Method and system of personalized blending for content recommendation
The present teaching relates to personalized content recommendation. A webpage is contrasted for a user having a plurality of slots each of which is to be allocated with a content item. For each of the plurality of slots, a plurality of content items in a plurality of types of content are accessed. For each of the plurality of types of content, a personalized score is predicted for each content item in the type of content, wherein the personalized score is obtained based on a trained model trained. A recommended content item of the type of content is selected based on personalized scores. An overall recommended content item is selected and allocated to a slot based on criteria associated with the personalized scores of the recommended content items and a business rule. The webpage with the plurality of slots allocated with content items is provided to the user.
1 . A method, implemented on a machine having at least one processor, storage, and a communication platform capable of connecting to a network for training a model used for estimating a reward for a content item, the method comprising:
obtaining, by a training engine, training data which provides a guided exploration mechanism that efficiently balances between exploration and exploitation strategies;
training, by the training engine via machine learning, the model by performing training iterations each of which includes:
selecting, from a plurality of actions, a first action associated with the content item according to a policy, wherein the first action has been previously selected, and wherein the first action corresponds to selecting the content item from one of a plurality of data corpuses and allocating the selected content item at a first slot on a first webpage, wherein the plurality of data corpuses store different types of content items, respectively,
obtaining a reward associated with the first action,
updating the model based on the obtained reward,
updating the policy based on at least one of the first action, the obtained reward, and the obtained training data,
choosing a second action and obtaining information about the second action, wherein the second action has not been previously chosen, and
updating the model and the policy based on the information about the second action;
selecting, based on the model and from the plurality of data corpuses, at least one content item to be displayed at one of slots on a second webpage;
validating the at least one content item based on a rule indicating that a type of the content item is allowed to be displayed at the one slot on the second webpage;
allocating the validated content item to the one slot; and
repeating the selecting, validating and allocating until all the slots on the second webpage are allocated with the corresponding validated content items.
2 . The method of claim 1 , further comprising revealing contextual information in each of the training iterations, wherein the selecting the first action in the training interaction is further based on the revealed contextual information.
3 . The method of claim 2 , wherein the policy is mapping from the revealed contextual information to the first action.
4 . The method of claim 1 , wherein the reward is a proxy for a level of user satisfaction.
5 . The method of claim 1 , wherein the training data comprises a plurality of training data-records each of which is associated with a corresponding weight.
6 . A non-transitory, computer-readable medium having information recorded thereon for training a model used for estimating a reward for a content item, wherein the information, when read by at least one processor, effectuate operations comprising:
obtaining training data which provides a guided exploration mechanism that efficiently balances between exploration and exploitation strategies;
training, via machine learning, the model by performing training iterations each of which includes:
selecting, from a plurality of actions, a first action associated with the content item according to a policy, wherein the first action has been previously selected, and wherein the first action corresponds to selecting the content item from one of a plurality of data corpuses and allocating the selected content item at a first slot on a first webpage, wherein the plurality of data corpuses store different types of content items, respectively,
obtaining a reward associated with the first action,
updating the model based on the obtained reward,
updating the policy based on the first action, the obtained reward, and the obtained training data,
choosing a second action and obtaining information about the second action, wherein the second action has not been previously chosen, and
updating the model and the policy based on the information about the second action;
selecting, based on the model and from the plurality of data corpuses, at least one content item to be displayed at one of slots on a second webpage;
validating the at least one content item based on a rule indicating that a type of the content item is allowed to be displayed at the one slot on the second webpage;
allocating the validated content item to the one slot; and
repeating the selecting, validating and allocating until all the slots on the second webpage are allocated with the corresponding validated content items.
7 . The medium of claim 6 , wherein the operations further comprise revealing contextual information in each of the training iterations, and
wherein the selecting the first action in the training interaction is further based on the revealed contextual information.
8 . The medium of claim 7 , wherein the policy is mapping from the revealed contextual information to the first action.
9 . The medium of claim 6 , wherein the reward is a proxy for a level of user satisfaction.
10 . The medium of claim 6 , wherein the training data comprises a plurality of training data-records each of which is associated with a corresponding weight.
11 . A system for training a model used for estimating a reward for a content item, the system comprising:
memory storing computer program instructions; and
one or more processors that, in response to executing the computer program instructions, effectuate operations comprising:
obtaining training data which provides a guided exploration mechanism that efficiently balances between exploration and exploitation strategies;
training, via machine learning, the model by performing training iterations each of which includes:
selecting, from a plurality of actions, a first action associated with the content item according to a policy, wherein the first action has been previously selected, and wherein the first action corresponds to selecting the content item from one of a plurality of data corpuses and allocating the selected content item at a first slot on a first webpage, wherein the plurality of data corpuses store different types of content items, respectively,
obtaining a reward associated with the first action,
updating the model based on the obtained reward,
updating the policy based on at least one of the first action, the obtained reward, and the obtained training data,
choosing a second action and obtaining information about the second action, wherein the second action has not been previously chosen, and
updating the model and the policy based on the information about the second action;
selecting, based on the model and from the plurality of data corpuses, at least one content item to be displayed at one of slots on a second webpage;
validating the at least one content item based on a rule indicating that a type of the content item is allowed to be displayed at the one slot on the second webpage;
allocating the validated content item to the one slot; and
repeating the selecting, validating and allocating until all the slots on the second webpage are allocated with the corresponding validated content items.
12 . The system of claim 11 , wherein the operations further comprise revealing contextual information in each of the training iterations, and
wherein the selecting the first action in the training interaction is further based on the revealed contextual information.
13 . The system of claim 12 , wherein the policy is mapping from the revealed contextual information to the first action.
14 . The system of claim 11 , wherein the reward is a proxy for a level of user satisfaction.
15 . The method of claim 1 , wherein the different types comprise images, videos, documents, and slideshows.
16 . The medium of claim 6 , wherein the different types comprise images, videos, documents, and slideshows.
17 . The system of claim 11 , wherein the different types comprise images, videos, documents, and slideshows.