Method and apparatus for classifying object based on social networking service, and storage medium

Patent metadata

Field	Value
Publication number	US-9875294-B2
Application number	US-201515112491-A
Country	US
Kind code	B2
Filing date	Apr 28, 2015
Priority date	Apr 29, 2014
Publication date	Jan 23, 2018
Grant date	Jan 23, 2018

Abstract

Described are a method and an apparatus for classifying an object based on a social networking service. Feature information of an object may be acquired; a semantic vector of the feature information may be computed, the semantic vector being used to reflect a relevancy feature when segmented words exist in the feature information simultaneously; and the semantic vector of the feature information of the object may be input to a predetermined classifier, to obtain an initial category of the object after the object is classified.

First claim

What is claimed is:

1. A method for classifying an object based on a social networking service, the method being run in a computing device, and the computing device comprising a storage apparatus, one or more processors, and program instructions stored in the storage apparatus and executed by the one or more processors, the method comprising:
acquiring feature information of an object, the feature information comprising multiple segmented words;
computing a semantic vector of the feature information, the semantic vector being used to reflect a relevancy feature when the segmented words exist in the feature information simultaneously;
inputting the semantic vector of the feature information of the object to a predetermined classifier, to obtain an initial category of the object after the object is classified;
wherein the object comprises a user and/or a group, and after the computing a semantic vector of the feature information, the method comprises:
determining, for each object by using a semantic vector of feature information of the object, a predetermined number of similar objects corresponding to the object;
forming an object pair by using the object and each of the similar objects;
constructing a bipartite graph by using the object pair and a correspondence between a group and each user in the group; and
clustering an object in the bipartite graph by using a topic modeling algorithm, wherein
the bipartite graph at least comprises a correspondence formed by a first element and a second element; and when the first element in the correspondence is a group, the second element is a user in the group or a group that is in a same object pair of the group; or, when the first element in the correspondence is a user, the second element is a user that is in a same object pair of the user.

2. The method according to claim 1 , wherein after the clustering an object in the bipartite graph by using a topic modeling algorithm, the method further comprises:
computing the number of objects that belong to a same initial category after the clustering;
determining an initial category having the largest number of objects as a spread category; and
determining categories of the other objects after the clustering as the spread category.

3. The method according to claim 1 , wherein the determining, for each object by using a semantic vector of feature information of the object, a predetermined number of similar objects corresponding to the object comprises:
computing a similarity between the object and an object having a same type by using semantic vectors of feature information of the objects;
sequencing the objects having the same type in a descending order according to the similarity; and
sequentially selecting a predetermined number of objects from the objects having the same type after the sequencing, and determining the selected predetermined number of objects as similar objects of the object, wherein
the same type refers to that, when the object is a user, an object having a type the same as the type of the object is a user; and when the object is a group, an object having a type the same as the type of the object is a group.

4. The method according to claim 3 , wherein the computing a similarity between the object and an object having a same type by using semantic vectors of feature information of the objects comprises:
computing a cosine value between the semantic vector of the object and the semantic vector of the object having the same type; and
determining the cosine value as the similarity.

5. The method according to claim 1 , wherein after the inputting the semantic vector of the feature information of the object to a predetermined classifier, to obtain an initial category of the object after the object is classified, the method further comprises:
acquiring a confidence value that is determined when the predetermined classifier classifies the object; and
when the confidence value is greater than a predetermined confidence threshold, categorizing the object corresponding to the confidence value as a type determined by the predetermined classifier.

6. The method according to claim 1 , wherein the computing a semantic vector of the feature information comprises:
by using a pre-stored corpus, collecting statistics on a probability that a segmented word exists behind a predetermined number of specified segmented words, the segmented word and the predetermined number of specified segmented words forming feature information;
solving a predetermined mathematical model by using a back propagation learning algorithm, to obtain a semantic vector of each segmented word in the feature information; and
performing normalization after the semantic vectors of the segmented words in the same feature information are added, to obtain the semantic vector of the feature information.

7. The method according to claim 6 , wherein the predetermined mathematical model is:
y =softmax( U ·tan h ( Hx+d )+ Wx+b ),

wherein, y is a matrix formed by probabilities that the segmented word exists in pieces of feature information, x is a vector obtained by connecting semantic vectors of a predetermined number of specified segmented words in each piece of feature information in a head-to-tail manner, d and b are offset items, tan h and softmax are activation functions, U is a parameter from a hidden layer to an output layer of the predetermined mathematical model, and W is linear transformation from an input layer to the output layer of the predetermined mathematical model.

8. An apparatus for classifying an object based on a social networking service, comprising:
a storage apparatus;
one or more processors; and
one or more program modules, stored in the storage apparatus and executed by the one or more processors, the one or more program modules comprising:
a first acquisition module, configured to acquire feature information of an object, the feature information comprising multiple segmented words;
a logical operation module, configured to compute a semantic vector of the feature information, the semantic vector being used to reflect a relevancy feature when the segmented words exist in the feature information simultaneously;
a classifying module, configured to input the semantic vector of the feature information of the object to a predetermined classifier, to obtain an initial category of the object after the object is classified;
a first determining module, configured to determine, for each object by using a semantic vector of feature information of the object, a predetermined number of similar objects corresponding to the object;
a formation module, configured to form an object pair by using the object and each of the similar objects;
a construction module, configured to construct a bipartite graph by using the object pair formed by the formation module and a correspondence between a group and each user in the group; and
a clustering module, configured to cluster an object in the bipartite graph by using a topic modeling algorithm, wherein
the bipartite graph at least comprises a correspondence formed by a first element and a second element; and when the first element in the correspondence is a group, the second element is a user in the group or a group that is in a same object pair of the group; or, when the first element in the correspondence is a user, the second element is a user that is in a same object pair of the user.

9. The apparatus according to claim 8 , further comprising:
a computation module, configured to compute the number of objects that belong to a same initial category after the clustering;
a second determining module, configured to determine an initial category having the greatest number, computed by the

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Yu Jianxing
• Yi Lingling
• He Peng
• Chen Chuan

Classifications

• G06Q10/40

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

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06F16/9024

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F17/16

  Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)} · CPC title

• G06F16/285 Primary

  Clustering or classification · CPC title