Cross-entity categorization of heterogenous data
The techniques described herein relate to constructing and using seed audiences. In an embodiment, a method includes loading, by a processing device, a user event sequence, the user event sequence including a plurality of user events and a plurality of corresponding conversions; generating, by the processing device, a plurality of conversion neighborhoods based on the user event sequence, a given conversion neighborhood in the plurality of conversion neighborhood including at least one conversion rule and a set of user events from the plurality of user events; annotating, by the processing device, each conversion neighborhood in the plurality of conversion neighborhoods with categorical labels; and generating, by the processing device, seed audiences for each conversion neighborhood, a given seed audience including a ranked list of user events for each conversion rule associated with the conversion neighborhood.
1. A method comprising:
loading, by a processing device, data representing a user event sequence, the user event sequence comprising a plurality of user events and a plurality of corresponding conversions;
generating, by the processing device, a plurality of conversion neighborhoods based on the user event sequence, a given conversion neighborhood in the plurality of conversion neighborhood comprising at least one conversion rule and a set of user events from the plurality of user events;
annotating, by the processing device, each conversion neighborhood in the plurality of conversion neighborhoods with categorical labels; and
generating, by the processing device, seed audiences for each conversion neighborhood, a given seed audience comprising a ranked list of user events for each conversion rule associated with the conversion neighborhood.
2. The method of claim 1 , wherein generating a plurality of conversion neighborhoods comprises computing embeddings for the plurality of user events and the plurality of corresponding conversions and computing similarities among the plurality of user events and the plurality of corresponding conversions.
3. The method of claim 2 , wherein computing similarities comprises using an approximate neighbor search algorithm to compute the similarities.
4. The method of claim 1 , wherein annotating a respective conversion neighborhood in the plurality of conversion neighborhoods with categorical labels comprises classifying each user event in the respective conversion neighborhood based on a pre-built taxonomy of category labels.
5. The method of claim 4 , wherein classifying each user event in the respective conversion neighborhood comprises inputting each user event into a random forest model.
6. The method of claim 1 , further comprising generating the ranked list of user events for each conversion rule by computing similarities between embeddings representing the user events and embeddings representing the conversion rules.
7. The method of claim 1 , further comprising:
expanding, by the processing device, a seed audience in the seed audiences into a set of users;
computing, by the processing device, conversion probabilities for the set of users;
segmenting, by the processing device, the set of users into segments based on the conversion probabilities; and
returning, by the processing device, a subset of the segments to a user.
8. A device comprising:
a processor; and
a storage medium for tangibly storing thereon program logic for execution by the processor, the program logic comprising:
logic, executed by the processor, for loading data representing a user event sequence, the user event sequence comprising a plurality of user events and a plurality of corresponding conversions;
logic, executed by the processor, for generating a plurality of conversion neighborhoods based on the user event sequence, a given conversion neighborhood in the plurality of conversion neighborhood comprising at least one conversion rule and a set of user events from the plurality of user events;
logic, executed by the processor, for annotating each conversion neighborhood in the plurality of conversion neighborhoods with categorical labels; and
logic, executed by the processor, for generating seed audiences for each conversion neighborhood, a given seed audience comprising a ranked list of user events for each conversion rule associated with the conversion neighborhood.
9. The device of claim 8 , wherein generating a plurality of conversion neighborhoods comprises computing embeddings for the plurality of user events and the plurality of corresponding conversions and computing similarities among the plurality of user events and the plurality of corresponding conversions.
10. The device of claim 9 , wherein computing similarities comprises using an approximate neighbor search algorithm to compute the similarities.
11. The device of claim 8 , wherein annotating a respective conversion neighborhood in the plurality of conversion neighborhoods with categorical labels comprises classifying each user event in the respective conversion neighborhood based on a pre-built taxonomy of category labels.
12. The device of claim 11 , wherein classifying each user event in the respective conversion neighborhood comprises inputting each user event into a random forest model.
13. The device of claim 11 , the program logic further comprising logic, executed by the processor, for generating the ranked list of user events for each conversion rule by computing similarities between embeddings representing the user events and embeddings representing the conversion rules.
14. The device of claim 11 , the program logic further comprising:
logic, executed by the processor, for expanding a seed audience in the seed audiences into a set of users;
logic, executed by the processor, for computing conversion probabilities for the set of users;
logic, executed by the processor, for segmenting the set of users into segments based on the conversion probabilities; and
logic, executed by the processor, for returning a subset of the segments to a user.
15. A non-transitory computer-readable storage medium for tangibly storing computer program instructions capable of being executed by a computer processor, the computer program instructions defining steps of:
loading data representing a user event sequence, the user event sequence comprising a plurality of user events and a plurality of corresponding conversions;
generating a plurality of conversion neighborhoods based on the user event sequence, a given conversion neighborhood in the plurality of conversion neighborhood comprising at least one conversion rule and a set of user events from the plurality of user events;
annotating each conversion neighborhood in the plurality of conversion neighborhoods with categorical labels; and
generating seed audiences for each conversion neighborhood, a given seed audience comprising a ranked list of user events for each conversion rule associated with the conversion neighborhood.
16. The non-transitory computer-readable storage medium of claim 15 , wherein generating a plurality of conversion neighborhoods comprises computing embeddings for the plurality of user events and the plurality of corresponding conversions and computing similarities among the plurality of user events and the plurality of corresponding conversions.
17. The non-transitory computer-readable storage medium of claim 16 , wherein computing similarities comprises using an approximate neighbor search algorithm to compute the similarities.
18. The non-transitory computer-readable storage medium of claim 15 , wherein annotating a respective conversion neighborhood in the plurality of conversion neighborhoods with categorical labels comprises classifying each user event in the respective conversion neighborhood based on a pre-built taxonomy of category labels.
19. The non-transitory computer-readable storage medium of claim 18 , wherein classifying each user event in the respective conversion neighborhood comprises inputting each user event into a random forest model.
20. The non-transitory computer-readable storage medium of claim 15 , further comprising generating the ranked list of user events for each conversion rule by computing similarities between embeddings representing the user events and embeddings representing the conversion rules.