CN110162692B - User label determination method and device, computer equipment and storage medium - Google Patents

User label determination method and device, computer equipment and storage medium Download PDF

Info

Publication number
CN110162692B
CN110162692B CN201811503599.5A CN201811503599A CN110162692B CN 110162692 B CN110162692 B CN 110162692B CN 201811503599 A CN201811503599 A CN 201811503599A CN 110162692 B CN110162692 B CN 110162692B
Authority
CN
China
Prior art keywords
user
current
target
label
tag
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201811503599.5A
Other languages
Chinese (zh)
Other versions
CN110162692A (en
Inventor
陈培炫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN201811503599.5A priority Critical patent/CN110162692B/en
Publication of CN110162692A publication Critical patent/CN110162692A/en
Application granted granted Critical
Publication of CN110162692B publication Critical patent/CN110162692B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a user label determination method, a device, computer equipment and a storage medium, wherein the method comprises the steps of obtaining a user set, wherein the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels; determining an initial label value corresponding to each user according to the label corresponding to each user in the user set; calculating to obtain target label reference degrees among the users according to the social association degrees and the feature similarity among the users in the user set; for the current user in the second sub-user set, obtaining a target label value corresponding to the current user according to a target label reference degree between the current user and the user in the user set and an initial label value corresponding to the user in the user set; and determining a target label corresponding to the current user according to the target label value. The label obtained by the method has high accuracy and saves network resources.

