Recommending content information based on user behavior

• US 9,400,995 B2
• Filed: 08/15/2014
• Issued: 07/26/2016
• Est. Priority Date: 08/16/2011
• Status: Active Grant

• Current Assignee: Alibaba Group Holding Ltd.
• Original Assignee: Alibaba Group Holding Ltd.
• Inventors: Gu, Xiangyu, Fu, Qile, Zu, Zhonglin, Ning, Wei
• Primary Examiner(s): Richardson, James E
• Application Number:
  US14/461,104
• Publication Number:
  US 20150026001A1
• Time in Patent Office: 711 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/435 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06Q 30/0201 Market modelling; Market analysis; Collecting market data
• CPC Class Codes: G06Q 30/0282 Rating or review of business operators or products
• CPC Class Codes: G06Q 30/0631 Item recommendations

Abstract

Content recommendation includes: determining a plurality of user behavior categories pertaining to a plurality of user behaviors by a plurality of users within a period of time, and a plurality of user behavior frequencies corresponding to the plurality of user behavior categories; determining whether the number of user behavior categories exceeds a preset category threshold; in the event that the number of user behavior categories exceeds the preset threshold, excluding one or more user behavior categories such that the number of remaining user behavior categories does not exceed the preset threshold; determining, based at least on the remaining user behavior categories and corresponding user behavior frequencies, user preference data comprising frequency distribution information of a plurality of pieces of content information targeted by user behaviors in the remaining user behavior categories; and recommending content information to a user based on the user preference data.

Claims

1. A content recommendation system, comprising:

one or more computer processors to;

determine a plurality of user behavior categories pertaining to a plurality of user behaviors within a preset time period, and a plurality of user behavior frequencies corresponding to the plurality of user behavior categories;

determine, based at least on the user behavior categories and corresponding user behavior frequencies, user preference data comprising frequency distribution information of a plurality of pieces of content information targeted by user behaviors in the user behavior categories;

divide the preset time period into a plurality of time segments;

analyze content information targeted by user behaviors in each time segment to determine user preference data for each time segment;

determine whether a drifting user preference, a sudden user preference, or a combination thereof exists based on the analyzed content information;

in the event that the drifting user preference exists, remove the drifting user preference from the content information;

in the event that the sudden user preference exists, remove the sudden user preference from the content information;

recommend content information to a user based at least in part on the user preference data, wherein the user is a target user;

identify a plurality of similar users who are similar to the target user, comprising to;

determine whether a geographical region corresponding to a current Internet Protocol (IP) address of the target user matches a registered geographical region of the target user;

in the event that the geographical region corresponding to the current IP address of the target user matches the registered geographical region of the target user, select a reference user set among other users with the same registered geographical region as the registered geographical region of the target user;

in the event that the geographical region corresponding to the current IP address of the target user does not match the registered geographical region of the target user, select a reference user set among other users with the same registered geographical region as the geographical region corresponding to the IP address of the target user; and

identify the plurality of similar users by comparing the target user with users in the reference user set;

obtain user behavior categories and corresponding user behavior frequencies of the plurality of similar users;

determine, based on the user behavior categories and corresponding user behavior frequencies of the plurality of similar users, population preference data comprising frequency distribution of a plurality of pieces of content information targeted by user behaviors of the plurality of similar users; and

recommend content information to the target user based at least in part on the population preference data and

one or more memories coupled to the one or more processors, configured to provide the one or more processors with instructions.

2. The system of claim 1, wherein the one or more processors are further configured to:

determine whether a user behavior frequency of a user behavior category exceeds a frequency threshold; and

in the event that the user behavior frequency of the user behavior category exceeds the frequency threshold, replace the value of the user behavior frequency with the value of the frequency threshold.

3. The system of claim 1, wherein the one or more processors are further configured to:

apply time weights to user behavior data of the plurality of user behaviors; and

wherein

a higher time weight is applied to more recent user behavior data than to more distant user behavior data.

4. The system of claim 3, wherein the time weights are computed according to:

time weight=1/log(D_off);

wherein D_off is a time difference between a time when a user behavior occurred and an end of the preset time period.

5. The system of claim 1, wherein to determine whether the drifting user preference exists comprises to identify first content information that appears in an earlier time segment but not in a later time segment.

6. The system of claim 1, wherein to determine whether the sudden user preference exists comprises to identify second content information that is targeted with abnormally high frequency.

7. The system of claim 1, wherein:

to recommend content information to the user based at least in part on the user preference data includes to;

select a set of users to form a rule-setting users set;

analyze behaviors of users in the rule-setting users set to obtain user preference data of the rule-setting users;

perform training based on user preference data of the rule-setting users to obtain a rule set comprising one or more rules predicting content information corresponding to a type of user preference that is to be recommended given a pattern of past user preferences; and

apply the rule set to user preference data of the target user to obtain recommended content information.

8. The system of claim 7, wherein to apply the rule set to the user preference data of the target user to obtain recommended content information includes normalizing the user preference data of the target user with the user preference data of the rule-setting users.

