Method, device, and server for friend recommendation

Patent metadata

Field	Value
Publication number	US-2016307101-A1
Application number	US-201415101314-A
Country	US
Kind code	A1
Filing date	Dec 12, 2014
Priority date	Dec 13, 2013
Publication date	Oct 20, 2016
Grant date	—

Abstract

Methods, devices, and servers for friend recommendation are provided. A user association set of a target user is obtained. Original data of each associated user in the user association set is obtained. The original data include location relationship data, associated friend data, time relationship data, or combinations thereof, between each associated user and the target user. The original data of each associated user is screened to obtain feature data to form a feature collection for each associated user. A pre-configured N-Tree prediction model is used to process the feature collection for a prediction calculation to obtain an association-predicting value for each associated user. According to the association-predicting value of each associated user, a friend user for the target user from the user association set is determined and recommended to the target user.

First claim

1 . A method for friend recommendation, comprising:
obtaining a user association set of a target user, wherein the user association set comprises a plurality of associated users that are associated with the target user; obtaining original data of each associated user in the user association set of the target user, wherein the original data comprise location relationship data, associated friend data, time relationship data, or combinations thereof, between each associated user and the target user; screening the original data of each associated user to obtain feature data to form a feature collection for each associated user; using a pre-configured N-Tree prediction model to process the feature collection for a prediction calculation to obtain an association-predicting value for each associated user; and according to the association-predicting value of each associated user, determining a friend user for the target user from the user association set and recommending the friend user to the target user.

2 . The method according to claim 1 , wherein the N-Tree prediction model comprises a prediction model based on gradient boosting decision tree (GBDT).

3 . The method according to claim 1 , wherein the step of using the pre-configured N-Tree prediction model to obtain the association-predicting value for each associated user comprises:
(a) determining feature collections of a plurality of predicted users; (b) configuring an initial weight value in the pre-configured N-Tree prediction model to respectively process the feature collection of each predicted user for the prediction calculation to determine the association-predicting value corresponding to the feature collection of each predicted user; (c) ranking each predicted user according to an amount of the association-predicting value to form a ranking list; (d) when a same-ranking rate between the obtained ranking list and a pre-set ranking list for the predicted users reaches a threshold value, outputting the N-Tree prediction model configured with the initial weight value as the pre-configured N-Tree prediction model; and (e) when the same-ranking rate between the obtained ranking list and the pre-set ranking list for the predicted users does not reach the threshold value, adjusting the weight value, and repeating steps (b)-(e) by: configuring the adjusted weight value in the pre-configured N-Tree prediction model to respectively process the feature collection of each predicted user for the prediction calculation to determine the association-predicting value corresponding to the feature collection of each predicted user, until the same-ranking rate between the obtained ranking list and the pre-set ranking list for the predicted users reaches the threshold value, and outputting the N-Tree prediction model configured with a finally-adjusted weight value as the pre-configured N-Tree prediction model.

4 . The method according to claim 3 , wherein the step of according to the association-predicting value of each associated user, determining a friend user for the target user from the user association set and recommending the friend user to the target user comprises:
determining an associated user having the association-predicting value greater than a pre-set prediction threshold value as the friend user of the target user; and depending on an amount of the association-predicting value of each determined friend user, ranking the determined friend users to provide a ranking result, and recommending one or more determined friend users to the target user according to the ranking result.

5 . The method according to claim 4 , wherein the step of obtaining the user association set of the target user comprises:
extracting marking information of the target user, and determining the plurality of associated users in the user association set according to the marking information, wherein the marking information comprises account information of the target user, location information of the target user, and combinations thereof.

6 . A non-transitory computer-readable storage medium comprising instructions stored thereon, wherein, when being executed, the instructions cause one or more processors of a device to perform the method according to claim 1 .

7 . A device for friend recommendation, comprising:
an obtaining module, configured to obtain a user association set of a target user, wherein the user association set comprises a plurality of associated users that are associated with the target user, wherein the obtaining module is further configured to obtain original data of each associated user in the user association set of the target user, wherein the original data comprise location relationship data, associated friend data, time relationship data, or combinations thereof, between each associated user and the target user; a collecting module, configured to screen the original data of each associated user to obtain feature data to form a feature collection for each associated user; a processing module, configured to use a pre-configured N-Tree prediction model to process the feature collection for a prediction calculation to obtain an association-predicting value for each associated user; and a recommending module configured, according to the association-predicting value of each associated user, to determine a friend user for the target user from the user association set and recommend the friend user to the target user.

8 . The device according to claim 7 , further comprising:
a pre-configuring module, configured to pre-configure the N-Tree prediction model comprising a prediction model based on gradient boosting decision tree (GBDT).

9 . The device according to claim 8 , wherein the pre-configuring module comprises:
a collection-determining unit, configured to determine feature collections of the plurality of predicted users; a first calculating unit, configured to respectively process the feature collection of each predicted user for the prediction calculation using the pre-configured N-Tree prediction model having an initial weight value to determine the association-predicting value corresponding to the feature collection of each predicted user; a ranking unit, configured to rank each predicted user according to an amount of the association-predicting value to form a ranking list; a first outputting unit configured, when a same-ranking rate between the obtained ranking list and a pre-set ranking list for the predicted users reaches a threshold value, to output the N-Tree prediction model configured with the initial weight value as the pre-configured N-Tree prediction model; a second calculating unit configured:
when the same-ranking rate between the obtained ranking list and the pre-set ranking list for the predicted users does not reach the threshold value, to adjust the weight value, and to use the pre-configured N-Tree prediction model having the adjusted weight value to respectively process the feature collection of each predicted user for the prediction calculation to determine the association-predicting value corresponding to the feature collection of each predicted user, and
to repeat adjustment of the weight value until the same-ranking rate between the obtained ranking list and the pre-set ranking list for the predicted users reaches the threshold value; and
a second outputting module, configured to output the N-Tree prediction model having a finally-adjusted weight value as the pre-configured N-Tree prediction model.

10 . The device according to claim 9 , wherein the recommending module comprises:
a friend determining unit, configured to determine an associated user having the association-predicting value greater than a pre-set predi

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Zhang Wenlong
• Jiao Feng
• Wang Bin
• Zeng Lei
• Chen Xiaohui

Classifications

• G06Q10/40 Primary

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

• G06F16/24578

  using ranking · CPC title

• H04L67/306

  User profiles · CPC title

• H04W4/21 Primary

  for social networking applications · CPC title

• G06N5/045 Primary

  Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence · CPC title