Description

User label determination method and device, computer equipment and storage medium
Technical Field
The present invention relates to the field of information processing, and in particular, to a user tag determination method, apparatus, computer device, and storage medium.
Background
With the development of the internet, users on an internet platform are more and more, and in many cases, the tags of the users need to be determined so as to manage and maintain the users according to the user tags. At present, when a user's label is determined, the user usually writes the label by the user's consciousness, but the data written by the user in the internet is often incomplete, and the label is marked by the user's manpower, when the label is marked by the user's manpower, the server sends the relevant information of the user to the client, the client returns the label to the server after the user's label is determined by the manpower according to the relevant information of the user, the server sends the relevant information of the next user to the client for label judgment, however, the label is judged by the manpower and is influenced by the subjective state of the person, the accuracy of the obtained label is low, and the terminal and the server need to interact for many times, which wastes network.
Disclosure of Invention
Therefore, it is necessary to provide a user tag determination method, an apparatus, a computer device, and a storage medium for solving the problems that the obtained tag has low accuracy, and a terminal and a server need to interact for many times, which wastes network resources.
A method of user tag determination, the method comprising: acquiring a user set, wherein the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels; determining an initial label value corresponding to each user according to the label corresponding to each user in the user set; calculating to obtain target label reference degrees among the users according to the social association degrees and the feature similarity among the users in the user set; for the current user in the second sub-user set, obtaining a target label value corresponding to the current user according to a target label reference degree between the current user and the user in the user set and an initial label value corresponding to the user in the user set; and determining a target label corresponding to the current user according to the target label value.
A user tag determination apparatus, the apparatus comprising: the system comprises a user set acquisition module, a label matching module and a label matching module, wherein the user set acquisition module is used for acquiring a user set, and the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels; an initial tag value determining module, configured to determine, according to a tag corresponding to each user in the user set, an initial tag value corresponding to each user; the target tag reference degree calculation module is used for calculating target tag reference degrees among users according to the social association degrees and the feature similarity among the users in the user set; a target label value calculation module, configured to, for a current user in the second sub-user set, obtain a target label value corresponding to the current user according to a target label reference between the current user and users in the user set and an initial label value corresponding to the users in the user set; and the target label determining module is used for determining a target label corresponding to the current user according to the target label value.
In one embodiment, the target tag reference degree comprises a first target tag reference degree and a second target tag reference degree, and the target tag reference degree calculation module is configured to: calculating to obtain a first target tag reference degree between users according to the social association degree between the users in the user set, and calculating to obtain a second target tag reference degree between the users according to the feature similarity degree between the users in the user set; the target tag value calculation module includes: a first influence label value calculating unit, configured to, for a current user in the second sub-user set, obtain a first influence label value corresponding to the current user according to a first target label reference degree corresponding to the current user and users in the user set and an initial label value corresponding to the users in the user set; a second influence label value calculating unit, configured to, for a current user in the second sub-user set, obtain a second influence label value corresponding to the current user according to a second target label reference degree corresponding to the current user and users in the user set and an initial label value corresponding to the users in the user set; and the target label value calculating unit is used for determining the target label value corresponding to the current user according to the first influence label value and the second influence label value corresponding to the current user.
In one embodiment, the second influence tag value calculation unit includes: a feature graph constructing subunit, configured to construct a feature graph by using each user in the user set as a node and using the second target label reference degree as an edge weight of an edge; a node obtaining subunit, configured to obtain a current node corresponding to the current user and a neighboring node of the current node in the feature map; an updating subunit, configured to use an initial tag value corresponding to each user as a current tag value of a node corresponding to the feature map, and update the current tag value of the current node according to the current tag value of the neighboring node in the feature map and the edge weight between the current node and the neighboring node; and the returning subunit is configured to return to the step of updating the current label value of the current node according to the current label value of the adjacent node in the feature map and the edge weights of the current node and the adjacent node until a convergence condition is met, so as to obtain a second influence label value corresponding to the current user.
In one embodiment, the target tag reference calculation module comprises: a target reference user set obtaining unit, configured to obtain a current user from the second sub-user set, and filter from the user set according to a social association relationship of the current user to obtain a target reference user set corresponding to the current user; the second target tag reference degree calculating unit is used for calculating to obtain second target tag reference degrees of the current user and the target reference users according to the feature similarity of the current user and each target reference user; the feature map construction subunit is configured to: and taking each user in the user set as a node, taking a node corresponding to the target reference user as an adjacent node of the current node, and taking the second target label reference degree of the current user and the target reference user as the weight of the edge to construct and obtain a feature graph.
In one embodiment, the target reference user set obtaining unit is configured to: acquiring a first direct association user set corresponding to the current user from the user set according to the social association relation of the current user; screening a first reference user from the first direct association user set according to the feature similarity of the current user and a first direct association user; acquiring a second direct association user set corresponding to the first reference user from the user set according to the social association relationship of the first reference user; screening a second reference user from the second direct association user set according to the feature similarity of the first reference user and the second direct association user; and taking the first reference user and the second reference user as target reference users in the target reference user set.
In one embodiment, the second target tag reference calculation unit is configured to: counting the feature similarity of the current user and each target reference user to obtain a feature similarity counting result; and acquiring the current feature similarity of the current user and the target reference user, and acquiring a second target label reference degree of the current user and the target reference user according to the ratio of the current feature similarity to the feature similarity statistical result.
In one embodiment, the first influencing tag value calculation unit is to: taking each user in the user set as a node, and taking the first target label reference degree as the edge weight of an edge to obtain a social association graph; acquiring a current node corresponding to the current user and an adjacent node of the current node in the social association graph; updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node; and returning to the step of updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node until a convergence condition is met, and obtaining a first influence label value corresponding to the current user.
In one embodiment, the target tag value calculation unit is to: acquiring a first weight corresponding to the first influence label value and a second weight corresponding to the second influence label value; and carrying out weighted summation according to the first influence label value and the corresponding first weight, the second influence label value and the corresponding second weight to obtain a target label value corresponding to the second user.
In one embodiment, the initial tag value determination module comprises: a target label category obtaining unit, configured to obtain a target label category corresponding to each first user; a first initial tag vector obtaining unit, configured to obtain an initial tag vector corresponding to each first user according to a target tag category corresponding to the first user; and a second initial tag vector obtaining unit, configured to obtain a preset tag vector as an initial tag vector corresponding to a second user in the second sub-user set, where values of the preset tag vector are consistent.
In one embodiment, the first initial tag vector deriving unit is configured to: for each target label category corresponding to the first user, acquiring a corresponding first preset value as a corresponding vector value, and for each non-target label category corresponding to the first user, acquiring a corresponding second preset value as a corresponding vector value; and combining the vector value corresponding to each target label category and the vector value corresponding to each non-target label category into an initial label vector corresponding to the first user, wherein the first preset value is different from the second preset value.
A computer device comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to carry out the steps of the above-mentioned user tag determination method.
A computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, causes the processor to carry out the steps of the above-mentioned user tag determination method.
According to the user label determining method, the user label determining device, the computer equipment and the storage medium, when the label of the user is determined, the user without the determined label is predicted according to the user with the determined label, the feature similarity and the association degree between the users are obtained through calculation, and the label of the user can be obtained without manual labeling, so that the obtained label is high in accuracy and network resources are saved.
Drawings
FIG. 1 is a diagram of an application environment for a user tag determination method provided in one embodiment;
FIG. 2 is a flow diagram of a method for user tag determination in one embodiment;
fig. 3 is a flowchart illustrating that, in an embodiment, for a current user in the second sub-user set, a target tag value corresponding to the current user is obtained through calculation according to a target tag reference degree corresponding to each user in the user set and an initial tag value corresponding to each user in the user set;
fig. 4A is a flowchart illustrating obtaining a second influence tag value corresponding to a current user according to a second target tag reference degree corresponding to the current user and a user in a user set and an initial tag value corresponding to the user in the user set in one embodiment;
FIG. 4B is a schematic illustration of a feature map in one embodiment;
FIG. 5 is a flow diagram that illustrates computing a second target tag degree of reference between users based on a degree of feature similarity between users in the set of users, under an embodiment;
fig. 6A is a flowchart illustrating that, for a current user in the second sub-user set, a first influence tag value corresponding to the current user is obtained according to a first target tag reference degree corresponding to the current user and users in the user set and an initial tag value corresponding to the users in the user set in one embodiment;
FIG. 6B is a diagram of a social association graph, under an embodiment;
FIG. 7A is a flow diagram of a method for user tag determination in one embodiment;
FIG. 7B is a schematic diagram illustrating a method for user tag determination in one embodiment;
FIG. 8 is a block diagram showing the structure of a user tag determination apparatus in one embodiment;
FIG. 9 is a block diagram showing the structure of a user tag determination unit in one embodiment;
FIG. 10 is a block diagram showing an internal configuration of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
It will be understood that, as used herein, the terms "first," "second," and the like may be used herein to describe various elements, but these elements are not limited by these terms unless otherwise specified. These terms are only used to distinguish one element from another. For example, the first set of sub-users may be referred to as a second set of sub-users, and similarly, the second set of sub-users may be referred to as the first set of sub-users, without departing from the scope of the present application.
Fig. 1 is a diagram of an application environment of a user tag determination method provided in an embodiment, as shown in fig. 1, in the application environment, including a terminal 110 and a server 120. The server 120 may store target tags corresponding to each user in the user set, and the server 120 may manage the user according to the target tags corresponding to each user in the user set, for example, send corresponding push information to the terminal 110 corresponding to the user according to the target tags of the user, and if the target tag corresponding to the user is a favorite car, push information related to the car to the terminal 110. And if the target tag corresponding to the user likes to travel, pushing the information related to the tourist attractions to the terminal 110. The target tag corresponding to each user in the user set is obtained by the server 120 according to the user tag determination method provided by the embodiment of the present invention.
The server 120 may be an independent physical server, or may be a server cluster formed by a plurality of physical servers, and may be a cloud server providing basic cloud computing services such as a cloud server, a cloud database, a cloud storage, and a CDN. The terminal 110 may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, and the like. The terminal 110 and the server 120 may be connected through communication connection manners such as bluetooth, USB (Universal Serial Bus), or network, which is not limited herein.
As shown in fig. 2, in an embodiment, a method for determining a user tag is provided, and this embodiment is mainly illustrated by applying the method to the server 120 in fig. 1. The method specifically comprises the following steps:
step S202, a user set is obtained, wherein the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels.
Specifically, the user set includes a plurality of users, the plurality refers to more than two (including two), the number of the users in the first sub-user set and the second sub-user set may also include a plurality, and the specific number may be set according to needs. For example, for social application WeChat, assuming there are eight hundred million users on the WeChat, the number of user sets may be eight hundred million. Of course, part of the user group user can be selected from the eight hundred million users, for example, users in the same city can be selected as the user group. The tags are used to represent characteristics of the user and may be used to classify the user. The user's tag may include a variety of types, and may have different definitions according to the specific situation. For credit evaluation, for example, the user's label may include categories of very good credit, medium credit, and poor credit. For the advertisement click prediction, the user's tag may include both click and no click categories. For resource recommendations, such as financial product recommendations, the user's tags may include conservative financial users as well as adventure financial users. A user who has determined a tag means that the tag of the user is known. A user with an undetermined tag means that the tag of the user is unknown and needs further determination. The label of each first user in the first sub-user set is determined, and the label of each second user in the second sub-user set is not determined, and needs to be determined according to the label of the first user in the first sub-user set.
Step S204, determining an initial label value corresponding to each user according to the label corresponding to each user in the user set.
Specifically, the initial tag value is obtained according to a tag corresponding to a user, for a first user with a determined tag, different tags may correspond to different initial tag values, and for a second user without a determined tag, a preset tag value may be obtained as the initial tag value.
In one embodiment, determining an initial tag value corresponding to each user according to a tag corresponding to each user in the user set includes: acquiring a target label category corresponding to each first user in a first sub-user set; obtaining an initial label vector corresponding to each first user according to the target label category corresponding to each first user; and acquiring a preset label vector as an initial label vector corresponding to a second user in the second sub-user set, wherein the values of the preset label vector are consistent.
Specifically, the initial tag vector may be a vector value of a class of tags corresponding to a preset position. For the target users with determined labels, the initial label vectors corresponding to different target label categories are different, so that the labels of the users can be distinguished. And for the user without the label, acquiring a preset label vector, wherein the preset label vector has consistent values, which indicates that the user label is not distinguished. The preset tag vector may be a vector whose vector values are all 0.
