1. A / B test
Recommended models before the new on-line to an A / B test, compared with the old algorithm. AB testing is a very common experimental methods online evaluation algorithm. It rules by certain users randomly divided into several groups, and to different groups of users using different algorithms, and then through various statistical indicators evaluating different groups of users to compare different algorithms,
For example, users can set different statistics of hits, click-through rate by comparing the performance of different algorithms.
After the user enters the site, flow distribution system needs to determine whether the user is an A / B test, if required, flow distribution system will give users what to tag packets belonging to the test. The user then browse the web, and user behavior while browsing web pages will be sent back by the background logging database log system. At this point, if
Households label test packet, then the label will be sent back to the back-end database. In the background, first of all laboratory personnel work flow distribution system configuration, user decide what conditions to meet to participate in any kind of test. Second, the need to test statistics log data in the database, generating a test report different packet system user through the evaluation and comparison and evaluation results. After completion of the AB test, according to the results of indicators, better than before if the recommendation algorithm, the old algorithm can be replaced with a new one.
3. References
https://juejin.im/post/5d75ad14f265da03c02c2827