysis result to the processing end, the processing end can perform fault removal, so that the fault removal efficiency is improved.

Description

technical field [0001] The present invention relates to the technical field of data processing, in particular to a data anomaly attribution analysis method and device. Background technique [0002] With the popularization of the Internet, companies in all walks of life are producing and depositing massive amounts of data all the time. In order to use these data, different companies extract a large number of different data indicators from the data according to the company's business preferences, which are used to measure the company's business development, the company's product quality, and so on. [0003] Taking the e-commerce website as an example, the data indicators to measure the company's business include: number of daily active users, daily page views, monthly active users, monthly page views, user click rate, user transaction conversion rate, user registration conversion rate, average user page views Volume, average browsing depth, average stay time, page stay time, ...

AI Technical Summary

Problems solved by technology

[0005] For the monitoring and analysis of data indicators, there are a lot of needs in all walks of life. The traditional way is to organize these indicators into tables or visualize them as curves, histograms or pie charts, etc., and manually view and analyze them; or monitor with simple statistical indicators Mainly, the judgment of abnormal data is relatively simple and depends on subjective experience

[0006] There are several problems here: first, some companies have a large number of business indicators (such as word of mouth), relying on analysts to observe and analyze one by one, the efficiency is very low, and it is easy to cause omissions due to negligence; Judgment mainly depends on personal experience, which leads to different judgment standards for data abnormalities by different analysts, which will eventually lead to different analysis results; third, when a certain data indicator is abnormal, such as the order conversion rate of an e-commerce website drops abnormally, When the cause is very direct and obvious, it may be easier to draw a conclusion by relying on the analyst's observation and analysis, but when the cause is intricate and relatively not obvious, it is often impossible to obtain an effective quantifiable attribution only by relying on the analyst's experience in conclusion

When there are relatively few data tables (fields) that need to be monitored, and the business experience is sufficient, and the above-mentioned simple statistics can show abnormalities that need to be monitored, relatively good results can be achieved; and for a large number of tables (fields) When monitoring, and the business experience is relatively insufficient, and the purpose of abnormal monitoring cannot be achieved only by simple statistics, effective data monitoring cannot be carried out (for example, the task of discovering abnormal store operations on the Koubei platform, due to the large number of stores (million level), The fluctuations of the corresponding indicators of each industry are inconsistent, and the fluctuation rules of the transaction volume of each store are also inconsistent, which makes it difficult to manually customize the rules, and the one-size-fits-all rules are often set unreasonably)

Images