In one embodiment, obtaining the initial tag vector corresponding to the first user according to the target tag category corresponding to each first user includes: for each target label category corresponding to the first user, acquiring a corresponding first preset value as a corresponding vector value, and for each non-target label category corresponding to the first user, acquiring a corresponding second preset value as a corresponding vector value; and combining the vector value corresponding to each target label category and the vector value corresponding to each non-target label category into an initial label vector corresponding to the first user.
Specifically, the first preset value and the second preset value can be set as required, and the first preset value is different from the second preset value. In one embodiment, the first preset value may be 1, and the second preset value may be 0. For example, assuming that the first sub-user set includes L first users, the second sub-user set includes U users, and the tag has m categories, where m is a positive integer greater than or equal to 2, the tag value corresponding to a user may be represented by a vector, the vector has m vector values, one category of the tag corresponds to a vector value at a preset position, if the tag corresponding to a user is an f-th category of the tag, an f-th vector value in the initial tag vector is a first preset value, for example, 1, and the others are second preset values, for example, 0. For a second user for whom the tag is not determined, each vector value of the corresponding vector may be a second preset value, for example, 0. For an actual example, assuming that the label corresponding to the user a is the third label, the vector corresponding to the user a is (0, 0, 1, 0).
In one embodiment, a matrix F composed of initial label vectors corresponding to respective users may be used for the matrix (1) representation, in the matrix (1), one row represents one initial label vector, and n is equal to the sum of the number L of the first users and the number U of the second users, which represents the number of users in the user set.
Figure BDA0001898854000000081
And step S206, calculating the reference degree of the target label among the users according to the social association degree and the feature similarity among the users in the user set.
Specifically, the tag reference degree indicates the influence degree of the tag of one user on the tag of another user, and the greater the tag reference degree, the greater the influence of the tag of one user on the tag of another user. The target tag reference degree between the user a and the user b may include a target tag reference degree of the user a to the user b and a target tag reference degree of the user b to the user a, and the target tag reference degree of the user a to the user b and the target tag reference degree of the user b to the user a may be the same or different. If the difference is different, when the label of the user a is calculated according to the label of the user b, calculating the reference degree of the target label of the user a by adopting the user b; and when the label of the user b is calculated according to the label of the user a, calculating the target label reference degree of the user b by adopting the user a. The social relevance degree represents the degree of relevance of the user on the society, and the greater the social relevance degree, the more compact the relevance is. The social association degree may be obtained according to an association relationship of the user on the social network, where the association relationship refers to a relationship between the user and the user on the social network, and may include a direct association relationship and an indirect association relationship. Such as friend relationships on social applications, users of the same social group, and teammate relationships of game players on the gaming platform, among others. The social network refers to a relationship network formed based on social contact, such as WeChat, QQ, microblog and relationship networks in real society. The Social application may be an instant messaging application, an SNS (Social Network Services) application, a bar application, or the like. Instant messaging applications may include WeChat, QQ, and MSN, among others. SNS applications may include, but are not limited to, personal networks, Facebook, and the like. The feature similarity represents the degree of similarity of the user on the features, and the greater the feature similarity, the more similar the features representing the user. The target label reference degree between the two users is calculated according to the social association degree and the feature similarity between the two users. The social relevance and the feature similarity may be integrated to obtain an integrated target tag reference degree, and the target tag reference degree may also include a first target tag reference degree and a second target tag reference degree, where the first target tag reference degree is calculated according to the social relevance, and the social relevance and the first target tag reference degree form a positive correlation. The second target label reference degree is obtained through calculation according to the feature similarity, and the feature similarity and the first target label reference degree form a positive correlation relationship. The negative correlation relationship means: the two variables have different changing directions, and when one variable changes from large to small, the other variable changes from small to large. The positive correlation relationship means that: the two variables have the same changing direction, and when one variable changes from large to small, the other variable changes from large to small.
In one embodiment, the target tag reference degree between the user and each user in the user set can be calculated, and in order to reduce the calculation amount, the target tag reference degree between partial users can also be calculated.
In an embodiment, for a current user, a user having a social association relationship with the current user may be acquired from a user set according to the social association relationship between the users as a reference user, a target tag reference degree between the current user and the reference user is calculated according to the social association degree between the current user and the reference user and a feature similarity, and for a non-reference user of the current user in the user set, the target tag reference degree between the current user and the non-reference user may be a preset value, for example, 0. As a practical example, assume that the user set includes 4 users: a1, a2, a3 and a 4. Target tag references between users a1 and a2, a1 and a3, a1 and a4, a2 and a3, a2 and a4, a3 and a4 can be calculated according to the social relevance and the feature similarity. However, in order to reduce the calculation amount, assuming that there is no association between a2 and a3 or the social association degree is relatively small, the target tag reference degrees of a2 and a3 are not calculated according to the social association degree and the feature similarity degree, and the target tag reference degree between a2 and a3 is a preset value, for example, 0.
In one embodiment, a tag referential degree between a second user in the second sub-user set and a first user in the first sub-user set may be calculated to obtain a tag of the second user by referring to the tag of the first user and a corresponding target tag referential degree. For example, assume that the first set of sub-users includes 2 users: a1, a2, the second sub-user set comprises 2 users: a3 and a4, then the target label reference degree between a3 and a1, the target label reference degree between a3 and a2, the target label reference degree between a4 and a1, and the target label reference degree between a4 and a2 can be calculated.
In one embodiment, a first user having a social association relationship with a second user may be selected from the first sub-user set as a reference user according to the social association relationship of the second user, and a target tag reference degree between the first user and the reference user is calculated according to the social association degree and the feature similarity between the second user and the reference user.
In one embodiment, a label referential between a second user in the second set of sub-users and a first user in the first set of sub-users is calculated, and a label referential between a second user in the second set of sub-users and a second user in the second set of sub-users is calculated. Thus, when determining the label of the second user, the label of the second user may be obtained by referring to the labels of the users in the first sub-user set and the labels of the users in the second sub-user set.
In one embodiment, a first user having a social association relationship with a second user may be selected from the first sub-user set as a first reference user according to the social association relationship of the second user, and a second user having a social association relationship with the second user may be selected from the second sub-user set as a second reference user according to the social association relationship of the second user. The number of the first reference users and the number of the second reference users may be the same or different. And calculating to obtain a target tag reference degree between the current user and the first reference user according to the social association degree and the feature similarity between the current user and the first reference user, and calculating to obtain a target tag reference degree between the current user and the second reference user according to the social association degree and the feature similarity between the current user and the second reference user.
In one embodiment, a first user in the first sub-user set, whose social relevance to the second user is greater than a first relevance threshold, is taken as a first reference user. And taking the second user with the social relevance degree larger than a second relevance degree threshold value with the second user in the second sub-user set as a second reference user. Wherein the second relevancy threshold may be greater than the first relevancy threshold. Since the tag of the first user is determined and the tag of the second user is obtained by referring to the tag of the user, the accuracy of the tag of the first user is high, so that the second association threshold is greater than the first association threshold, when the tag of the second user is determined, more tags of the first user who has determined the tag and has a social association relationship with the second user can be referred to, and the accuracy of the obtained tag is improved.
In one embodiment, the social relevance is derived from the incidence between users. For example, the determination may be based on at least one of the user's interaction behavior and the length of the social relationship chain. The interactive behavior may include the number of contacts between users, the frequency of contacts, and so on. The social relationship chains are sequentially formed according to the social association sequence of the intermediate users which connect the two users and are between the two users. And (3) regarding the user as a node, and connecting the nodes with the direct association relationship to obtain a social relationship chain. The length of the relationship chain can be obtained by the number of paths. For example, for a1 and a4, assuming that the friend of a1 is a2, the friend of a2 is a3, and the friend of a3 is a4, a2 and a3 are intermediate users, the target relationship chain is a1 → a2 → a3 → a4, and the length of the relationship chain is 3. The greater the social relevance if the interactive behavior indicates that the user is more closely connected, and the greater the social relevance if the length of the relationship chain is shorter. For example, it may be set that if the relationship chain length is 1, the social association degree is 1, and if the relationship chain length is 2, the social association degree is 0.6. The social relevance is 0.8 if the frequency of connections between users a year is 2 days, and 0.3 if the frequency of connections between users a year is 10 days.
In one embodiment, for users with a friend relationship, the social association is 1, otherwise it is 0.
In one embodiment, the feature similarity is obtained according to the distance between one or more features of the users, the features are used for representing the attributes of the users, and the features corresponding to the calculated feature similarity can be determined according to needs. Attributes such as age, gender, and consumption level of the user may be selected as the characteristics. The feature similarity and the feature distance can be used to measure whether the images are similar. The smaller the feature distance, the more similar the images, and the greater the similarity, the more similar the images. The feature distance may be in a negative correlation with the feature similarity, i.e., if the feature distance is large, the feature similarity is small. And calculating the feature similarity according to the feature distance. For example, the reciprocal of the feature distance may be the feature similarity. For another example, the feature similarity w between the user i and the user j is calculatedijThe corresponding formula can be expressed as follows:
Figure BDA0001898854000000111
where exp refers to an exponential function with a natural constant e as the base in higher mathematics, | xi-xj||2Refers to the combined distance of the features of user i and user j, such as the sum or average of the respective feature distances of the users. The distance of the feature may be set as needed, for example, a relationship between an age difference value of the user and the feature distance may be set, and the greater the age difference value, the greater the feature distance. The relationship of the gender of the user to the characteristic distance may also be set. After the distances corresponding to the respective features are obtained, the distances corresponding to the respective features, for example, the sum of the distances, are integrated as a total distance. For example, if the sexes are the same, the distance is 0, and if the sexes are different, the distance is 0.05. The characteristic distance may be 0.02 assuming that the i and j age differences are 10 years, and the user distance may be 0.1 if the i and j age differences are 20 years. a ofThe value may be set as desired but cannot be 0, e.g. a may be 1.
In an embodiment, the first target tag reference degree between the users may be calculated according to the social association degrees between the users in the user set, for example, the social association degree may be used as the first target tag reference degree, and the social association degree may also be further processed to obtain the first target tag reference degree, for example, the social association degree is normalized to obtain the first target tag reference degree. For example, the first target tag reference of user j to user i can be formulated
Figure BDA0001898854000000121
Is shown in the specification, wherein wijRepresenting the social relevance of i and j, and n is the number of users in the user set. The matrix P1 composed of the first target tag reference may be as shown in matrix (2), where in matrix (2), P isijRepresents the first target label referential degree of the user j to the user i, namely the probability that the label is transferred from the user j to the user i. It is understood that the value of the k-th row and the k-th column in the matrix P1, i.e. the user's label reference to himself/herself, may be a default value, such as 1.
Figure BDA0001898854000000122
In one embodiment, a second target tag referential degree between the users can be calculated according to the feature similarity degree between the users in the user set. For example, the feature similarity may be used as a second target tag reference degree, and the feature similarity may also be further processed to obtain the second target tag reference degree, for example, the feature similarity is normalized to obtain the second target tag reference degree. For example, the second target label reference degree of the user j and the user i can be formulated by
Figure BDA0001898854000000123
Is shown in the specification, wherein wij' represents the feature similarity of i and j, and n is the number of users in the user set. Then composed of the second target tag referenceThe matrix P2 may be as shown in matrix (3), where in matrix (3), P'ijAnd the second target label reference degree of the user j to the user i is represented, namely the probability of transferring the label from the user j to the user i. It is understood that the value of the k-th row and the k-th column in the matrix P2, i.e. the user's label reference to himself/herself, may be a default value, such as 1.
Figure BDA0001898854000000131
Step S208, for the current user in the second sub-user set, obtaining a target label value corresponding to the current user according to the target label reference degree between the current user and the user in the user set and the initial label value corresponding to the user in the user set.
Specifically, the current user is a current second user when the target tag value is calculated, and since the second sub-user set includes a plurality of second users, when the target tag value corresponding to the user a is determined, the user a is the current user, and when the target tag value corresponding to the user b is determined, the user b is the current user. When the target label value is calculated, the target label value is calculated by using the target label reference degree of the current user and the initial label value corresponding to the user. If the initial tag value of the current user is a value, the target tag value can be obtained according to the sum of products obtained by multiplying the reference degree of the target tag by the initial tag value, and if the initial tag value of the current user is an initial tag vector formed by a plurality of vector values, and one class of tags corresponds to the vector value of the preset position, the target tag value is the target vector value.
In one embodiment, when determining the target tag value corresponding to the current user, the initial tag values of one or more users may be referred to. If the initial label value of one user is referred, the target label value corresponding to the current user can be obtained according to the initial label value of the user and the target label reference degree between the user and the current user. And if the initial label values of a plurality of reference users are referred, calculating by integrating the target label reference degrees between the current user and each reference user and the initial label values of the reference users to obtain the target label value corresponding to the current user. For example, the products of the initial tag values of the respective reference users and the target tag reference degrees between the corresponding current user and the reference user may be calculated, and the respective products are added to obtain the target tag value corresponding to the current user.
