BIG DATA BASED CROSS-DOMAIN RECOMMENDATION METHOD AND APPARATUS

• US 20190050484A1
• Filed: 06/20/2016
• Published: 02/14/2019
• Est. Priority Date: 12/23/2015
• Status: Active Grant

• Current Assignee: Baidu Online Network Technology (Beijing) Co., Ltd (Baidu Incorporated)
• Original Assignee: Baidu Online Network Technology (Beijing) Co., Ltd (Baidu Incorporated)
• Inventors: Liu, Zhiqiang, Shen, Zhiyong
• Granted Patent: US 10,459,996 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/313 Selection or weighting of terms for indexing
• CPC Class Codes: G06F 16/3346 using probabilistic model
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06Q 10/063 Operations research, analysis or management
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0282 Rating or review of business operators or products

Abstract

The present disclosure discloses a big data based cross-domain recommendation method and apparatus. Correlation between domains and correlation between user behavior characteristics are obtained by connection and cross analysis of a user'"'"'s online input and offline behaviors in different domains, and content is recommended to the user according to the established correlation. The technical solution of the present disclosure is applied to the precise consumer brand recommendation to users and the precise positioning of potential consumer brand customers in the Internet+retail area. The technical solution can solve a series of problems, such as cross-drainage of users in multiple domains, precise marketing and precise positioning of potential customers, and the effect is very obvious. From the offline simulation test and online real consumption test, the brand recommendation and user positioning accuracy is largely improved, while the GMV of the offline retail is largely improved.

Claims

1. A big data based cross-domain recommendation method, comprising:

modeling a topic separately based on an online input record and an offline behavior record of a user in a specific user set, the user in the specific user set having both the online input record and the offline behavior record;

determining a transition probability of transitioning from each online input topic to each offline behavior topic according to a topic modeling result; and

recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user.

2. The big data based cross-domain recommendation method according to claim 1, wherein the offline behavior record includes an offline consumption record;

and

the online input record includes a search record obtained by a search engine and/or an input record in a particular application obtained by an input method or a voice receiving device; and

a type of the particular application includes at least one of the following;

chat software, a search engine, social software and online shopping client software.

3. The big data based cross-domain recommendation method according to claim 2, wherein the offline consumption record includes an offline consumption topic and an offline consumption brand;

in case the online input record is a search record obtained by the search engine, the modeling a topic separately based on an online input record and an offline behavior record of the user in the specific user set comprises;

modeling a topic based on the offline consumption record of the user in the specific user set to obtain a probability P(brand consumption topic) of consuming each brand in each consumption topic and a probability P(consumption topic|user) of consuming each consumption topic by each user; and

modeling a topic based on an online search record of the user in the specific user set to obtain a probability P(search word|search topic) of inputting each search word in each search topic and a probability P(search topic|user) of searching each search topic by each user.

4. The big data based cross-domain recommendation method according to claim 3, wherein the determining a transition probability of transitioning from each online input topic to each offline behavior topic according to a topic modeling result comprises:

using a two-level probability graph model to obtain a transition probability P(consumption topic|search topic) from each search topic to each consumption topic, based on the P(search topic|user) corresponding to each user and the brand consumption data of the user in the specific user set, or based on the P(brand|consumption topic), the P(search topic|user) corresponding to the each user and the brand consumption data of the user in the specific user set.

5. The big data based cross-domain recommendation method according to claim 4, wherein the recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user comprises:

A1;

determining a probability P₀(brand|user) of consuming each brand by the target user in accordance with a following formula;

6. The big data based cross-domain recommendation method according to claim 4, wherein in case the target user has an offline consumption record, the recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user comprises:

B1;

determining a probability P₀(brand|user) of consuming each brand by the target user in accordance with a following formula;

7. A big data based cross-domain recommendation apparatus, comprising:

at least one processor; and

a memory storing instructions, which when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising;

modeling a topic separately based on an online input record and an offline behavior record of a user in a specific user set, the user in the specific user set having both the online input record and the offline behavior record;

determining a transition probability of transitioning from each online input topic to each offline behavior topic according to a topic modeling result; and

recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user.

8. The big data based cross-domain recommendation apparatus according to claim 7, wherein the offline behavior record includes an offline consumption record;

the online input record includes a search record obtained by a search engine, and/or an input record in a particular application obtained by an input method or a voice receiving device; and

a type of the particular application includes at least one of the following;

chat software, a search engine, social software and online shopping client software.

9. The big data based cross-domain recommendation apparatus according to claim 8, wherein the offline consumption record includes an offline consumption topic and an offline consumption brand;

in case the online input record is a search record obtained by the search engine, the modeling a topic separately based on an online input record and an offline behavior record of the user in the specific user set comprises;

modeling a topic based on the offline consumption record of the user in the specific user set to obtain a probability P(brand|consumption topic) of consuming each brand in each consumption topic, and a probability P(consumption topic|user) of consuming each consumption topic by each user; and

modeling a topic based on an online search record of the user in the specific user set to obtain a probability P(search word|search topic) of inputting each search word in each search topic and a probability P(search topic|user) of searching each search topic by each user.

10. The big data based cross-domain recommendation apparatus according to claim 9, wherein the determining a transition probability of transitioning from each online input topic to each offline behavior topic according to a topic modeling result comprises:

using a two-level probability graph model to obtain a transition probability P(consumption topic|search topic) from each search topic to each consumption topic, based on the P(search topic|user) corresponding to each user and the brand consumption data of the user in the specific user set, or based on the P(brand|consumption topic), the P(search topic|user) corresponding to the each user and the brand consumption data of the user in the specific user set.

11. The big data based cross-domain recommendation apparatus according to claim 10, wherein the recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user comprises:

determining a probability P₀(brand|user) of consuming each brand by the target user in accordance with a following formula;

12. The big data based cross-domain recommendation apparatus according to claim 10, wherein in case the target user has an offline consumption record, the recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user comprises:

B1;

determining a probability P₀(brand|user) of consuming each brand by the target user in accordance with a following formula;

13. (canceled)

14. A non-volatile computer storage medium storing a computer-readable instruction capable of being executed by a processor, wherein when the computer-readable instruction is executed, the processor executes a big data based cross-domain recommendation method, the method comprising:

modeling a topic separately based on an online input record and an offline behavior record of a user in a specific user set, the user in the specific user set having both the online input record and the offline behavior record;

determining a transition probability of transitioning from each online input topic to each offline behavior topic according to a topic modeling result; and

recommending content of an offline behavior to a target user based on the transition probability and an online input record of the target user.