Incremental model training for advertisement targeting using streaming data
Incremental model training for advertisement targeting is performed using streaming data. A model for targeting advertisements of an advertising campaign is initialized. A data stream including data corresponding to converters and data corresponding to non-converters is received. The model is then applied to the data corresponding to the converter and data corresponding to the non-converter (or other ratio of converter to non-converters) to obtain a predicted score for each. The predicted score is compared to the observed score (e.g., an observed score of 1 for a converter, and 0 for a non-converter). The difference between the predicted and observed scores is computed, and the model is incrementally updated based on this difference. Models can optionally be built separately on multiple modeling servers that are geographically dispersed in order to support bidding on advertising opportunities in a real-time bidding environment.
1. A method comprising:
responsive to receiving a signal of conversion corresponding to a converter at a first modeling server of a distributed modeling system comprising a plurality of modeling servers, pairing the converter and a non-converter and performing for the converter and the non-converter:
applying a model at the first modeling server to data corresponding to the converter to obtain a predicted score for the converter;
applying the model to data corresponding to the non-converter to obtain a predicted score for the non-converter;
comparing the predicted score for the converter to an observed score for the converter to compute a first difference;
comparing the predicted score for the non-converter to an observed score for the non-converter to compute a second difference; and
updating the model at the first modeling server based on the first and second differences when the first difference exceeds a threshold;
combining the updated model with at least one other model from at least a second modeling server of the distributed modeling system; and
redistributing the combined model to at least the first and second modeling servers of the distributed modeling system.
2. The method of claim 1 , wherein at least the applying, comparing, and updating steps are performed by a plurality of modeling servers concurrently on data corresponding to different converters.
3. The method of claim 1 , wherein combining the updated model with at least one other model from at least a second modeling server comprises taking a weighted average of the models based on respective amounts of converters recently incorporated into each model.
4. The method of claim 1 , further comprising downsampling data corresponding to non-converters.
5. The method of claim 1 , wherein the model comprises a feature set and a respective weight for each feature in the feature set.
6. The method of claim 5 , further comprising penalizing infrequently observed features to reduce the feature set in the model.
7. The method of claim 1 , further comprising discarding the data corresponding to the converter and the data corresponding to the non-converter.
8. A non-transitory computer readable storage medium executing computer program instructions, the computer program instructions comprising instructions for:
responsive to receiving a signal of conversion corresponding to a single converter at a first modeling server of a distributed modeling system comprising a plurality of modeling servers, pairing the converter and a non-converter and performing for the converter and the non-converter:
applying a model at the first modeling server to data corresponding to the converter to obtain a predicted score for the converter;
applying the model to data corresponding to the non-converter to obtain a predicted score for the non-converter;
comparing the predicted score for the converter to an observed score for the converter to compute a first difference;
comparing the predicted score for the non-converter to an observed score for the non-converter to compute a second difference;
and updating the model at the first modeling server based on the first and second differences when the first difference exceeds a threshold;
combining the updated model with at least one other model from at least a second modeling server of the distributed modeling system; and
redistributing the combined model to at least the first and second modeling servers of the distributed modeling system.
9. The medium of claim 8 , wherein combining the updated model with at least one other model from at least a second modeling server comprises taking a weighted average of the models based on respective amounts of converters recently incorporated into each model.
10. The medium of claim 8 , further comprising instructions for downsampling data corresponding to non-converters.
11. The medium of claim 8 , wherein the model comprises a feature set and a respective weight for each feature in the feature set.
12. The medium of claim 11 , further comprising instructions for penalizing infrequently observed features to reduce the feature set in the model.
13. The medium of claim 8 , further comprising instructions for discarding the data corresponding to the converter and the data corresponding to the non-converter.
14. A system comprising:
a processor;
a computer readable storage medium storing processor-executable computer program instructions, the instructions comprising instructions for:
responsive to receiving a signal of conversion corresponding to a converter at a first modeling server of a distributed modeling system comprising a plurality of modeling servers, pairing the converter and a non-converter and performing for the converter and the non-converter:
applying a model at the first modeling server to data corresponding to the converter to obtain a predicted score for the converter;
applying the model to data corresponding to the non-converter to obtain a predicted score for the non-converter;
comparing the predicted score for the converter to an observed score for the converter to compute a first difference;
comparing the predicted score for the non-converter to an observed score for the non-converter to compute a second difference; and
updating the model at the first modeling server based on the first and second differences when the first difference exceeds a threshold;
combining the updated model with at least one other model from at least a second modeling server of the distributed modeling system; and
redistributing the combined model to at least the first and second modeling servers of the distributed modeling system.
15. The system of claim 14 , wherein combining the updated model with at least one other model from at least a second modeling server comprises taking a weighted average of the models based on respective amounts of converters recently incorporated into each model.
16. The system of claim 14 , the medium further comprising instructions for downsampling data corresponding to non-converters.
17. The system of claim 14 , wherein the model comprises a feature set and a respective weight for each feature in the feature set.
18. The system of claim 17 , the medium further comprising instructions for penalizing infrequently observed features to reduce the feature set in the model.
19. The system of claim 14 , the medium further comprising instructions for discarding the data corresponding to the converter and the data corresponding to the non-converter.