For example, when the target tag reference degree between the second user in the second sub-user set and the first user in the first sub-user set is obtained, the target tag value corresponding to the current user may be obtained according to the sum of products of the target tag reference degree between the current user and the first user and the initial tag value of the corresponding first user. To take a practical example, assume that the first user has 2: a2 and a3, the target label reference degree between the current user a1, a1 and a2, a1 and a3 is p respectively12、p13If the initial tag value of a2 is h2 and the initial tag value of a3 is h3, the target tag value corresponding to a1 can be calculated as p12*h2+p13*h3。
For example, when the label reference degree between the second user in the second sub-user set and the first user in the first sub-user set and the label reference degree between the second user in the second sub-user set and the second user in the second sub-user set are obtained, the target label value may be obtained according to the sum of the products of the target label reference degree between the current user and the first user and the corresponding initial label value of the first user, and the sum of the products of the target label reference degree between the current user and the second user and the corresponding initial label value of the second user.
In an embodiment, when the target tag reference degree includes a first target tag reference degree and a second target tag reference degree, a first influence tag value corresponding to the current user may be obtained through calculation according to the first target tag reference degree and an initial tag value corresponding to a user in the user set, a second influence tag value corresponding to the current user may be obtained through calculation according to a second target tag reference degree corresponding to a user in the user set and an initial tag value corresponding to a user in the user set, and the target tag value may be obtained according to the first influence tag value and the second influence tag value.
In one embodiment, when one type of tag corresponds to a vector value at a preset position of an initial tag vector, a target tag reference degree between a current user and a user may be multiplied by the vector value corresponding to each tag of the user to obtain a product, and then the products corresponding to the same type of tag are added to obtain a vector value corresponding to the tag of the type in the target tag vector.
As a practical example, assume that the reference users include a2, a3 and a4, and the first target tag reference degrees between the current users are a1, a1 and a2, a1 and a3, a1 and a4 are p12、p13And p14,The label categories are three categories, so the vector values of the initial label vector are 3, the first vector value is the value corresponding to the first label category, the second vector value is the value corresponding to the second label category, and the third vector value is the value corresponding to the third label category. Assuming that initial label vectors corresponding to a1, a2, a3 and a4 are (b11, b12, b13), (b21, b22, b23), (b31, b32, b33) and (b41, b42, b43), respectively, a first influence label value corresponding to a current user a1 is a vector, which is represented as (e11, e12, e13), and e11 is b11+ p12*b21+p13*b31+p14*b41,e12=b12+p12*b22+p13*b32+p14*b42,e13=b13+p12*b23+p13*b33+p14B 43. Similarly, the second influence label value corresponding to the current user a1 can be obtained as a vector, which is expressed as (g11, g12, g13), and the target label vector can be (e11+ g11, e12+ g12, e13+ g13)
When the matrix formed by the first target tag reference degree is the matrix P1, and the matrix formed by the second target tag reference degree is the matrix P2, the matrix is expressed as follows: multiplying the matrix P1 with the matrix F to obtain a first target label matrix, multiplying the matrix P2 with the matrix F to obtain a second target label matrix, and combining the first target label matrix and the second target label matrix to obtain a target label matrix, wherein one row of the target label matrix can represent a target label value corresponding to one user. For each label class according to the principle of matrix division when performing matrix multiplicationAnd the label value is obtained by calculation according to the target label reference degree and the label value of the same label. That is, the vector value corresponding to the r-th tag of the current user is obtained according to the target tag reference degrees of the current user and the vector value corresponding to the r-th tag of the user. This can be embodied by the rule of matrix multiplication, for example, when the matrix P1 is multiplied by the matrix F to obtain the matrix F1, the value F of the h row and r column in F1hrEqual to the sum of the products of multiplication of the matrix value of the h-th row of the matrix P1 with the matrix value of the r-th column of the matrix F.
In one embodiment, matrix P1 is multiplied by matrix F to obtain a first target label matrix, and matrix P2 is multiplied by matrix F to obtain a second target label matrix. And combining the first target label matrix and the second target label matrix to obtain a target label matrix. Obtaining a target label matrix FEyes of a userThe formula of (a) can be shown as follows, wherein q1 and q2 are weights, and specific numerical values are set according to needs, and can be 1, for example.
F1=P1*F (4)
F2=P2*F (5)
FEyes of a user=F1*q1+F2*q2 (6)
In one embodiment, only one calculation or multiple iterations may be performed when the target tag value corresponding to the current user is obtained through calculation according to the target tag reference degree corresponding to the current user and the users in the user set and the initial tag value corresponding to the users in the user set, and during the iterations, an intermediate tag value may be obtained according to the target tag reference degree corresponding to the users and the initial tag value corresponding to the users in the user set, and then the target tag value is obtained through calculation according to the intermediate tag value and the target tag reference degree corresponding to the users.
And step S210, determining a target label corresponding to the current user according to the target label value.
Specifically, after the target tag value is obtained, the target tag is determined according to the size of the target tag value. For example, for the target label matrix, a target label vector corresponding to each current user may be obtained, and the target label is determined according to the magnitude of each vector value in the target label vector. The label corresponding to the largest vector value in the vector values may be used as the target label, or a label corresponding to a vector value larger than a preset value, for example, 0.8, may be used as the target label. As an actual example, if the target label vector corresponding to the current user is (0.1, 0.3, 0.6), the label corresponding to 0.6 represents the credit difference, the label corresponding to 0.3 represents the credit medium, and the label corresponding to 0.1 represents that the information is good, the target label corresponding to the current user is the credit difference.
In one embodiment, after obtaining the target tag corresponding to the second user, the second user may be stored in association with the corresponding target tag.
In one embodiment, information can be pushed to a terminal corresponding to the second user according to the target tag of the second user. For example, for conservative financing users, funds with low financing risk are recommended, and for aggressive financing users, funds with high financing risk but high return are recommended.
In one embodiment, different user management policies may be implemented based on the target tag of the second user, e.g., for a user with a poor credit tag, the user is added to a blacklist of loaned products.
According to the user label determining method provided by the embodiment of the invention, when the label of the user is determined, the user without the determined label is predicted according to the user with the determined label, and the user label is obtained by calculating the characteristic similarity and the association degree among the users in a comprehensive manner without manual labeling, so that the obtained label has high accuracy and network resources are saved.
In an embodiment, when the target tag reference degree includes a first target tag reference degree and a second target tag reference degree, as shown in fig. 3, for the current user in the second sub-user set, the obtaining the target tag value corresponding to the current user according to the target tag reference degree corresponding between the current user and the user in the user set and the initial tag value corresponding to the user in the user set in step S208 includes:
step S302, for the current user in the second sub-user set, obtaining a first influence tag value corresponding to the current user according to the first target tag reference degree corresponding to the current user and the users in the user set and the initial tag value corresponding to the users in the user set.
Specifically, for a current user, a first target tag reference degree between the current user and the user is obtained, an initial tag value corresponding to the user is obtained, an influence tag value of the user on the current user is obtained according to the first target tag reference degree and the initial tag value corresponding to the user, and the first influence tag value is obtained by combining the influence tag value of the user on the current user, for example, the first influence tag value may be obtained according to the sum of the influence tag values of the reference users on the current user. Multiple iterations of the calculation may be performed in calculating the first impact tag value. For example, for each current user, during the first calculation, a middle tag value may be obtained according to the first target tag reference degree corresponding to the user in the current user and the user in the user set and the initial tag value corresponding to the user in the user set, then the middle tag value is updated according to the updated middle tag value of the user and the first target tag reference degree corresponding to the user in the current user and the user in the user set, and the step of updating the middle tag value according to the updated middle tag value of the current user and the first target tag reference degree corresponding to the user in the current user and the user in the user set is repeated one or more times to obtain the first impact tag value corresponding to the current user. The method of iterative computation is represented by a matrix as follows: for the matrix P1 composed of the first target tag reference degree, the initial tag matrix F composed of the initial tag vector, P1 may be multiplied by F to obtain updated F, then P1 is multiplied by the updated F, and the step of multiplying P1 by the updated F is repeated until a convergence condition is satisfied, where the convergence condition includes at least one of the repetition number reaching a preset number and the change of the previous updated F and the current updated F being less than a preset value. Wherein the change of the F of the previous update and the F of the current update can be represented by the sum of squares of the difference values of the matrix values at the same position. Due to multiple iterations, useful information can be learned from partially tagged users and from untagged users according to the social relevance to determine the tags of the untagged users. That is, when the first influence tag value corresponding to the current user is calculated, the tags corresponding to the first user and the second user both influence the tag value of the current user, and the influence degree of the tag value, that is, the first target tag reference degree, is obtained according to the social association degree, so that for users with closer social connections, the mutual influence degree is greater, and the tags of the users with closer social connections can be the same or similar through multiple tag value propagation.
Step S304, for the current user in the second sub-user set, obtaining a second influence label value corresponding to the current user according to a second target label reference degree corresponding to the current user and the users in the user set and the initial label value corresponding to the users in the user set.
Specifically, for a current user, a second target tag reference degree between the current user and the user is obtained, an initial tag value corresponding to the user is obtained, an influence tag value of the user on the current user is obtained according to the second target tag reference degree and the initial tag value corresponding to the user, and a second influence tag value is obtained by combining the influence tag value of the user on the current user, for example, the second influence tag value may be obtained according to a sum of the influence tag values of the user on the current user. Multiple iterations of the calculation may be performed in calculating the second impact tag value. For example, for each current user, during the first calculation, a middle tag value may be obtained by updating according to a second target tag reference degree corresponding to the current user and the user in the user set and an initial tag value corresponding to the user in the user set, then the middle tag value may be updated according to the updated middle tag value of the user and a second target tag reference degree corresponding to the current user and the user in the user set, and the step of updating the middle tag value according to the updated middle tag value of the current user and the second target tag reference degree corresponding to the current user and the user in the user set is repeated one or more times to obtain a second impact tag value corresponding to the current user. The method of iterative computation is represented by a matrix as follows: for the matrix P2 composed of the second target tag reference degrees and the initial tag matrix F composed of the initial tag vectors, P2 may be multiplied by F to obtain updated F, then P2 is multiplied by the updated F, and the step of multiplying P2 by the updated F is repeated until a convergence condition is satisfied, wherein the convergence condition includes at least one of the repetition number reaching a preset number and the change of the previous updated F and the current updated F being smaller than a preset value. Wherein the change of the F of the previous update and the F of the current update can be represented by the sum of squares of the difference values of the matrix at the same position. Through multiple iterations, useful information can be learned from partial users with tags and users without tags according to the social association degree to determine the tags of the users without tags, that is, when a second influence tag value corresponding to the current user is calculated, the tags corresponding to the first user and the second user both influence the tag value of the current user, and the influence degree of the tag value, that is, the second target tag reference degree, is obtained according to the feature similarity degree.
In one embodiment, each time the matrix F is updated, since the label of the first user with the determined label is known, when the updated matrix F is obtained, the label vector corresponding to the first user in the updated matrix F may be reset to the initial label vector.
Step S306, determining a target tag value corresponding to the current user according to the first influence tag value and the second influence tag value corresponding to the current user.
Specifically, after the first influence tag value and the second influence tag value are obtained, the target tag value corresponding to the current user is determined according to the first influence tag value and the second influence tag value. For example, the target tag value may be obtained according to a sum of the first impact tag value and the second impact tag value, the sum of the first impact tag value and the second impact tag value may be used as the target tag value, or the sum of the first impact tag value and the second impact tag value may be obtained and then added to a preset tag value to obtain the target tag value.
In one embodiment, determining the target tag value corresponding to the current user according to the first impact tag value and the second impact tag value corresponding to the current user includes: acquiring a first weight corresponding to the first influence label value and a second weight corresponding to the second influence label value; and carrying out weighted summation according to the first influence label value and the corresponding first weight, second influence label value and corresponding second weight to obtain a target label value corresponding to the current user.
Specifically, the first weight and the second weight may be set as needed, for example, the first weight may be 0.6, and the second weight may be 0.7. And after a first weight corresponding to the first influence label value and a second weight corresponding to the second influence label value are obtained, carrying out weighted summation according to the first influence label value, the corresponding first weight, the corresponding second influence label value and the corresponding second weight, and obtaining a target label value corresponding to the current user.
In an embodiment, as shown in fig. 4A, the step S304 of calculating a second influence tag value corresponding to the current user according to the second target tag reference degree corresponding to the current user and the user in the user set and the initial tag value corresponding to the user in the user set includes:
and S402, constructing a feature graph by taking each user in the user set as a node and the reference degree of the second target label as the weight of the edge.
Specifically, the graph is composed of nodes and edges between the nodes, and the feature graph is obtained by taking the user as the node and the second target label reference degree as the weight of the edge.
As shown in FIG. 4B, which is a feature diagram in an embodiment, assuming that the user set includes 5 users a 1-a 5, there are 5 nodes in the feature diagram, and the line segments between the 5 nodes represent edges.
Step S404, acquiring a current node corresponding to the current user in the feature map and an adjacent node of the current node.
