Characterizing an entity in an identifier space based on behaviors of unrelated entities in a different identifier space
Models are built based on existing histories in one identifier space to infer features of entities in a different identifier space. A source model is built using features of an archetypical population in a given identifier space and the standard population. A join panel, i.e., a set of entities operating across both the given identifier space and a second disjoined identifier space, is scored using the source model. Based on the scores and features associated with the entities in the join panel within the second identifier space, a target model specific to the second identifier space is built. An audience of entities within the second identifier space can then be scored using the target model to identify entities that are similar to the archetypical population.
1. A computer-implemented method of determining a similarity between entities across different identifier spaces, the method comprising:
building a first model specific to a first identifier space using a first set of features correlated with an archetypical population having made a product purchase in the first identifier space as opposed to a standard population, both the archetypical population and the standard population operating in the first identifier space, the first set of features associated with the archetypical population in the first identifier space;
identifying a join panel of entities that each operates in both the first identifier space and a second identifier space, each entity of the join panel having a respective first identifier of an Internet browser of the first identifier space mapped to a respective second identifier of a mobile application of the second identifier space;
applying the first model to each entity of the join panel to compute a score for each respective entity of the join panel, each respective score reflective of the similarity between the respective entity of the join panel and the archetypical population;
selecting a set of contributing entities comprising a plurality of entities from the join panel, each of the contributing entities having a respective score above a threshold;
building a second model specific to the mobile application of the second identifier space by selecting a second set of features correlated with the set of contributing entities as opposed to a second standard population, both the contributing entities and the second standard population operating in the second identifier space, the second set of features associated with the contributing entities in the second identifier space;
predicting the similarity between a target entity operating in the second identifier space and the archetypical population operating in the first identifier space by applying the second model to the target entity operating in the second identifier space, wherein an identifier associated with the target entity in the second identifier space is not mapped to an identifier in the first identifier space;
responsive to the predicted similarity indicating the target entity is likely to be similar to the archetypical population, targeting the target entity to receive advertising content related to the product; and
sending the advertising content to the mobile application of the target entity.
2. The method of claim 1 , wherein:
each entity operating in the first identifier space is associated with an identifier specific to the first identifier space, and each entity operating in the second identifier space is associated with an identifier specific to the second identifier space.
3. The method of claim 1 wherein:
each entity in the join panel is associated with a respective identifier specific to the first identifier space and a respective identifier specific to the second identifier space, the respective identifier specific to the first identifier space mapped to the respective identifier specific to the second identifier space.
4. The method of claim 1 wherein:
the target entity operating in the second identifier space also operates in the first identifier space, and an identifier associated with the target entity and specific to the first identifier space is not mapped to an identifier associated with the target entity and specific to the second identifier space.
5. The method of claim 1 further comprising:
selecting the archetypical population in the first identifier space according to pre-defined criteria, wherein each entity in the archetypical population fulfills the pre-defined criteria.
6. The method of claim 1 further comprising:
selecting the archetypical population by analyzing histories of entities operating in the first identifier space.
7. The method of claim 1 wherein:
the second set of features are found in the histories of the contributing entities associated with the second identifier space.
8. The method of claim 1 wherein:
building the second model comprises weighting each feature in the second set of features according to the scores of the contributing entities having that feature.
9. A non-transitory computer readable storage medium executing computer program instructions for determining a similarity between entities across different identifier spaces, the computer program instructions comprising instructions for:
building a first model specific to a first identifier space using a first set of features correlated with an archetypical population having made a product purchase in the first identifier space as opposed to a standard population, both the archetypical population and the standard population operating in the first identifier space, the first set of features associated with the archetypical population in the first identifier space;
identifying a join panel of entities that each operates in both the first identifier space and a second identifier space, each entity of the join panel having a respective first identifier of an Internet browser of the first identifier space mapped to a respective second identifier of a mobile application of the second identifier space;
applying the first model to each entity of the join panel to compute a score for each respective entity of the join panel, each respective score reflective of the similarity between the respective entity of the join panel and the archetypical population;
selecting a set of contributing entities comprising a plurality of entities from the join panel, each of the contributing entities having a respective score above a threshold;
building a second model specific to the mobile application of the second identifier space by selecting a second set of features correlated with the set of contributing entities as opposed to a second standard population, both the contributing entities and the second standard population operating in the second identifier space, the second set of features associated with the contributing entities in the second identifier space;
predicting the similarity between a target entity operating in the second identifier space and the archetypical population operating in the first identifier space by applying the second model to the target entity operating in the second identifier space, wherein an identifier associated with the target entity in the second identifier space is not mapped to an identifier in the first identifier space;
responsive to the predicted similarity indicating the target entity is likely to be similar to the archetypical population, targeting the target entity to receive advertising content related to the product; and
sending the advertising content to the mobile application of the target entity.
10. The medium of claim 9 wherein:
each entity operating in the first identifier space is associated with an identifier specific to the first identifier space, and each entity operating in the second identifier space is associated with an identifier specific to the second identifier space.
11. The medium of claim 9 wherein:
each entity in the join panel is associated with a respective identifier specific to the first identifier space and a respective identifier specific to the second identifier space, the respective identifier specific to the first identifier space mapped to the respective identifier specific to the second identifier space.
12. The medium of claim 9 wherein:
the target entity operating in the second identifier space also operates in the first identifier space, and an identifier associated with the target entity and specific to the first identifier space is not mapped to an identifier associated with the target entity and specific to the second identifier space.
13. The medium of claim 9 wherein the instructions further comprise:
selecting the archetypical population in the first identifier space according to pre-defined criteria, wherein each entity in the archetypical population fulfills the pre-defined criteria.
14. The medium of claim 9 wherein the instructions further comprise:
selecting the archetypical population by analyzing histories of entities operating in the first identifier space.
15. The medium of claim 9 wherein the instructions further comprise:
building the second model by weighting each feature in the second set of features according to the scores of the contributing entities having that feature.
16. A system comprising:
a processor;
a computer readable storage medium storing processor-executable computer program instructions for predicting the similarity between entities across different identifier spaces, the computer program instructions comprising instructions for:
building a first model specific to a first identifier space using a first set of features correlated with an archetypical population having made a product purchase in the first identifier space as opposed to a standard population, both the archetypical population and the standard population operating in the first identifier space, the first set of features associated with the archetypical population in the first identifier space;
identifying a join panel of entities that each operates in both the first identifier space and a second identifier space, each entity of the join panel having a respective first identifier of an Internet browser of the first identifier space mapped to a respective second identifier of a mobile application of the second identifier space;
applying the first model to each entity of the join panel to compute a score for each respective entity of the join panel, each respective score reflective of the similarity between the respective entity of the join panel and the archetypical population;
selecting a set of contributing entities comprising a plurality of entities from the join panel, each of the contributing entities having a respective score above a threshold;
building a second model specific to the mobile application of the second identifier space by selecting a second set of features correlated with the set of contributing entities as opposed to a second standard population, both the contributing entities and the second standard population operating in the second identifier space, the second set of features associated with the contributing entities in the second identifier space;
predicting the similarity between a target entity operating in the second identifier space and the archetypical population operating in the first identifier space by applying the second model to the target entity operating in the second identifier space, wherein an identifier associated with the target entity in the second identifier space is not mapped to an identifier in the first identifier space;
responsive to the predicted similarity indicating the target entity is likely to be similar to the archetypical population, targeting the target entity to receive advertising content related to the product; and
sending the advertising content to the mobile application of the target entity.