This is an open question. Dynamically adjusting standards should try to avoid the influence of history and reflect the latest customer recognition in a timely manner. You might as well define the customer's current recognition of a restaurant as the ratio of the restaurant's positive reviews after being recommended during the two dynamic adjustments. Just dynamically adjust the recommendation probability in proportion to the current recognition level.

For example, during this dynamic adjustment, the positive reviews of 5 restaurants since the last adjustment are:

• A: 10/40 (recommended 40 times, praised 10 times) = 0.25

• B: 20/30 = 0.67

• C: 3/10 = 0.3

• D: 10/15 = 0.67

• E: 2/5 = 0.4

Then the adjusted recommendation probability is updated to:

• A: 0.25 / (0.25 + 0.67 + 0.3 + 0.67 + 0.4) = 0.25 / 2.29 = 11%

• B: 0.67 / 2.29 = 29%

• C: 0.3 / 2.29 = 13%

• D: 0.67 / 2.29 = 29%

• E: 0.4 / 2.29 = 18%