Specifically, the adjacent nodes are nodes with edge connection, if the second target label reference degree between the nodes is not 0, an edge exists between the nodes, and if the second target label reference degree between the nodes is 0, an edge does not exist between the nodes. As shown in fig. 4B, if the current node is a1, the neighboring nodes of a1 are a2, a4, and a 5.
Step S406, updating the current label value of the current node according to the current label value of the adjacent node in the feature map and the edge weight of the current node and the adjacent node.
Specifically, the tag values corresponding to the nodes in the feature map may be updated through one or more rounds, and the tag values corresponding to the second users are updated in each round of updating. And when the label value of the node in the feature graph is subjected to one round of updating, the updated label value is used as the current label value. And when updating for the first time, taking the initial label value corresponding to the user as the current label value of the corresponding node in the feature map. And the label value of the current node is updated according to the current label value of the adjacent node and the edge weight of the edge between the current node and the corresponding adjacent node, and the current label value after the current node is updated in the characteristic diagram is obtained by performing weighted summation according to the edge weight of the edge between the current node and the adjacent node and the current label value of the corresponding adjacent node. And after the nodes corresponding to the second users in the feature graph are all used as the current nodes, the step S408 is performed.
Step S408, judging whether a convergence condition is met; if not, returning to the step of updating the current label value of the current node according to the current label value of the adjacent node in the feature graph and the edge weight of the current node and the adjacent node, and if so, entering the step S410 to obtain a second influence label value corresponding to the current user.
Specifically, after the label value of the current node corresponding to each second user in the feature graph is updated, the step of updating the current label value of the current node according to the current label value of the adjacent node in the feature graph and the edge weight of the current node and the adjacent node is continuously returned to perform the next round of updating on the label value of the node in the feature graph until a convergence condition is met, and the label value of the node obtained by the last updating is used as a second influence label value, where the convergence condition may include at least one of the fact that the number of times of return reaches a preset number of times and the change of the weight of the edge updated last time in the feature graph and the weight of the edge updated currently is smaller than the preset value. The change in the weights of the previously updated edge and the currently updated edge in the feature map can be represented by the sum of squares of the differences of the edges at the same position.
In an embodiment, the label value of the node corresponding to the first user in the feature map may be updated according to the label value updating method of the current node corresponding to the second user, or may not be updated.
In one embodiment, as shown in fig. 5, calculating the second target tag reference degree between the users according to the feature similarity among the users in the user set comprises:
step S502, the current user is obtained from the second sub-user set, and a target reference user set corresponding to the current user is obtained by screening from the user set according to the social association relationship of the current user.
Specifically, the social association relationship means that a social connection exists between the user and the user, the social system stores a social association chain, and the social association chain records the social association relationship established by the user in the social system and records information of other users having social association with the user. The social relation chains are sequentially arranged according to the user association sequence. And (3) regarding the user as a node, and connecting the nodes with the direct association relationship to obtain a social relationship chain. For example, for a1 and a4, assuming that the friend of a1 is a2, the friend of a2 is a3, and the friend of a3 is a4, the social relationship chain is a1 → a2 → a3 → a 4. The social associations may include direct associations as well as indirect associations. The direct association relationship means that the user is directly associated with the user and is a friend relationship. An indirect relationship refers to a relationship between a user and a user through one or more intermediary users, for example, the user and the user have friends in common. For example, in the social relationship chain a1 → a2 → a3 → a4, a1 is directly related to a2, a2 is related to a3, and a3 is related to a4, and a1 is indirectly related to a3, a1 is related to a4, and a2 is related to a 4. The target reference users in the target reference user set are users which are obtained by screening from the user set and have social association with the current user. The users in the user set who all have social association with the current user may be, or some users who have social association. For example, the target reference users in the target reference user set may be friends within M degrees of the current user, and the social association relationship between the users may be represented by degrees, where the relationship between the degrees and the middle users is equal to the number of the middle users plus 1. For example, if the current user and a certain user are friends, the middle user is 0, and the certain user is a 1-degree friend of the current user. If the current user is a friend of a certain user, the intermediate user is 1, and the certain user is a friend of 2 degrees of the current user.
In one embodiment, the step of screening the target reference user set corresponding to the current user from the user set according to the social association relationship of the current user includes: screening an initial reference user set corresponding to the current user from the user set according to the social association relation of the current user, and calculating the feature similarity between the corresponding current user and the initial reference user according to the current feature corresponding to the current user and the reference feature corresponding to each initial reference user; and screening the target reference users from the initial reference user set according to the feature similarity of the current user and each initial reference user.
Specifically, the current feature refers to a feature of a current user, and the reference feature refers to a feature of an initial reference user. The feature similarity may be calculated by using a distance of the feature, and a calculation method of the distance of the feature may be set as needed, for example, calculated by using a euclidean distance algorithm. The initial reference users with the feature similarity larger than the preset feature similarity and/or with the feature similarity ranking within the preset ranking can be used as the target reference users. The feature similarity ranking is sorted from big to small, and the higher the feature similarity is, the higher the ranking is. For example, friends within M degrees of the current user may be obtained as initial reference users, and then, users with 10 top ranked feature similarity degrees are obtained as target reference users through calculation and screening according to the feature similarity degrees of the current user and the friends within M degrees.
In one embodiment, the step of screening the target reference user set corresponding to the current user from the user set according to the social association relationship of the current user includes: acquiring a first direct association user set corresponding to a current user from a user set according to the social association relation of the current user; screening a first reference user from a first direct correlation user set according to the feature similarity of the current user and the first direct correlation user; acquiring a second direct associated user set corresponding to the first reference user from the user set according to the social association relation of the first reference user; screening a second reference user from a second direct association user set according to the feature similarity of the first reference user and the second direct association user; and taking the first reference user and the second reference user as target reference users in the target reference user set.
Specifically, a first direct associated user in the first direct associated user set is a user having a direct association relationship with a current user, after the first direct associated user set is obtained, feature similarity between the current user and each first direct associated user is calculated, and the first direct associated user whose feature similarity is greater than a preset feature similarity and/or whose feature similarity rank is within a preset rank is used as a first reference user. The feature similarity ranking is sorted from big to small, and the higher the feature similarity is, the higher the ranking is. The preset ranking may be set as desired, for example, 10. After the first reference user is obtained, the direct associated users of the first reference user are obtained to form a second direct associated user set, and the second direct associated users, which have feature similarity greater than the preset feature similarity with the first reference user and/or have feature similarity ranking within the preset ranking, are used as the second reference users, where the preset ranking may be, for example, 8, and the preset feature similarity may be, for example, 0.6. Taking the first reference user and the second reference user as target reference users, it can be understood that a direct associated user of the second reference user may also be continuously obtained, a third reference user is screened according to the feature similarity of the second reference user and the direct associated user thereof, and the third reference user is also taken as the target reference user. Taking an actual example, the 1-degree friend of the current user may be obtained as the first direct association user, then the user whose feature similarity with the current user is ranked 10 in the 1-degree friends is obtained as the first reference user, and then the 1-degree friend of each user in the 10 first reference users is searched. For 1-degree friends of each user in the 10 first reference users, the user with the characteristic similarity ranking 10 before each first reference user is used as a second reference user, a target reference user is obtained according to the algorithm, the calculation complexity can be reduced, the label of the current user can be obtained by referring to the labels of the users with social association relation and similar characteristics to the current user, the obtained target label of the current user is high in accuracy, and the current user can be accurately classified.
Step S504, calculating according to the feature similarity of the current user and each target reference user to obtain a second target label reference degree of the current user and the target reference user.
Specifically, the feature similarity between the current user and the target reference user may be used as the second target tag reference degree, or the feature similarity may be further processed to obtain the second target tag reference degree, for example, the feature similarity is normalized to obtain the second target tag reference degree.
In one embodiment, the feature similarity between the current user and each target reference user can be counted to obtain a feature similarity statistical result; and acquiring the current feature similarity of the current user and the target reference user, and acquiring the second target label reference degree of the current user and the target reference user according to the ratio of the current feature similarity to the feature similarity statistical result. For example, the target reference users of the current user are a2, a3 and a4, and the similarities of the current user and the target reference users are a2, a3 and a4 are 0.8, 0.7 and 0.6, respectively. The similarity statistic is 0.8+0.7+0.6 ≈ 2.1, and the second target label reference between the current user and a2 may be 0.8/2.1 ≈ 0.38.
In one embodiment, the step of constructing the feature graph by using each user in the user set as a node and the second target label reference degree as a weight of an edge includes: and taking each user in the user set as a node, taking a node corresponding to the target reference user as an adjacent node of the current node, and taking the second target label reference degree of the current user and the target reference user as the weight of the edge to construct and obtain the characteristic graph.
Specifically, after the second target label reference degrees of the current user and the target reference user are obtained, when the feature graph is constructed, the node corresponding to the target reference user is used as an adjacent node of the current node, and the second target label reference degrees of the current user and the target reference user are used as the weight of the edge.
In the embodiment of the invention, when the second target label reference degree between the users is calculated, the second target label reference degree between the users is calculated by screening partial users from the user set according to the social association relationship of the users, the feature similarity is calculated by selecting the users with the social association relationship, and the target reference users are further screened by utilizing the feature similarity, so that the calculation amount of the second target label reference degree is reduced, and meanwhile, the labels can be effectively propagated due to the existence of the social association relationship and the strong reference value of the users with the large feature similarity.
It is to be understood that, since the matrix P2 represents the second target label reference degree between the users, the weight of the edge in the feature graph, i.e. the second target label reference degree, can be represented by the matrix P2, where the jth vector value in the ith row in the matrix P2 represents the weight of the edge between the i node and the j node, if the vector value is not 0, it represents that the i node and the j node in the feature graph are neighboring nodes, and if the vector value is 0, it represents that the i node and the j node in the feature graph are not neighboring nodes. The update process for the label values of the nodes in the feature graph can be represented by the following steps:
step 11: and (3) carrying out label propagation: fFurthermore, the utility model=P2*FWhen in use(ii) a I.e. according to the matrix P2 and the current label matrix FWhen in useCalculating to obtain an updated label matrix FFurthermore, the utility modelAnd when the label matrix is calculated for the first time, the current label matrix is the initial label matrix.
Step 12: will update the tag matrix FFurthermore, the utility modelThe label vector corresponding to the first user is reduced to the initial label vector to obtain the current label matrix FWhen in useAnd returning to the step 11 until fullThe convergence condition is preset.
In an embodiment, as shown in fig. 6A, for the current user in the second sub-user set, the step S304 of obtaining the first influence tag value corresponding to the current user according to the first target tag reference degree corresponding to the current user and the user in the user set and the initial tag value corresponding to the user in the user set may specifically include the following steps:
step S602, taking each user in the user set as a node, and taking the first target label reference degree as the edge weight of the edge, so as to obtain the social association graph.
Specifically, the graph is composed of nodes and edges between the nodes, and the social association graph is obtained by taking the user as the node and the first target tag reference degree as the weight of the edge.
As shown in FIG. 6B, which is a social association graph in one embodiment, assuming that the user set includes 5 users a 1-a 5, there are 5 nodes in the social association graph, and the line segments between the 5 nodes represent edges.
Step S604, a current node corresponding to the current user in the social association graph and an adjacent node of the current node are obtained.
Specifically, the adjacent nodes are nodes with edge connection, if the first target label reference degree between the nodes is not 0, an edge exists between the nodes, and if the first target label reference degree between the nodes is 0, an edge does not exist between the nodes. As shown in fig. 6B, if the current node is a1, the neighboring nodes of a1 are a3 and a 5. The node corresponding to the user having the direct association relation with the current user may be taken as the neighboring node of the current node.
Step S606, updating the current label value of the current node according to the current label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node.
Specifically, the tag values corresponding to the nodes in the social association graph may be updated through one or more rounds, and the tag values corresponding to the second users may be updated in each round of updating. And when the label value of the node in the social association graph is subjected to a round of updating, the updated label value is used as the current label value. And when the social association graph is updated for the first time, taking the initial tag value corresponding to the user as the current tag value of the corresponding node in the social association graph. And the label value of the current node is updated according to the label values of the adjacent nodes and the edge weights of the edges between the current node and the corresponding adjacent nodes, and the label value of the current node after updating in the social association diagram is obtained by performing weighted summation according to the edge weights of the edges between the current node and the adjacent nodes and the label values of the corresponding adjacent nodes. After the nodes corresponding to the second users in the social association graph are all updated as the current nodes, the step S608 is performed.
Step S608, judging whether a convergence condition is met, if not, returning to the step of updating the current label value of the current node according to the current label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node, if so, entering step S610, and obtaining a first influence label value corresponding to the current user.
