| 摘要 |
PROBLEM TO BE SOLVED: To improve an advertising effect in a case where a lot of advertising contents exist as to one commercial material by predicting a click through rate (CRT) of each advertising content.SOLUTION: A content distribution device is configured to include means of: generating a characteristic amount of an advertising content for an individual advertising content of an advertising database managing a plurality of advertising contents as to one commercial material; selecting a group of advertising contents from the advertising database and initializing an initial distribution probability and a coefficient vector for each advertising content; calculating the coefficient vector and a prediction CTR; distributing the advertising content and recording an actual CTR; performing a reinforcement learning so as to maximize a distribution probability of one advertising content of a plurality of advertising contents with the prediction CTR during a predetermined period of time for each advertising content as a remuneration and leave a minimum distribution probability as to remaining advertising contents to calculate a new distribution probability for each advertising content and setting the new distribution probability; and updating the coefficient vector on the basis of the actual CTR. |
