System and method for using deep learning to identify purchase stages from a microblog post

1. A method of targeting information, the method comprising:

collecting one or more digital posts that identify a product and are associated with a digital account;

identifying, by a neural network, a stage of a decision model associated with each of the one or more digital posts based on a feature representation generated for each digital post; and

transmitting product-related information to the digital account, wherein the transmitting is based on the stage of the decision model identified for each of the one or more digital posts,

wherein the identifying the stage of the decision model comprises:

generating, by another neural network, a feature representation associated with each of the one or more digital posts based on at least one content feature associated with each of the one or more digital posts, the other neural network being different from the neural network; and

identifying the stage of the decision model based on the generated feature representation associated with each of the one or more digital post.

2. The method of claim 1 , further comprising pre-processing the one or more digital posts prior to identifying the stage of the decision model associated with the one or more digital posts, wherein the pre-processing comprises at least one of:

filtering the one or more digital posts to remove at least one digital post indicative of spam; and

cleaning at least one digital posts to remove inconsistent tokenization and irregular elongation of media content.

3. The method of claim 1 , wherein the identifying the stage of the decision model comprises:

extracting the at least one content feature associated with each of the one or more digital posts.

4. The method of claim 1 , wherein the identifying the stage of the decision model further comprises at least one of:

generating a sentiment representation associated with each of the one or more digital posts based on the at least one content feature associated with each of the one or more digital posts and identifying the stage of the decision model based on the generated sentiment representation and the generated feature representation associated with each of the one or more digital posts; and

generating a relevance representation associated with each of the one or more digital posts based on the at least one content feature associated with each of the one or more digital posts and identifying the stage of the decision model based on the generated relevance representation and the generated feature representation associated with each of the one or more digital posts.

5. The method of claim 1 , wherein the neural network is a unidirectional gated recurrent neural network;

wherein the one or more digital posts comprises a chronological sequence of digital posts comprising:

a first post associated with a first posting time; and

a second post associated with a second posting time after the first posting time; and

wherein, the identifying the stage of the decision model comprises:

identifying the stage of the decision model stage associated with the second post based on the feature representation associated with the first post and the feature representation associated with the second post.

6. The method of claim 1 , wherein the neural network is a bidirectional gated recurrent neural network;

wherein the one or more digital posts comprises a chronological sequence of digital posts comprising:

a first post associated with a first posting time;

a second post associated with a second posting time after the first posting time; and

a third post associated with a third posting time after the first positing time and after the second posting time; and

wherein, the identifying the stage of the decision model comprises

identifying the stage of the decision model associated with the second post based on the feature representation associated with the first post, the feature representation associated with the second post, and the feature representation associated with the third post.

7. The method of claim 1 , wherein the stages of decision model comprise one or more of:

an awareness stage;

an interested stage;

a desire stage;

an action stage;

a negative sentiment stage; and

a not relevant stage.

8. The method of claim 1 , the identifying, by the neural network, a stage of a decision model associated with each of the one or more digital posts comprises: using ranking of classes as a loss function to allow the neural network to compensate for class imbalances.

9. A non-transitory computer readable medium having stored therein a program for making a computer execute a method of targeting information, the method comprising:

collecting one or more digital posts that identify a product and are associated with a digital account;

identifying, by a neural network, a stage of a decision model associated with each of the one or more digital posts based on a feature representation generated for each digital post; and

transmitting product-related information to the digital account, wherein the transmitting is based on the stage of the decision model identified for each of the one or more digital posts,

wherein the identifying the stage of the decision model comprises:

generating, by another neural network, a feature representation associated with each of the one or more digital posts based on at least one content feature associated with each of the one or more digital posts, the other neural network being different from the neural network; and

identifying the stage of the decision model based on the generated feature representation associated with each of the one or more digital post.

10. The non-transitory computer readable medium of claim 9 , further comprising pre-processing the one or more digital posts prior to identifying the stage of the decision model associated with the one or more digital posts, wherein the pre-processing comprises at least one of:

filtering the one or more digital posts to remove at least one digital post indicative of spam; and

cleaning at least one digital posts to remove inconsistent tokenization and irregular elongation of media content.

11. The non-transitory computer readable medium of claim 9 , wherein the identifying the stage of the decision model comprises:

extracting the at least one content feature associated with each of the one or more digital posts.