9. A method of recommending content information to a user, comprising:

determining, using one or more computer processors, a plurality of user behavior categories pertaining to a plurality of user behaviors within a preset time period, and a plurality of user behavior frequencies corresponding to the plurality of user behavior categories;

determining, based at least on the user behavior categories and corresponding user behavior frequencies, user preference data comprising frequency distribution information of a plurality of pieces of content information targeted by user behaviors in the user behavior categories;

dividing the preset time period into a plurality of time segments;

analyzing content information targeted by user behaviors in each time segment to determine user preference data for each time segment;

determining whether a drifting user preference, a sudden user preference, or a combination thereof exists based on the analyzed content information;

in the event that the drifting user preference exists, removing the drifting user preference from the content information;

in the event that the sudden user preference exists, removing the sudden user preference from the content information;

recommending content information to the user based on the user preference data, wherein the user is a target user;

identifying a plurality of similar users who are similar to the target user, comprising;

determining whether a geographical region corresponding to a current Internet Protocol (IP) address of the target user matches a registered geographical region of the target user;

in the event that the geographical region corresponding to the current IP address of the target user matches the registered geographical region of the target user, selecting a reference user set among other users with the same registered geographical region as the registered geographical region of the target user;

in the event that the geographical region corresponding to the current IP address of the target user does not match the registered geographical region of the target user, selecting a reference user set among other users with the same registered geographical region as the geographical region corresponding to the IP address of the target user; and

identifying the plurality of similar users by comparing the target user with users in the reference user set;

obtaining user behavior categories and corresponding user behavior frequencies of the plurality of similar users;

determining, based on the user behavior categories and corresponding user behavior frequencies of the plurality of similar users, population preference data comprising frequency distribution of a plurality of pieces of content information targeted by user behaviors of the plurality of similar users; and

recommending content information to the target user based at least in part on the population preference data.

10. The method of claim 9, further comprising:

determining whether a user behavior frequency of a user behavior category exceeds a frequency threshold; and

in the event that the user behavior frequency of the user behavior category exceeds the frequency threshold, replacing the value of the user behavior frequency with the value of the frequency threshold.

11. The method of claim 9, further comprising:

applying time weights to user behavior data of the plurality of user behaviors; and

wherein

a higher time weight is applied to more recent user behavior data than to more distant user behavior data.

12. The method of claim 11, wherein the time weights are computed according to:

time weight=1/log (D_off), wherein D_off is a time difference between a time when a user behavior occurred and an end of the preset time period.

13. The method of claim 9, wherein the determining of whether the drifting user preference exists comprises identifying first content information that appears in an earlier time segment but not in a later time segment.

14. The method of claim 9, wherein the determining of whether the sudden user preference exists comprises identifying second content information that is targeted with abnormally high frequency.

15. The method of claim 9, wherein:

recommending content information to the user based at least in part on the user preference data includes;

selecting a set of users to form a rule-setting users set;

analyzing behaviors of users in the rule-setting users set to obtain user preference data of the rule-setting users;

performing training based on user preference data of the rule-setting users to obtain a rule set comprising one or more rules predicting content information corresponding to a type of user preference that is to be recommended given a pattern of past user preferences; and

applying the rule set to user preference data of the target user to obtain recommended content information.

16. The method of claim 15, wherein applying the rule set to the user preference data of the target user to obtain recommended content information includes normalizing the user preference data of the target user with the user preference data of the rule-setting users.

17. A computer program product for content recommendation, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

determining, using one or more computer processors, a plurality of user behavior categories pertaining to a plurality of user behaviors within a period of time, and a plurality of user behavior frequencies corresponding to the plurality of user behavior categories;

determining, based at least on the user behavior categories and corresponding user behavior frequencies, user preference data comprising frequency distribution information of a plurality of pieces of content information targeted by user behaviors in the user behavior categories;

dividing the preset time period into a plurality of time segments;

analyzing content information targeted by user behaviors in each time segment to determine user preference data for each time segment;

determining whether a drifting user preference, a sudden user preference, or a combination thereof exists based on the analyzed content information;

in the event that the drifting user preference exists, removing the drifting user preference from the content information;

in the event that the sudden user preference exists, removing the sudden user preference from the content information;

recommending content information to a user based on the user preference data, wherein the user is a target user;

identifying a plurality of similar users who are similar to the target user, comprising;

determining whether a geographical region corresponding to a current Internet Protocol (IP) address of the target user matches a registered geographical region of the target user;

in the event that the geographical region corresponding to the current IP address of the target user matches the registered geographical region of the target user, selecting a reference user set among other users with the same registered geographical region as the registered geographical region of the target user;

in the event that the geographical region corresponding to the current IP address of the target user does not match the registered geographical region of the target user, selecting a reference user set among other users with the same registered geographical region as the geographical region corresponding to the IP address of the target user; and

identifying the plurality of similar users by comparing the target user with users in the reference user set;

obtaining user behavior categories and corresponding user behavior frequencies of the plurality of similar users;

determining, based on the user behavior categories and corresponding user behavior frequencies of the plurality of similar users, population preference data comprising frequency distribution of a plurality of pieces of content information targeted by user behaviors of the plurality of similar users; and

recommending content information to the target user based at least in part on the population preference data.