Specifically, after the tag value of the current node corresponding to each second user in the social association graph is updated, the step of updating the current tag value of the current node according to the current tag value of the adjacent node in the social association graph and the edge weights of the current node and the adjacent node is continuously returned to perform next round of updating on the tag value of the node in the social association graph until a convergence condition is met, and the tag value of the node obtained by the last updating is used as a first influence tag value, where the convergence condition may include at least one of the fact that the number of times of returning reaches a preset number of times and the change of the weight of the edge updated last time in the social association graph and the weight of the edge updated currently is smaller than the preset value. The change in the weight of the previously updated edge and the weight of the currently updated edge in the social relevance graph may be represented by the sum of the squares of the differences of the co-located edges.
In an embodiment, the tag value of the node corresponding to the first user in the social association graph may be updated according to the tag value updating method of the current node corresponding to the second user, or may not be updated.
It is to be understood that since the matrix P1 represents the first target tag reference degree between the users, the first target tag reference degree, which is the weight of the edge in the social relevance graph, can be represented by the matrix P1, where the jth vector value in the ith row in the matrix P1 represents the weight of the edge between the i node and the j node, if the vector value is not 0, it represents that the i node and the j node in the social relevance graph are neighboring nodes, and if the vector value is 0, it represents that the i node and the j node in the social relevance graph are not neighboring nodes. The update process for the tag values of the nodes in the social association graph can be represented by the following steps:
step 21: and (3) carrying out label propagation: fFurthermore, the utility model=P1*FWhen in use(ii) a I.e. according to the matrix P1 and the current label matrix FWhen in useCalculating to obtain an updated label matrix FFurthermore, the utility modelAnd when the label matrix is calculated for the first time, the current label matrix is the initial label matrix.
Step 22: will update the tag matrix FFurthermore, the utility modelThe label vector corresponding to the first user is reduced to the initial label vector to obtain the current label matrix FWhen in useAnd returning to the step 21 until the preset convergence condition is met.
The following describes a user tag determination method provided in an embodiment of the present invention with reference to fig. 7A in a specific embodiment:
step S702, an input user set and a relationship chain between users are obtained, wherein the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels.
The user set may be all users in the social application, or may be a set composed of users whose social association degree is greater than a preset association degree.
Step S704, determining an initial label value corresponding to each user according to the label corresponding to each user in the user set;
step S706, obtaining users within M-degree friends in each user in the user set, and forming an initial reference user set corresponding to each user;
where M may be set as desired, for example, 2.
Step 708, acquiring initial reference users with the first W feature similarities with the current user in the initial reference user set as target reference users;
step S710, calculating according to the feature similarity of the current user and each target reference user to obtain a second target label reference degree of the current user and the target reference user;
step 712, using each user in the user set as a node, using the target reference user as an adjacent node of the current node, and using the second target label reference degree as the weight of the edge between the corresponding nodes to obtain a feature graph;
step S714, repeating the step of updating the label value of the current node once or for many times according to the label values of the adjacent nodes in the characteristic diagram and the weight of the corresponding edge until the convergence condition is met to obtain a second influence label value;
step S716, calculating a first target tag reference degree of the current user and each user according to the social relevance degree of the current user and each user in the user set;
step S718, using each user in the user set as a node, using the first target label reference degree as the weight of an edge between corresponding nodes, and obtaining a social association graph;
step S720, repeating the step of updating the current label value of the current node once or for many times according to the current label value of the adjacent node in the social association graph and the weight of the corresponding edge until a convergence condition is met to obtain a first influence degree label value;
step S722, obtaining a target tag value according to the first influence degree tag value and the second influence degree tag value;
step S724, determining a target label corresponding to the current user according to the target label value.
In an embodiment, as shown in fig. 7B, when a user tag needs to be determined, the input data includes a user identifier, a user characteristic, a social relationship chain, and a tag corresponding to a part of the user. The server may construct an initial label matrix F based on the user labels, search for a friend set of each user within M degrees based on the social relationship chain, obtain K users with the most similar characteristics to the user from the friend set of each user within M degrees, construct a feature map based on the feature similarity after finding K associated users with similar characteristics corresponding to each user, where the K associated users with similar characteristics corresponding to each user are adjacent nodes of the user, and then perform label propagation based on the matrix P2 and the feature map, where the number of label propagation times may be multiple times until F converges. In addition, a social association graph is also constructed based on the social relationship chain, users with direct association are adjacent to each other, and then label propagation can be performed based on the matrix P2 and the social association graph, wherein the label propagation times can be multiple times until F converges. And then, the results propagated by the two labels can be fused to obtain and output the target label. During fusion, the two target label matrixes obtained by final propagation can be added.
As shown in fig. 8, in an embodiment, a user tag determination apparatus is provided, which may be integrated in the server 120 described above, and specifically may include a user set obtaining module 802, an initial tag value determining module 804, a target tag referential degree calculating module 806, a target tag value calculating module 808, and a target tag determination module 810.
A user set obtaining module 802, configured to obtain a user set, where the user set includes a first sub-user set with determined tags and a second sub-user set with undetermined tags;
an initial tag value determining module 804, configured to determine an initial tag value corresponding to each user according to a tag corresponding to each user in the user set;
a target tag reference degree calculating module 806, configured to calculate a target tag reference degree between users according to the social association degree and the feature similarity among the users in the user set;
a target tag value calculating module 808, configured to calculate, for a current user in the second sub-user set, a target tag value corresponding to the current user according to a target tag reference degree corresponding to each user in the current user and each user in the user set and an initial tag value corresponding to each user in the user set;
and the target tag determining module 810 is configured to determine a target tag corresponding to the current user according to the target tag value.
In one embodiment, the target tag reference degree includes a first target tag reference degree and a second target tag reference degree, and the target tag reference degree calculating module 806 is configured to: calculating to obtain a first target tag reference degree between users according to the social association degree between the users in the user set, and calculating to obtain a second target tag reference degree between the users according to the feature similarity degree between the users in the user set;
as shown in fig. 9, the target tag value calculation module 808 includes:
a first influence tag value calculating unit 808A, configured to, for a current user in the second sub-user set, obtain a first influence tag value corresponding to the current user according to a first target tag reference degree corresponding to the current user and a user in the user set and an initial tag value corresponding to the user in the user set;
a second influence tag value calculating unit 808B, configured to, for a current user in the second sub-user set, obtain a second influence tag value corresponding to the current user according to a second target tag reference degree corresponding to the current user and a user in the user set and an initial tag value corresponding to the user in the user set;
and the target label value calculating unit 808C is configured to determine a target label value corresponding to the current user according to the first impact label value and the second impact label value corresponding to the current user.
In one embodiment, the second influence tag value calculation unit 808 includes:
the feature graph constructing subunit is used for constructing a feature graph by taking each user in the user set as a node and taking the second target label reference degree as the edge weight of the edge;
the node acquisition subunit is used for acquiring a current node corresponding to a current user in the characteristic diagram and adjacent nodes of the current node;
the updating subunit is used for updating the label value of the current node according to the label values of the adjacent nodes in the characteristic diagram and the edge weights of the current node and the adjacent nodes;
and the returning subunit is used for returning to the step of updating the label value of the current node according to the label values of the adjacent nodes in the characteristic diagram and the edge weights of the current node and the adjacent nodes until a convergence condition is met, and obtaining a second influence label value corresponding to the current user.
In one embodiment, the target tag reference calculation module 806 includes:
the target reference user set acquisition unit is used for acquiring the current user from the second sub-user set and screening the target reference user set corresponding to the current user from the user set according to the social association relationship of the current user;
the second target label reference degree calculating unit is used for calculating and obtaining a second target label reference degree of the current user and the target reference user according to the feature similarity of the current user and each target reference user;
the feature map construction subunit is configured to:
and taking each user in the user set as a node, taking a node corresponding to the target reference user as an adjacent node of the current node, and taking the second target label reference degree of the current user and the target reference user as the weight of the edge to construct and obtain the characteristic graph.
In one embodiment, the target reference user set obtaining unit is configured to:
acquiring a first direct association user set corresponding to a current user from a user set according to the social association relation of the current user;
screening a first reference user from a first direct correlation user set according to the feature similarity of the current user and the first direct correlation user;
acquiring a second direct associated user set corresponding to the first reference user from the user set according to the social association relation of the first reference user;
screening a second reference user from a second direct association user set according to the feature similarity of the first reference user and the second direct association user;
and taking the first reference user and the second reference user as target reference users in the target reference user set.
In one embodiment, the second target tag reference calculation unit is configured to:
counting the feature similarity of the current user and each target reference user to obtain a feature similarity counting result;
and acquiring the current feature similarity of the current user and the target reference user, and acquiring the second target label reference degree of the current user and the target reference user according to the ratio of the current feature similarity to the feature similarity statistical result.
In one embodiment, the first influencing tag value calculation unit 808A is to:
taking each user in the user set as a node, and taking the first target label reference degree as the edge weight of an edge to obtain a social association graph;
acquiring a current node corresponding to a current user in a social association graph and an adjacent node of the current node;
updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node;
and returning to the step of updating the label value of the current node according to the label values of the adjacent nodes in the social association graph and the edge weights of the current node and the adjacent nodes until the convergence condition is met, and obtaining a first influence label value corresponding to the current user.
In one embodiment, the target tag value calculation unit 808C is to:
acquiring a first weight corresponding to the first influence label value and a second weight corresponding to the second influence label value;
and carrying out weighted summation according to the first influence label value and the corresponding first weight, second influence label value and corresponding second weight to obtain a target label value corresponding to the second user.
In one embodiment, the initial tag value determination module 804 includes:
the target label type obtaining unit is used for obtaining target label types corresponding to the first users;
a first initial tag vector obtaining unit, configured to obtain an initial tag vector corresponding to each first user according to a target tag category corresponding to each first user;
and the second initial label vector obtaining unit is used for obtaining the preset label vector as the initial label vector corresponding to the second user in the second sub-user set, and the values of the preset label vector are consistent.
In one embodiment, the first initial tag vector deriving unit is configured to:
for each target label category corresponding to the first user, acquiring a corresponding first preset value as a corresponding vector value, and for each non-target label category corresponding to the first user, acquiring a corresponding second preset value as a corresponding vector value, wherein the first preset value is different from the second preset value;
and combining the vector value corresponding to each target label category and the vector value corresponding to each non-target label category into an initial label vector corresponding to the first user.
FIG. 10 is a diagram illustrating an internal structure of a computer device in one embodiment. The computer device may specifically be the server 120 in fig. 1. As shown in fig. 10, the computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program that, when executed by the processor, causes the processor to implement the user tag determination method. The internal memory may also have stored therein a computer program that, when executed by the processor, causes the processor to perform a user tag determination method.
Those skilled in the art will appreciate that the architecture shown in fig. 10 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, the user tag determination apparatus provided herein may be implemented in the form of a computer program that is executable on a computer device such as that shown in fig. 10. The memory of the computer device may store various program modules constituting the user tag determination apparatus, such as the user set acquisition module 802, the initial tag value determination module 804, the target tag referential calculation module 806, the target tag value calculation module 808 and the target tag determination module 810 shown in fig. 8. The program modules constitute computer programs that cause a processor to execute the steps in the user tag determination methods of the embodiments of the present application described in the present specification.
For example, the computer device shown in fig. 10 may obtain, by the user set obtaining module 802 in the user tag determining apparatus shown in fig. 8, a user set, where the user set includes a first sub-user set with determined tags and a second sub-user set with undetermined tags; determining an initial tag value corresponding to each user according to a tag corresponding to each user in the user set by an initial tag value determining module 804; calculating target label reference degrees among the users according to the social association degrees and the feature similarity degrees among the users in the user set through a target label reference degree calculating module 806; calculating a target label value corresponding to the current user according to the target label reference degree corresponding to the current user and each user in the user set and the initial label value corresponding to each user in the user set for the current user in the second sub-user set through a target label value calculating module 808; and determining a target tag corresponding to the current user according to the target tag value through the target tag determination module 810.
In an embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the above-described user tag determination method. Here, the steps of the user tag determination method may be steps in the user tag determination methods of the above embodiments.
In an embodiment, a computer readable storage medium is provided, storing a computer program which, when executed by a processor, causes the processor to perform the steps of the above-described user tag determination method. Here, the steps of the user tag determination method may be steps in the user tag determination methods of the above embodiments.
It should be understood that, although the steps in the flowcharts of the embodiments of the present invention are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in various embodiments may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware related to instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, the computer program can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (22)

