Predictive recommendation system using contextual relevance
In general, embodiments of the present invention provide systems, methods and computer readable media for ranking promotions selected for recommendation to consumers based on predictions of promotion performance and consumer behavior. In embodiments, a set of promotions to be recommended to a consumer can be sorted and/or ranked according to respective relevance scores representing a probability that the consumer's behavior in response to the promotion will match a ranking target. In embodiments, calculating scores is based on a relevance model (a predictive function) derived from one or more contextual data sources representing attributes of promotions and consumer behavior. In embodiments, an absolute relevance score represents an absolute prediction of a ranking target variable. In embodiments, absolute relevance may be used to determine personalized local merchant discovery frontiers; featured result set thresholding for impressions; and/or promotion notification triggers. In embodiments, predictive models based on gross revenue may be optimized using promotion category-dependent price boosting.
1. A system, comprising one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to:
receive, from a computing device associated with a consumer identifier, input data representing a request for available promotions;
receive user-promotion interaction data describing attributes of historical interactions with respect to at least one promotion of a set of available promotions;
in circumstances where an activation state of the computing device satisfies a defined criterion, generate a contextual relevance score for each available promotion of the set of available promotions based at least in part on a gross revenue relevance model and features representing one or more of the input data or the user-promotion interaction data;
determine a subset of available promotions from the set of available promotions, wherein the subset of available promotions comprises available promotions with highest contextual relevance scores among contextual relevance scores calculated for the set of available promotions; and
transmit the subset of available promotions to the computing device.
2. The system of claim 1 , wherein the gross revenue relevance model is a random forest model that employs the features representing one or more of the input data or the user-promotion interaction data.
3. The system of claim 1 , wherein the gross revenue relevance model is a neural network model that employs the features representing one or more of the input data or the user-promotion interaction data.
4. The system of claim 1 , wherein the one or more storage devices store instructions that are operable, when executed by the one or more computers, to further cause the one or more computers to:
generate a promotion freshness feature vector representing the features; and
map the promotion freshness feature vector, based at least in part on the gross revenue relevance model, to a probability class of a set of probability classes.
5. The system of claim 4 , wherein the one or more storage devices store instructions that are operable, when executed by the one or more computers, to further cause the one or more computers to:
calculate the contextual relevance score based at least in part on the probability class.
6. The system of claim 1 , wherein the gross revenue relevance model is generated based at least in part on at least one contextual data source associated with at least one location related to the computing device.
7. The system of claim 1 , wherein the gross revenue relevance model is generated based at least in part on at least one contextual data source associated with at least one category of interest related to at least one location of the computing device.
8. The system of claim 1 , wherein the gross revenue relevance model is generated based at least in part on at least one contextual data source associated with at least one historical promotion category interest level related to the user- promotion interaction data.
9. A computer-implemented method, comprising:
receiving, by a device comprising a processor, input data from a computing device associated with a consumer identifier, the input data representing a request for available promotions;
receiving, by the device, user-promotion interaction data describing attributes of historical interactions with respect to at least one promotion of a set of available promotions;
in circumstances where an activation state of the computing device satisfies a defined criterion, generating, by the device, a contextual relevance score for each available promotion of the set of available promotions based at least in part on a gross revenue relevance model and features representing one or more of the input data or the user-promotion interaction data;
determining, by the device, a subset of available promotions from the set of available promotions, wherein the subset of available promotions comprises available promotions with highest contextual relevance scores among contextual relevance scores calculated for the set of available promotions; and
transmitting, by the device, the subset of available promotions to the computing device.
10. The computer-implemented method of claim 9 , wherein the gross revenue relevance model is a random forest model, and in circumstances where the activation state of the computing device satisfies the defined criterion, the calculating comprises calculating the contextual relevance score for each available promotion of the set of available promotions based at least in part on the random forest model and the features representing one or more of the input data or the user-promotion interaction data.
11. The computer-implemented method of claim 9 , wherein the gross revenue relevance model is a neural network model, and in circumstances where the activation state of the computing device satisfies the defined criterion, the calculating comprises calculating the contextual relevance score for each available promotion of the set of available promotions based at least in part on the neural network model and the features representing one or more of the input data or the user-promotion interaction data.
12. The computer-implemented method of claim 9 , further comprising:
generating, by the device, a promotion freshness feature vector representing the features; and
mapping, by the device, the promotion freshness feature vector, based at least in part on the gross revenue relevance model, to a probability class of a set of probability classes.
13. The computer-implemented method of claim 12 , wherein the one or more storage devices store instructions that are operable, when executed by the one or more computers, to further cause the one or more computers to:
calculating, by the device, the contextual relevance score based at least in part on the probability class.
14. The computer-implemented method of claim 9 , further comprising:
generating, by the device, the gross revenue relevance model based at least in part on at least one contextual data source associated with at least one location related to the computing device.
15. The computer-implemented method of claim 9 , further comprising:
generating, by the device, the gross revenue relevance model based at least in part on at least one contextual data source associated with at least one category of interest related to at least one location of the computing device.
16. The computer-implemented method of claim 9 , further comprising:
generating, by the device, the gross revenue relevance model based at least in part on at least one contextual data source associated with at least one historical promotion category interest level related to the user-promotion interaction data.
17. A computer program product, stored on a computer readable medium, comprising instructions that when executed by one or more computers cause the one or more computers to:
receive, from a computing device associated with a consumer identifier, input data representing a request for available promotions;
receive user-promotion interaction data describing attributes of historical interactions with respect to at least one promotion of a set of available promotions;
in circumstances where an activation state of the computing device satisfies a defined criterion, generate a contextual relevance score for each available promotion of the set of available promotions based at least in part on a gross revenue relevance model and features representing one or more of the input data or the user-promotion interaction data;
determine a subset of available promotions from the set of available promotions, wherein the subset of available promotions comprises available promotions with highest contextual relevance scores among contextual relevance scores calculated for the set of available promotions; and
transmit the subset of available promotions to the computing device.
18. The computer program product of claim 17 , further comprising instructions that when executed by the one or more computers cause the one or more computers to:
generate a promotion freshness feature vector representing the features; and
map the promotion freshness feature vector, based at least in part on the gross revenue relevance model, to a probability class of a set of probability classes.
19. The computer program product of claim 18 , further comprising instructions that when executed by the one or more computers cause the one or more computers to:
calculate the contextual relevance score based at least in part on the probability class.
20. The computer program product of claim 17 , wherein the gross revenue relevance model is a random forest model that employs the features representing one or more of the input data or the user-promotion interaction data.