Behavior prediction on social media using neural networks

Patent metadata

Field	Value
Publication number	US-10453099-B2
Application number	US-201514966438-A
Country	US
Kind code	B2
Filing date	Dec 11, 2015
Priority date	Dec 11, 2015
Publication date	Oct 22, 2019
Grant date	Oct 22, 2019

Abstract

Example implementations include a system and method of recognizing behavior of a user. In example implementations, a first post and at least one subsequent post indicative of a product and associated with a first social media account is obtained. A relevance probability is calculated for each of the obtained first post and the at least one subsequent post. The obtained first post and the at least one subsequent post are sequentially analyzed by a second neural network to determine output values relevant to probability of purchasing the product. A probability of purchasing the product is calculated based on the determined output values associated with each post and the calculated relevance probabilities. Product-related information is transmitted to the user associated with the obtained first post based on the determined probability of purchasing the product.

First claim

What is claimed is:

1. A method of recognizing a behavior of a specific user using a predictive engine having multiple processing layers, the method comprising:
receiving, by an input layer of the predictive engine, a plurality of sequential posts authored by the specific user indicative of a product and associated with a first social media account;
sequentially calculating, using a bottom layer of the predictive engine having a first neural network, a relevance probability for each of the obtained posts in the plurality of sequential posts by:
parsing, by the first neural network, each of the plurality of sequential posts into one or more digital content segments;
assigning, by the first neural network, a vector representation to each of the one or more content segments;
calculating, by the first neural network, a total vector representation for each of the plurality of sequential posts based on the vector representations assigned to each of the one or more content segments;
determining, using a hidden layer of the predicative engine having a second neural network, an output value relevant to probability of purchasing associated with each posts in the plurality of sequential posts by sequentially analyzing each post, wherein for each post in the plurality of sequential posts, each post is compared to a previously stored maximum output value, and the greater of the post and the previous stored maximum output value is newly stored as the maximum output value;
calculating a probability of purchasing the product based on the determined output values associated with each posts in the plurality of sequential posts and the stored maximum after all of the plurality of sequential posts is compared; and
transmitting product-related information to the specific user associated with the posts in the plurality of sequential posts based on the determined probability of purchasing the product exceeding a specified threshold.

2. The method of claim 1 , wherein the transmitting the product-related information comprises: identifying a second social media account associated with the specific user; detecting information associated with the specific user based on the identified second social media account; and tailoring the information relating to the product based on the detected information.

3. The method of claim 1 , wherein the calculating the relevance probability for each of the posts in the plurality of sequential posts comprises using a feed forward neural network to calculate a relevance probability for each of the posts in the plurality of sequential posts, based on content data associated with each of the posts in the plurality of sequential posts individually.

4. The method of claim 1 , wherein the calculating the relevance probability for each of the posts in the plurality of sequential posts comprises classifying at least one of the obtained posts in the plurality of sequential posts as relevant to determining a probability of purchasing the product.

5. The method of claim 1 , wherein the determining an output value relevant to probability of purchasing comprises using the second neural network with memory to sequentially analyze each post in the plurality of sequential posts.

6. The method of claim 5 , wherein the sequentially analyzing each post in the plurality of sequential posts comprises: determining a first output value associated with an obtained first post in the plurality of sequential posts based on the content of the first post; determining a second output value associated with each of the one or more subsequent posts in the plurality of sequential posts based on the content of the one or more subsequent posts and the determined first output value of the obtained first post; and determining a maximum output value based on the determined first output value associated with the obtained first post and the determined second output value associated with each of the one or more subsequent posts in the plurality of sequential posts.

7. The method of claim 6 , wherein the calculating a probability of purchasing the product further comprises: calculating the probability of purchasing the product using a SOFTMAX function; and classifying the user as being likely to purchase the product based on the calculated probability of purchasing exceeding a confidence level threshold; wherein the confidence level threshold varies based on one or more of: a number of subsequent posts obtained, calculated relevance probabilities of each of the obtained first post and the one or more subsequent posts, a desired high purchaser strength level; a desired purchaser conversion rate, and a product type.

8. The method of claim 1 , further comprising classifying a user as a predicted purchaser based on the calculated probability of purchasing the product.

9. The method of claim 1 , wherein the first social media account is a microblog account.

10. The method of claim 1 , wherein the obtaining the plurality of the sequential posts authored by the specific user comprises collecting a plurality of subsequent posts for a duration of time.

11. The method of claim 10 , wherein the duration is 60 days.

12. A non-transitory computer readable medium having stored therein a program for making a computer execute a method of recognizing behavior of a specific user using a predictive engine having multiple processing layers, said program including computer executable instructions for performing the method comprising:
receiving, by an input layer of the predictive engine, a plurality of sequential posts authored by the specific user indicative of a product and associated with a first social media account;
sequentially calculating, using a bottom layer of the predictive engine having a first neural network, a relevance probability for each of the obtained posts in the plurality of sequential posts by:
parsing, by the first neural network, each of the plurality of sequential posts into one or more digital content segments;
assigning, by the first neural network, a vector representation to each of the one or more content segments;
calculating, by the first neural network, a total vector representation for each of the plurality of sequential posts based on the vector representations assigned to each of the one or more content segments;
determining, using a hidden layer of the predicative engine having a second neural network, an output value relevant to probability of purchasing associated with each posts in the plurality of sequential posts by sequentially analyzing each post, wherein for each post in the plurality of sequential posts, each post is compared to a previously stored maximum output value, and the greater of the post and the previous stored maximum output value is newly stored as the maximum output value;
calculating a probability of purchasing the product based on the determined output values associated with each posts in the plurality of sequential posts and the stored maximum after all of the plurality of sequential posts is compared; and
transmitting product-related information to the specific user associated with the posts in the plurality of sequential posts based on the determined probability of purchasing the product exceeding a specified threshold.

13. The non-transitory computer readable medium of claim 12 , wherein the transmitting information comprises: identifying a second social media account associated with the specific user; detecting information associated with the specific user based on the identified second social media account; and tailoring the information relating to the product based on the detected information.

14. The non-transitory computer

Assignees

• Fuji Xerox Co Ltd

Inventors

• Korpusik Mandy Barrett
• Chen Francine
• Sakaki Shigeyuki
• Chen Yin-Ying

Classifications

• G06Q10/40

  Business processes related to social networking or social networking services · CPC title

• G06Q30/0269 Primary

  based on user profile or attribute · CPC title

• G06Q50/01

  Physics · mapped topic