1. A method of user tag determination, the method comprising:
acquiring a user set, wherein the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels;
determining an initial label value corresponding to each user according to the label corresponding to each user in the user set;
calculating to obtain a first target tag reference degree between users according to the social association degree between the users in the user set, and calculating to obtain a second target tag reference degree between the users according to the feature similarity degree between the users in the user set;
for the current user in the second sub-user set, calculating to obtain a first influence label value corresponding to the current user according to a first target label reference degree corresponding to the current user and the users in the user set and an initial label value corresponding to the users in the user set;
for the current user in the second sub-user set, calculating to obtain a second influence label value corresponding to the current user according to a second target label reference degree corresponding to the current user and the users in the user set and an initial label value corresponding to the users in the user set;
determining a target tag value corresponding to the current user according to a first influence tag value and a second influence tag value corresponding to the current user;
and determining a target label corresponding to the current user according to the target label value.
2. The method of claim 1, wherein the target tag reference is positively correlated with the social relevance and the target tag reference is positively correlated with the feature similarity.
3. The method of claim 1, wherein the calculating a second impact tag value corresponding to the current user according to a second target tag reference corresponding to the current user and a user in the user set and an initial tag value corresponding to a user in the user set comprises:
taking each user in the user set as a node, and taking the second target label reference degree as the edge weight of an edge to construct a feature graph;
acquiring a current node corresponding to the current user and an adjacent node of the current node in the feature map;
taking the initial label value corresponding to each user as the current label value of the corresponding node in the feature map, and updating the current label value of the current node according to the current label value of the adjacent node in the feature map and the edge weight of the current node and the adjacent node;
and returning to the step of updating the current label value of the current node according to the current label value of the adjacent node in the feature graph and the edge weight of the current node and the adjacent node until a convergence condition is met, and obtaining a second influence label value corresponding to the current user.
4. The method of claim 3, wherein the calculating a second target tag reference between users according to the feature similarity between users in the user set comprises:
obtaining a current user from the second sub-user set, and screening a target reference user set corresponding to the current user from the user set according to the social association relationship of the current user;
calculating to obtain a second target label reference degree of the current user and each target reference user according to the feature similarity of the current user and each target reference user;
the constructing the feature graph by using each user in the user set as a node and the second target label reference degree as a weight of an edge includes:
and taking each user in the user set as a node, taking a node corresponding to the target reference user as an adjacent node of the current node, and taking the second target label reference degree of the current user and the target reference user as the weight of the edge to construct and obtain a feature graph.
5. The method according to claim 4, wherein the obtaining of the target reference user set corresponding to the current user from the user set by filtering according to the social association relationship of the current user comprises:
acquiring a first direct association user set corresponding to the current user from the user set according to the social association relation of the current user;
screening a first reference user from the first direct association user set according to the feature similarity of the current user and a first direct association user;
acquiring a second direct association user set corresponding to the first reference user from the user set according to the social association relationship of the first reference user;
screening a second reference user from the second direct association user set according to the feature similarity of the first reference user and the second direct association user;
and taking the first reference user and the second reference user as target reference users in the target reference user set.
6. The method of claim 4, wherein the calculating a second target tag reference degree of the current user and the target reference user according to the feature similarity of the current user and each target reference user comprises:
counting the feature similarity of the current user and each target reference user to obtain a feature similarity counting result;
and acquiring the current feature similarity of the current user and the target reference user, and acquiring a second target label reference degree of the current user and the target reference user according to the ratio of the current feature similarity to the feature similarity statistical result.
7. The method of claim 1, wherein the calculating a first impact tag value corresponding to the current user according to a first target tag reference degree corresponding to the current user and a user in the user set and an initial tag value corresponding to the user in the user set comprises:
taking each user in the user set as a node, and taking the first target label reference degree as the edge weight of an edge to obtain a social association graph;
acquiring a current node corresponding to the current user and an adjacent node of the current node in the social association graph;
updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node;
and returning to the step of updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node until a convergence condition is met, and obtaining a first influence label value corresponding to the current user.
8. The method of claim 1, wherein determining the target tag value corresponding to the current user according to the first impact tag value and the second impact tag value corresponding to the current user comprises:
acquiring a first weight corresponding to the first influence label value and a second weight corresponding to the second influence label value;
and carrying out weighted summation according to the first influence label value and the corresponding first weight, the second influence label value and the corresponding second weight to obtain a target label value corresponding to the current user.
9. The method of claim 1, wherein the determining an initial tag value corresponding to each user according to the tag corresponding to each user in the user set comprises:
acquiring a target label category corresponding to each first user in the first sub-user set;
obtaining an initial label vector corresponding to each first user according to the target label category corresponding to the first user;
and acquiring a preset label vector as an initial label vector corresponding to a second user in the second sub-user set, wherein the values of the preset label vector are consistent.
10. The method of claim 9, wherein obtaining the initial tag vector corresponding to the first user according to the target tag category corresponding to each of the first users comprises:
for each target label category corresponding to the first user, acquiring a corresponding first preset value as a corresponding vector value, and for each non-target label category corresponding to the first user, acquiring a corresponding second preset value as a corresponding vector value, wherein the first preset value is different from the second preset value;
and combining the vector value corresponding to each target label category and the vector value corresponding to each non-target label category into an initial label vector corresponding to the first user.
11. A user tag determination apparatus, the apparatus comprising:
the system comprises a user set acquisition module, a label matching module and a label matching module, wherein the user set acquisition module is used for acquiring a user set, and the user set comprises a first sub-user set with determined labels and a second sub-user set with undetermined labels;
an initial tag value determining module, configured to determine, according to a tag corresponding to each user in the user set, an initial tag value corresponding to each user;
the target tag reference degree calculation module is used for calculating a first target tag reference degree between users according to social association degrees between the users in the user set and calculating a second target tag reference degree between the users according to feature similarity degrees between the users in the user set;
a target label value calculation module, configured to calculate, for a current user in the second sub-user set, a first influence label value corresponding to the current user according to a first target label reference degree corresponding to the current user and users in the user set and an initial label value corresponding to the users in the user set; for the current user in the second sub-user set, calculating to obtain a second influence label value corresponding to the current user according to a second target label reference degree corresponding to the current user and the users in the user set and an initial label value corresponding to the users in the user set; determining a target tag value corresponding to the current user according to a first influence tag value and a second influence tag value corresponding to the current user;
and the target label determining module is used for determining a target label corresponding to the current user according to the target label value.
12. The apparatus of claim 11, wherein the target tag reference is positively correlated with the social relevance and the target tag reference is positively correlated with the feature similarity.
13. The apparatus according to claim 11, wherein the second influence label value calculation unit includes:
a feature graph constructing subunit, configured to construct a feature graph by using each user in the user set as a node and using the second target label reference degree as an edge weight of an edge;
a node obtaining subunit, configured to obtain a current node corresponding to the current user and a neighboring node of the current node in the feature map;
an updating subunit, configured to use an initial tag value corresponding to each user as a current tag value of a node corresponding to the feature map, and update the current tag value of the current node according to the current tag value of the neighboring node in the feature map and the edge weight between the current node and the neighboring node;
and the returning subunit is configured to return to the step of updating the current label value of the current node according to the current label value of the adjacent node in the feature map and the edge weights of the current node and the adjacent node until a convergence condition is met, so as to obtain a second influence label value corresponding to the current user.
14. The apparatus of claim 13, wherein the target tag reference calculation module comprises:
a target reference user set obtaining unit, configured to obtain a current user from the second sub-user set, and filter from the user set according to a social association relationship of the current user to obtain a target reference user set corresponding to the current user;
the second target tag reference degree calculating unit is used for calculating to obtain second target tag reference degrees of the current user and the target reference users according to the feature similarity of the current user and each target reference user;
the feature map construction subunit is configured to: and taking each user in the user set as a node, taking a node corresponding to the target reference user as an adjacent node of the current node, and taking the second target label reference degree of the current user and the target reference user as the weight of the edge to construct and obtain a feature graph.
15. The apparatus of claim 14, wherein the target reference user set obtaining unit is configured to: acquiring a first direct association user set corresponding to the current user from the user set according to the social association relation of the current user; screening a first reference user from the first direct association user set according to the feature similarity of the current user and a first direct association user; acquiring a second direct association user set corresponding to the first reference user from the user set according to the social association relationship of the first reference user; screening a second reference user from the second direct association user set according to the feature similarity of the first reference user and the second direct association user; and taking the first reference user and the second reference user as target reference users in the target reference user set.
16. The apparatus of claim 14, wherein the second target tag reference calculation unit is configured to: counting the feature similarity of the current user and each target reference user to obtain a feature similarity counting result; and acquiring the current feature similarity of the current user and the target reference user, and acquiring a second target label reference degree of the current user and the target reference user according to the ratio of the current feature similarity to the feature similarity statistical result.
17. The apparatus of claim 11, wherein the first influencing tag value calculating unit is configured to: taking each user in the user set as a node, and taking the first target label reference degree as the edge weight of an edge to obtain a social association graph; acquiring a current node corresponding to the current user and an adjacent node of the current node in the social association graph; updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node; and returning to the step of updating the label value of the current node according to the label value of the adjacent node in the social association graph and the edge weight of the current node and the adjacent node until a convergence condition is met, and obtaining a first influence label value corresponding to the current user.
18. The apparatus of claim 11, wherein the target tag value computing unit is configured to: acquiring a first weight corresponding to the first influence label value and a second weight corresponding to the second influence label value; and carrying out weighted summation according to the first influence label value and the corresponding first weight, the second influence label value and the corresponding second weight to obtain a target label value corresponding to the current user.
19. The apparatus of claim 11, wherein the initial tag value determining module comprises: a target label category obtaining unit, configured to obtain a target label category corresponding to each first user in the first sub-user set; a first initial tag vector obtaining unit, configured to obtain an initial tag vector corresponding to each first user according to a target tag category corresponding to the first user; and a second initial tag vector obtaining unit, configured to obtain a preset tag vector as an initial tag vector corresponding to a second user in the second sub-user set, where values of the preset tag vector are consistent.
20. The apparatus of claim 19, wherein the first initial tag vector deriving unit is configured to: for each target label category corresponding to the first user, acquiring a corresponding first preset value as a corresponding vector value, and for each non-target label category corresponding to the first user, acquiring a corresponding second preset value as a corresponding vector value; and combining the vector value corresponding to each target label category and the vector value corresponding to each non-target label category into an initial label vector corresponding to the first user, wherein the first preset value is different from the second preset value.
21. A computer arrangement, comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to carry out the steps of the user tag determination method of any one of claims 1 to 10.
22. A computer-readable storage medium, having stored thereon a computer program which, when being executed by a processor, causes the processor to carry out the steps of the user tag determination method according to any one of claims 1 to 10.
CN201811503599.5A 2018-12-10 2018-12-10 User label determination method and device, computer equipment and storage medium Active CN110162692B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811503599.5A CN110162692B (en) 2018-12-10 2018-12-10 User label determination method and device, computer equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811503599.5A CN110162692B (en) 2018-12-10 2018-12-10 User label determination method and device, computer equipment and storage medium