12. The non-transitory computer readable medium of claim 9 , wherein the identifying the stage of the decision model further comprises at least one of:

generating a sentiment representation associated with each of the one or more digital posts based on the at least one content feature associated with each of the one or more digital posts and identifying the stage of the decision model based on the generated sentiment representation and the generated feature representation associated with each of the one or more digital posts; and

generating a relevance representation associated with each of the one or more digital posts based on the at least one content feature associated with each of the one or more digital posts and identifying the stage of the decision model based on the generated relevance representation and the generated feature representation associated with each of the one or more digital posts.

13. The non-transitory computer readable medium of claim 9 , wherein the neural network is a unidirectional gated recurrent neural network;

wherein the one or more digital posts comprises a chronological sequence of digital posts comprising:

a first post associated with a first posting time; and

a second post associated with a second posting time after the first posting time; and

wherein, the identifying the stage of the decision model comprises:

identifying the stage of the decision model stage associated with the second post based on the feature representation associated with the first post and the feature representation associated with the second post.

14. The non-transitory computer readable medium of claim 9 , wherein the neural network is a bidirectional gated recurrent neural network;

wherein the one or more digital posts comprises a chronological sequence of digital posts comprising:

a first post associated with a first posting time;

a second post associated with a second posting time after the first posting time; and

a third post associated with a third posting time after the first positing time and after the second posting time; and

wherein, the identifying the stage of the decision model comprises:

identifying the stage of the decision model associated with the second post based on the feature representation associated with the first post, the feature representation associated with the second post, and the feature representation associated with the third post.

15. A server apparatus configured to target information, the server apparatus comprising:

a memory storing one or more digital posts that identify a product and are associated with a digital account;

a processor executing a process comprising:

identifying, by a neural network, a stage of a decision model associated with each of the one or more digital posts based on a feature representation generated for each digital post; and

transmitting product-related information to the digital account, wherein the transmitting is based on the stage of the decision model identified for each of the one or more digital post,

wherein the identifying the stage of the decision model comprises:

generating, by another neural network, a feature representation associated with each of the one or more digital posts based on at least one content feature associated with each of the one or more digital posts, the other neural network being different from the neural network; and

identifying the stage of the decision model based on the generated feature representation associated with each of the one or more digital post.

16. The server apparatus of claim 15 , wherein the process further comprises pre-processing the one or more digital posts prior to identifying the stage of the decision model associated with the one or more digital posts, wherein the pre-processing comprises at least one of:

filtering the one or more digital posts to remove at least one digital post indicative of spam; and

cleaning at least one digital posts to remove inconsistent tokenization and irregular elongation of media content.

17. The server apparatus of claim 15 , wherein the identifying the stage of the decision model comprises:

extracting the at least one content feature associated with each of the one or more digital posts.

18. The server apparatus of claim 16 , wherein the identifying the stage of the decision model further comprises at least one of:

generating a sentiment representation associated with each of the one or more digital posts based on the at least one content feature associated with each of the one or more digital posts and identifying the stage of the decision model based on the generated sentiment representation and the generated feature representation associated with each of the one or more digital posts; and

generating a relevance representation associated with each of the one or more digital posts based on the at least one content feature associated with each of the one or more digital posts and identifying the stage of the decision model based on the generated relevance representation and the generated feature representation associated with each of the one or more digital posts.

19. The server apparatus of claim 16 , wherein the neural network is a unidirectional gated recurrent neural network;

wherein the one or more digital posts comprises a chronological sequence of digital posts comprising:

a first post associated with a first posting time; and

a second post associated with a second posting time after the first posting time; and

wherein, the identifying the stage of the decision model comprises:

identifying the stage of the decision model stage associated with the second post based on the feature representation associated with the first post and the feature representation associated with the second post.

20. The server apparatus of claim 16 , wherein the neural network is a bidirectional gated recurrent neural network;

wherein the one or more digital posts comprises a chronological sequence of digital posts comprising:

a first post associated with a first posting time;

a second post associated with a second posting time after the first posting time; and

a third post associated with a third posting time after the first positing time and after the second posting time; and

wherein, the identifying the stage of the decision model comprises:

identifying the stage of the decision model associated with the second post based on the feature representation associated with the first post, the feature representation associated with the second post, and the feature representation associated with the third post.