Generating dynamic content item recommendations

1. A method, comprising:

executing, on a processor of a computing device, instructions that cause the computing device to perform operations, the operations comprising:

generating machine generated messages, wherein the machine generated messages are generated by a machine to have information corresponding to a content item;

generating an extraction rule by executing a clustering method upon the machine generated messages;

performing online extraction of content item information from message data using the extraction rule, wherein the online extraction is performed using at least one of entity extraction or natural language processing;

aggregating the content item information extracted from the message data to calculate popularity and attributes of content items;

ranking the content items based upon the popularity and attributes of the content items to generate a ranked list of content items;

serving exploration traffic utilizing a set of eligible content items, selected from the ranked list of content items according to ranks of the content items, wherein a first model is used to generate scores for the set of eligible content items to be served for the exploration traffic based upon bid values and predicted interaction probabilities derived from ranks and attributes of the set of eligible content items;

in response to an eligible content item being served a threshold number of times, promoting the eligible content item for participation in auctions for serving non-exploration traffic using a second model, wherein the second model is trained with user engagement feedback indicating that one or more users at least one of engaged with or did not engage with one or more eligible content items of the set of eligible content items; and

in response to receiving a request for one or more content items:

based upon at least one of whether a sufficient amount of feedback has been received for the one or more content items or whether the second model has been adequately trained to recognize one or more types of users that will engage with the one or more content items, determining whether the request should be processed as (i) exploration traffic using the first model or (ii) non-exploration traffic for which user engagement is predicted using the second model trained with the user engagement feedback; and

processing the request to serve content using the first model or the second model in accordance with the determination.

2. The method of claim 1 , comprising:

excluding an attribute from the attributes of the content items based upon at least one of a manually generated exclusion rule or distributed hierarchical clustering.

3. The method of claim 1 , comprising:

excluding an attribute from the attributes of the content items based upon locality-sensitive hashing.

4. The method of claim 1 , comprising:

determining global popularity for a content item.

5. The method of claim 1 , comprising:

determining a first demographic popularity for a content item with respect to a first demographic of users and second demographic popularity for the content item with respect to a second demographic of users.

6. The method of claim 1 , comprising:

generating an array of attributes for a content item based upon extracted message data.

7. The method of claim 1 , comprising:

determining a category of a content item as an attribute for the content item based upon extracted message data.

8. The method of claim 1 , wherein the online extraction is performed using entity extraction.

9. The method of claim 1 , wherein the online extraction is performed using natural language processing.

10. A computing device comprising:

a processor; and

memory comprising processor-executable instructions that when executed by the processor cause performance of operations, the operations comprising:

generating machine generated messages, wherein the machine generated messages are generated by a machine to have information corresponding to a content item;

generating an extraction rule by executing a clustering method upon the machine generated messages;

performing online extraction of content item information from message data using the extraction rule, wherein the online extraction is performed using at least one of entity extraction or natural language processing;

assigning ranks to content items based upon popularity and attributes of the content items, identified from the content item information, to generate a ranked list of content items;

serving exploration traffic utilizing a set of eligible content items, selected from the ranked list of content items according to the ranks of the content items, wherein a first model is used to generate scores for the set of eligible content items to be served for the exploration traffic based upon bid values and predicted interaction probabilities derived from ranks and attributes of the set of eligible content items;

in response to an eligible content item being served a threshold number of times, promoting the eligible content item for participation in auctions for serving non-exploration traffic using a second model, wherein the second model is trained with user engagement feedback indicating that one or more users at least one of engaged with or did not engage with one or more eligible content items of the set of eligible content items; and

in response to receiving a request for one or more content items, processing the request as (i) exploration traffic using the first model or (ii) non-exploration traffic for which user engagement is predicted using the second model trained with the user engagement feedback, wherein the processing is based upon at least one of whether a sufficient amount of feedback has been received for the one or more content items or whether the second model has been adequately trained to recognize one or more types of users that will engage with the one or more content items.

11. The computing device of claim 10 , comprising:

determining an attribute of a content item as a price of the content item.

12. The computing device of claim 10 , comprising:

determining an attribute of a content item based upon whether the content item is a premium product having a price exceeding a threshold.

13. The computing device of claim 10 , comprising:

determining an attribute of a content intent identified from a message associated with a user based upon an age of the user.

14. The computing device of claim 10 , comprising:

determining an attribute of a content intent identified from a message associated with a user based upon a location of the user.

15. The computing device of claim 10 , comprising:

determining an attribute of a content intent identified from a message associated with a user based upon a gender of the user.

16. The computing device of claim 10 , wherein the online extraction is performed using entity extraction.

17. A non-transitory machine readable medium having stored thereon processor-executable instructions that when executed cause performance of operations, the operations comprising:

generating machine generated messages, wherein the machine generated messages are generated by a machine to have information corresponding to a content item;

generating an extraction rule by executing a clustering method upon the machine generated messages;

performing online extraction of content item information from message data using the extraction rule, wherein the online extraction is performed using at least one of entity extraction or natural language processing;

assigning ranks to content items based upon popularity and attributes of the content items, identified from the content item information, to generate a ranked list of content items;

serving exploration traffic utilizing a set of eligible content items, selected from the ranked list of content items according to the ranks of the content items, wherein a first model is used to generate scores for the set of eligible content items to be served for the exploration traffic based upon bid values and predicted interaction probabilities derived from ranks and attributes of the set of eligible content items;

selecting a subset of the set of eligible content items as a set of promoted content items based upon performance of the set of eligible content items for serving the exploration traffic, wherein a second model is trained with user engagement feedback indicating that one or more users at least one of engaged with or did not engage with one or more eligible content items of the set of eligible content items;

utilizing the second model trained with the user engagement feedback to generate scores for the set of promoted content items to be served for non-exploration traffic; and

in response to receiving a request for one or more content items, processing the request as (i) exploration traffic using the first model or (ii) non-exploration traffic for which user engagement is predicted using the second model trained with the user engagement feedback, wherein the processing is based upon at least one of whether a sufficient amount of feedback has been received for the one or more content items or whether the second model has been adequately trained to recognize one or more types of users that will engage with the one or more content items.

18. The non-transitory machine readable medium of claim 17 , wherein the operations comprise:

weighting the first model based upon the popularity of the content items.

19. The non-transitory machine readable medium of claim 17 , wherein the operations comprise:

weighting the first model based upon the attributes of the content items.

20. The non-transitory machine readable medium of claim 17 , wherein the message data comprises at least one of a receipt, a delivery confirmation, a recommendation, or an order confirmation, and wherein a content item comprises at least one of a product, a service, or a media item.