Publications (2)

Publication Number Publication Date
CN110162692A CN110162692A (en) 2019-08-23
CN110162692B true CN110162692B (en) 2021-05-25

Family

ID=67645265

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811503599.5A Active CN110162692B (en) 2018-12-10 2018-12-10 User label determination method and device, computer equipment and storage medium

Country Status (1)

Country Link
CN (1) CN110162692B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112685626A (en) * 2019-10-17 2021-04-20 北京星选科技有限公司 Method and device for determining object type, electronic equipment and storage medium
CN111275492A (en) 2020-02-07 2020-06-12 腾讯科技(深圳)有限公司 User portrait generation method, device, storage medium and equipment
CN111339425B (en) * 2020-03-05 2021-07-23 拉扎斯网络科技(上海)有限公司 Object marking method, device, server and storage medium
CN111552799B (en) * 2020-04-30 2021-07-09 腾讯科技(深圳)有限公司 Information processing method, information processing device, electronic equipment and storage medium
CN113159927A (en) * 2021-04-30 2021-07-23 中国银行股份有限公司 Method and device for determining client label
CN114596097B (en) * 2022-05-10 2022-08-26 富算科技(上海)有限公司 User identification method, device, electronic equipment and computer readable storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105069129A (en) * 2015-06-24 2015-11-18 合肥工业大学 Self-adaptive multi-label prediction method
CN107368534A (en) * 2017-06-21 2017-11-21 南京邮电大学 A kind of method for predicting social network user attribute
CN107368499A (en) * 2016-05-12 2017-11-21 中国移动通信集团广东有限公司 A kind of client's tag modeling and recommendation method and device
WO2018077039A1 (en) * 2016-10-27 2018-05-03 腾讯科技(深圳)有限公司 Community discovery method, apparatus, server, and computer storage medium
CN108307318A (en) * 2017-12-21 2018-07-20 北京三快在线科技有限公司 A kind of user information acquiring method and device, electronic equipment

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107220852A (en) * 2017-05-26 2017-09-29 北京小度信息科技有限公司 Method, device and server for determining target recommended user
CN108921189B (en) * 2018-05-23 2021-05-18 北京航空航天大学 Deduction method and device for social network user attributes

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105069129A (en) * 2015-06-24 2015-11-18 合肥工业大学 Self-adaptive multi-label prediction method
CN107368499A (en) * 2016-05-12 2017-11-21 中国移动通信集团广东有限公司 A kind of client's tag modeling and recommendation method and device
WO2018077039A1 (en) * 2016-10-27 2018-05-03 腾讯科技(深圳)有限公司 Community discovery method, apparatus, server, and computer storage medium
CN107368534A (en) * 2017-06-21 2017-11-21 南京邮电大学 A kind of method for predicting social network user attribute
CN108307318A (en) * 2017-12-21 2018-07-20 北京三快在线科技有限公司 A kind of user information acquiring method and device, electronic equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于主题模型的社交网络用户画像分析方法;马超;《中国优秀硕士学位论文全文数据库 信息科技辑》;20171115(第11期);第I139-24页 *

Also Published As

Publication number Publication date
CN110162692A (en) 2019-08-23

Similar Documents

Publication Publication Date Title
CN110162692B (en) User label determination method and device, computer equipment and storage medium
CN110569437B (en) Click probability prediction and page content recommendation methods and devices
US10846052B2 (en) Community discovery method, device, server and computer storage medium
US10936765B2 (en) Graph centrality calculation method and apparatus, and storage medium
CN107895038B (en) Link prediction relation recommendation method and device
CN109919316A (en) The method, apparatus and equipment and storage medium of acquisition network representation study vector
CN109460793A (en) A kind of method of node-classification, the method and device of model training
CN108304935B (en) Machine learning model training method and device and computer equipment
CN112861967A (en) Social network abnormal user detection method and device based on heterogeneous graph neural network
CN111177473B (en) Personnel relationship analysis method, device and readable storage medium
CN108491511A (en) Data digging method and device, model training method based on diagram data and device
CN112989169B (en) Target object identification method, information recommendation method, device, equipment and medium
WO2019019385A1 (en) Cross-platform data matching method and apparatus, computer device and storage medium
CN110659394B (en) Recommendation method based on bidirectional proximity
CN113656699B (en) User feature vector determining method, related equipment and medium
Liu et al. Evolving graph construction for successive recommendation in event-based social networks
CN113689285A (en) Method, device, equipment and storage medium for detecting user characteristics
CN104484365B (en) In a kind of multi-source heterogeneous online community network between network principal social relationships Forecasting Methodology and system
CN114493674A (en) Advertisement click rate prediction model and method
CN113468403A (en) User information prediction method based on big data mining and cloud computing AI (Artificial Intelligence) service system
CN112559640A (en) Training method and device of atlas characterization system
CN115758271A (en) Data processing method, data processing device, computer equipment and storage medium
KR102323424B1 (en) Rating Prediction Method for Recommendation Algorithm Based on Observed Ratings and Similarity Graphs
CN115329214A (en) User recommendation method and device, electronic equipment and storage medium
CN112085040B (en) Object tag determining method and device and computer equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant