CN106547921B - Label generation method and device - Google Patents

Abstract

The disclosure provides a kind of label generating method and device, for providing a kind of mode for preferably generating label for object.The described method includes: determining the access information that known object and unknown object are accessed by the user in preset duration, wherein the known object is the object to have labelled, and the unknown object is the object not labelled, and the label is used for classified object；According to the label of the known object and the access information, the label of the unknown object is determined.

Description

Label generation method and device

technical field

The present disclosure relates to the field of computer technology, and in particular, to a label generation method and device.

Background technique

In the contemporary society with a huge amount of information, objects can be described and classified through tags. For example, if the object can be news, then you can label the news to match its content, such as science and technology, people's livelihood, economy, military, and so on. Tags play an important role in information recommendation, and can also help us to recognize information resources more effectively, which is conducive to the discovery, management, dissemination and utilization of information.

At present, there are usually two methods for labeling objects: the first one is to manually label objects, which is difficult to deal with a large number of objects, and the quality of labels fluctuates due to human factors. The second is to analyze the content of the object itself (such as the content of the article, the detailed information of the product, etc.), and use data mining and machine learning algorithms to automatically generate tags from the text information. However, in practical applications, some objects often lack descriptive information. Or explanatory text, which may not be able to generate corresponding tags by means of data mining and machine learning algorithms.

It can be seen that there is currently no better way to label objects.

Contents of the invention

The purpose of the present disclosure is to provide a tag generation method and device for providing a better way of generating tags for objects.

According to a first aspect of an embodiment of the present disclosure, a tag generation method is provided, including:

Determine the access information of known objects and unknown objects accessed by users within a preset time period, wherein the known objects are tagged objects, the unknown objects are untagged objects, and the tags are used for attribution class object;

Determine the label of the unknown object according to the label of the known object and the access information.

Optionally, the known object includes multiple objects, the unknown object includes at least one object, and the user includes multiple users;

After determining the access information of known objects and unknown objects accessed by users within a preset time period, it also includes:

generating a behavior matrix A of the known object, wherein the behavior matrix A is used to indicate the access information that the plurality of users respectively visit each object included in the known object; and,

generating a behavior matrix B of the unknown object, wherein the behavior matrix B is used to indicate the access information that the plurality of users respectively access to each object included in the unknown object;

Determining the label of the unknown object according to the label of the known object and the access information includes:

According to the label matrix M of the known object, the behavior matrix A, and the behavior matrix B, generate the label matrix N of the unknown object to determine the label of the unknown object, wherein the label matrix is used to indicate Objects correspond to weights of different labels respectively.

Optionally, the first user is any user among the multiple users, and the first object is any object included in the known object or the unknown object;

Determining the access information of the first object being accessed by the first user within the preset time period includes:

Determining access information that the first object is accessed by the first user according to the number of times the first user accesses the first object within the preset time period and/or the duration of accessing the first object.

Optionally, generating the label matrix N of the unknown object according to the label matrix M of the known object, the behavior matrix A, and the behavior matrix B includes:

multiplying the label matrix M by the behavior matrix A to obtain the user label matrix W of the known object;

Multiplying the user label matrix W by the transpose matrix of the behavior matrix B to obtain the label matrix N of the unknown object.

Optionally, generating the label matrix N of the unknown object according to the label matrix M of the known object, the behavior matrix A, and the behavior matrix B includes:

According to the label matrix M of the known object, the behavior matrix A, the behavior matrix B, and the penalty function F(C), generate the label matrix N of the unknown object, wherein, C is the number of times the object is accessed, and n is the preset times threshold.

According to a second aspect of an embodiment of the present disclosure, a label generation device is provided, including:

The first determination module is used to determine the access information of known objects and unknown objects accessed by users within a preset time period, wherein the known objects are tagged objects, and the unknown objects are untagged objects , the label is used to classify objects;

The second determining module is configured to determine the label of the unknown object according to the label of the known object and the access information.

Optionally, the known object includes multiple objects, the unknown object includes at least one object, and the user includes multiple users;

The device also includes:

The first generation module is configured to generate the behavior matrix A of the known objects after determining the access information of the known objects and the unknown objects accessed by the user within a preset time period, wherein the behavior matrix A is used to indicate the The access information that the plurality of users respectively access to each object included in the known object; and,

generating a behavior matrix B of the unknown object, wherein the behavior matrix B is used to indicate the access information that the plurality of users respectively access to each object included in the unknown object;

The second generation module is used to generate the label matrix N of the unknown object according to the label matrix M of the known object, the behavior matrix A, and the behavior matrix B, so as to determine the label of the unknown object, Among them, the label matrix is used to indicate the weights of objects respectively corresponding to different labels.

Optionally, the first user is any user among the multiple users, and the first object is any object included in the known object or the unknown object;

The first determination module is used for:

Determining access information that the first object is accessed by the first user according to the number of times the first user accesses the first object within the preset time period and/or the duration of accessing the first object.

Optionally, the second generating module is used for:

multiplying the label matrix M by the behavior matrix A to obtain the user label matrix W of the known object;

Multiplying the user label matrix W by the transpose matrix of the behavior matrix B to obtain the label matrix N of the unknown object.

Optionally, the second generating module is used for:

According to the label matrix M of the known object, the behavior matrix A, the behavior matrix B, and the penalty function F(C), generate the label matrix N of the unknown object, wherein, C is the number of times the object is accessed, and n is the preset times threshold.

According to the third aspect of the embodiments of the present disclosure, there is provided a non-transitory computer-readable storage medium, when the instructions in the storage medium are executed by the processor of the computer, the computer equipment can execute a label generation method, so The methods described include:

Determine the access information of known objects and unknown objects accessed by users within a preset time period, wherein the known objects are tagged objects, the unknown objects are untagged objects, and the tags are used for attribution class object;

Determine the label of the unknown object according to the label of the known object and the access information.

Through the above technical solution, the known objects that have been tagged can be used to determine the tag of the unknown object through the user's access behavior to the known object and the unknown object that has not been tagged. For the same user, it can be considered that preferences are relatively invariant, and frequently visited objects are usually associated. Therefore, the method of determining the label of an unknown object by counting the user's access behavior to known objects and unknown objects , can accurately predict the labels of unknown objects without manual labeling, and for unknown objects that lack descriptive information, labels can also be generated more accurately through the user's access behavior.

Other features and advantages of the present disclosure will be described in detail in the detailed description that follows.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present disclosure, and constitute a part of the description, together with the following specific embodiments, are used to explain the present disclosure, but do not constitute a limitation to the present disclosure. In the attached picture:

Fig. 1 is a flow chart showing a label generation method according to an exemplary embodiment.

Fig. 2 is a block diagram of a label generating device according to an exemplary embodiment.

Fig. 3 is another block diagram of a label generating device according to an exemplary embodiment.

Detailed ways

Specific embodiments of the present disclosure will be described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to illustrate and explain the present disclosure, and are not intended to limit the present disclosure.

Fig. 1 is a flow chart showing a method for generating a label according to an exemplary embodiment. As shown in Fig. 1 , the method for generating a label can be applied to a computer, and includes the following steps.

Step S11: Determine the access information of known objects and unknown objects accessed by the user within a preset time period.

Step S12: Determine the label of the unknown object according to the label of the known object and the access information.

Among them, known objects are labeled objects, unknown objects are unlabeled objects, and labels can be used to classify objects.

The objects in the embodiments of the present disclosure may be any objects that can be tagged, for example, news on news websites, commodities on shopping websites, novels, movies, websites, and so on. Tags can be used to classify objects, for example, if the tag of a certain movie includes suspense, then the movie can be classified as suspense movie, and so on.

A known object is an object that has already been tagged. For example, the tags of a certain piece of news include economy, real estate, and financial management. An unknown object is an object that has not been tagged yet, and the embodiment of the present disclosure can tag the unknown object by using the tag of the known object and the user's access behavior.

Optionally, the first user is any user among multiple users, and the first object is any object included in a known object or an unknown object. Determining the access information of the first object being accessed by the first user within the preset time period may determine that the first object is accessed by the first user according to the number of times the first user visits the first object within the preset time period and/or the duration of accessing the first object A user's access information.

The preset duration may be any preset duration, for example, it may be nearly one month, or it may be nearly six months, etc., which is not limited in this embodiment of the present disclosure.

Access information is generated when each user accesses a known object or each object in an unknown object, and the access information can be used to characterize the user's behavior of accessing an object. Taking the first user's visit to the first object as an example, the access information can be determined according to the number of times the first user visits the first object within a preset time period, or according to the length of time the first user visits the first object within a preset time period To determine the access information, the access information may also be jointly determined according to the number of times the first user accesses the first object within a preset time period and the length of time.

For example, in the past month, user 1 browsed product 1 five times, and user 1 browsed product 2 ten times. The visit information can reflect the number of times the user visits the object, so it can be recorded as p ₁ = 5 for the visit information of user 1 visiting product 1, and p ₂ = 10 for the visit information of user 1 visiting product 2. At this time, p ₁ < _p2 .

Or for example, in the past month, user 1 spent a total of 30 minutes browsing product 1, and user 1 spent a total of 10 minutes browsing product 2. The access information can be the information reflecting the duration of the user’s access object. Then, the access information of user 1’s access to commodity 1 can be recorded as p ₁ =30, and the access information of user 1’s access to commodity 2 can be recorded as p ₂ =10. At this time, p ₁ > _p2 .

Or for example, in the past month, user 1 browsed product 1 five times, and it took a total of 30 minutes, and user 1 browsed product 2 twice, and it took a total of 10 minutes. The access information can be the information that comprehensively reflects the number and duration of the user's access to the object through numerical values. For example, 1 point is scored for each visit, and 1 point is scored for every 10 minutes of access, so the access information of user 1's access to product 1 can be recorded as p ₁ =8, record the access information of user 1's access to product 2 as p ₂ =3, and at this time p ₁ >p ₂ .

Through the above method, the user's behavior of accessing the object can be represented more accurately through the access information, which is conducive to further more accurately generating the label of the unknown object.

The following describes how to generate labels for unknown objects.

Optionally, the known object includes multiple objects, the unknown object includes at least one object, and the user includes multiple users. After determining that the known object and the unknown object are accessed by the user within a preset period of time, it is also possible to generate a known object. The behavior matrix A of the known object, wherein, the behavior matrix A is used to indicate the access information that multiple users respectively visit each object included in the known object; and, generate the behavior matrix B of the unknown object, wherein, the behavior matrix B uses Based on the access information indicating that multiple users visit each object included in the unknown object, the label matrix N of the unknown object can be generated according to the label matrix M, behavior matrix A, and behavior matrix B of the known object to determine the unknown object. The labels of the objects, where the label matrix is used to indicate the weights of the objects respectively corresponding to different labels.

In the embodiments of the present disclosure, the number of known objects and users may be relatively large. In practical applications, the larger the number of known objects and users is, the more it can reflect the user's access habits. Then the obtained unknown object Labels are also more accurate. Representing the tags of known objects and the user's access behavior in the form of a matrix can record a large amount of data more accurately and conveniently, which is conducive to more accurately generating tags of unknown objects.

Optionally, according to the label matrix M, behavior matrix A, and behavior matrix B of the known object, the label matrix N of the unknown object can be generated, and the label matrix M can be multiplied by the behavior matrix A to obtain the user label matrix of the known object W, the user label matrix W is multiplied by the transpose matrix of the behavior matrix B to obtain the label matrix N of the unknown object.

The label matrix can be used to indicate the weights of the objects corresponding to different labels. For example, in the label matrix, news 1 corresponds to the military label with a weight of 5, and news 1 corresponds to the social label with a weight of 2, so that means news A label of 1 is news with a greater weight than a label that is social, and so on.

It can be assumed that U is a user set, including multiple users, T is a label set, I is an object set of known objects, J is an object set of unknown objects, and the label matrix of known objects is M(i×t). The behavior matrix of the object is A(u×i), and the behavior matrix of the unknown object is B(u×j).

A matrix of user labels can be generated:

W(u×t)=A(u×i)×M(i×t) (1)

Further, the label matrix of the object set J of unknown objects can be generated:

N(j×t)=B ^T (u×j)×W(u×t) (2)

In practical applications, in order to obtain data that can more truly reflect the user's behavior, the matrix W(u×t) and matrix N(j×t) can also be divided by the access information of the user's access object.

Optionally, the label matrix N of the unknown object can be generated according to the label matrix M, behavior matrix A, behavior matrix B, and penalty function F(C) of the known object, where, C is the number of times the object is accessed, and n is the preset times threshold.

In the embodiment of the present disclosure, in order to avoid deviations caused by the actions of a small number of active users, a penalty function F(C) is introduced for users with too many actions. For example, a penalty function is introduced for users with more than 100 visits, that is, Set n=100.

After introducing the penalty function, formula (1) can be improved as:

W(u×t)=A(u×i)×M(i×t)/F(C) (3)

Formula (2) can be improved as:

N(j× ^t )=BT(u×j)×W(u×t)/F(C) (4)

The technical solutions in the embodiments of the present disclosure will be described below through specific embodiments.

Known objects include object A, object B, and object C, tag set T includes tag 1, tag 2, tag 3, tag 4, and tag 5, user set U includes user A, user B, and user C, and unknown objects include object D and object E.

The label matrix M(i×t) of known objects is:

label 1 label 2 label 3 label 4 label 5 Object A 2 5 0 3 1 Object B 2 0 4 3 2 Object C 3 1 3 4 3

The behavior matrix A(u×i) of all users in the user set U on known objects is:

Object A Object B Object C User A 2 1 3 User B 4 0 1 User C 0 4 3

User label matrix W(u×t)=A(u×i)×M(i×t)/access information can be generated, namely:

The behavior matrix of all users in the user set U on the unknown object is B(u×j):

Object D Object E User A 4 1 User B 3 2 User C 1 5

The label matrix N(j×t)=B ^T (u×j)×W(u×t)/access information of unknown objects can be generated, namely:

After obtaining the label matrix N of the unknown object, the label of each object in the unknown object can be determined, and then a corresponding label can be generated for the unknown object. Through the above method, the label of the unknown object can be predicted more accurately without manually labeling the object, and for the unknown object lacking descriptive information, the label can also be generated more accurately through the user's access behavior.

Referring to FIG. 2, based on the same inventive concept, an embodiment of the present disclosure provides a label generation device 100, which may include:

The first determination module 101 is used to determine the access information of known objects and unknown objects accessed by users within a preset time period, wherein the known objects are tagged objects, unknown objects are untagged objects, and tags are used to classify objects;

The second determination module 102 is configured to determine the label of the unknown object according to the label of the known object and the access information.

Optionally, referring to FIG. 3, the known object includes multiple objects, the unknown object includes at least one object, and the user includes multiple users;

The device 100 also includes:

The first generation module 103 is configured to generate a behavior matrix A of known objects after determining the access information of known objects and unknown objects accessed by users within a preset period of time, wherein the behavior matrix A is used to indicate that multiple users respectively access information for each object included in the known object; and,

generating a behavior matrix B of the unknown object, wherein the behavior matrix B is used to indicate the access information that multiple users respectively access to each object included in the unknown object;

The second generation module 104 is used to generate the label matrix N of the unknown object according to the label matrix M, behavior matrix A, and behavior matrix B of the known object, so as to determine the label of the unknown object, wherein the label matrix is used to indicate that the objects are respectively Weights corresponding to different labels.

Optionally, the first user is any user among multiple users, and the first object is any object included in a known object or an unknown object;

The first determination module 101 is used for:

The access information that the first object is accessed by the first user is determined according to the number of times the first user accesses the first object within a preset time period and/or the duration of accessing the first object.

Optionally, the second generating module 104 is used for:

Multiply the label matrix M by the behavior matrix A to obtain the user label matrix W of known objects;

Multiply the user label matrix W by the transpose of the behavior matrix B to get the label matrix N for unknown objects.

Optionally, the second generating module 104 is used for:

According to the label matrix M, behavior matrix A, behavior matrix B, and penalty function F(C) of the known object, generate the label matrix N of the unknown object, where, C is the number of times the object is accessed, and n is the preset times threshold.

In the embodiments provided in the present disclosure, it should be understood that the disclosed devices and methods may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be Incorporation may either be integrated into another system, or some features may be omitted, or not implemented.

Each functional module in each embodiment of the present application may be integrated into one processing unit, or each module may physically exist separately, or two or more modules may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory, read-only memory), RAM (Random Access Memory, random access memory), magnetic disk or optical disk and other media that can store program codes. .

As mentioned above, the above embodiments are only used to introduce the technical solutions of the present disclosure in detail, but the descriptions of the above embodiments are only used to help understand the methods and core ideas of the present disclosure, and should not be construed as limiting the present disclosure. Within the technical scope disclosed in the present disclosure, any changes or substitutions that can be easily conceived by those skilled in the art shall fall within the protection scope of the present disclosure.

Claims (8)

1. a kind of label generating method, which is characterized in that the described method includes:

Determine the access information that known object and unknown object are accessed by the user in preset duration, wherein the known object For the object to have labelled, the unknown object is the object not labelled, and the label is used for classified object；

According to the label of the known object and the access information, the label of the unknown object is determined；

The known object includes multiple objects, and the unknown object includes at least an object, and the user includes multiple use Family；

After determining the access information that known object and unknown object are accessed by the user in preset duration, further includes:

Generate the behavioural matrix A of the known object, wherein it is right respectively that the behavioural matrix A is used to indicate the multiple user The access information that each object that the known object includes accesses；And

Generate the behavioural matrix B of the unknown object, wherein it is right respectively that the behavioural matrix B is used to indicate the multiple user The access information that each object that the unknown object includes accesses；

According to the label of the known object and the access information, the label of the unknown object is determined, comprising:

According to the label matrix M of the known object, the behavioural matrix A and the behavioural matrix B, it is described unknown right to generate The label matrix N of elephant, with the label of the determination unknown object, wherein label matrix is used to indicate object and corresponds respectively to not With the weight of label.

2. the method according to claim 1, wherein the first user be the multiple user in any user, Any object that first object is the known object or the unknown object includes；

Determine the access information that first object is accessed in the preset duration by first user, comprising:

According to first user accessed in the preset duration first object number and/or access it is described first pair The duration of elephant determines the access information that first object is accessed by first user.

3. the method according to claim 1, wherein according to the label matrix M of the known object, the behavior Matrix A and the behavioural matrix B, generate the label matrix N of the unknown object, comprising:

By the label matrix M multiplied by the behavioural matrix A, to obtain the user tag matrix W of the known object；

By the user tag matrix W multiplied by the transposed matrix of the behavioural matrix B, to obtain the label square of the unknown object Battle array N.

4. the method according to claim 1, wherein according to the label matrix M of the known object, the behavior Matrix A and the behavioural matrix B, generate the label matrix N of the unknown object, comprising:

It is raw according to the label matrix M of the known object, the behavioural matrix A, the behavioural matrix B and penalty F (C) At the label matrix N of the unknown object, whereinC is the number that object is accessed, and n is pre- If frequency threshold value.

5. a kind of label generating means, which is characterized in that described device includes:

First determining module, the access information being accessed by the user in preset duration for determining known object and unknown object, Wherein, the known object is the object to have labelled, and the unknown object is the object not labelled, and the label is for returning Class object；

Second determining module, for according to the known object label and the access information, determine the unknown object Label；

The known object includes multiple objects, and the unknown object includes at least an object, and the user includes multiple use Family；

Described device further include:

First generation module, in the access information for determining that known object and unknown object are accessed by the user in preset duration Later, the behavioural matrix A of the known object is generated, wherein it is right respectively that the behavioural matrix A is used to indicate the multiple user The access information that each object that the known object includes accesses；And

Generate the behavioural matrix B of the unknown object, wherein it is right respectively that the behavioural matrix B is used to indicate the multiple user The access information that each object that the unknown object includes accesses；

Second generation module, for the label matrix M, the behavioural matrix A and the behavioural matrix according to the known object B generates the label matrix N of the unknown object, with the label of the determination unknown object, wherein label matrix is used to indicate Object corresponds respectively to the weight of different labels.

6. device according to claim 5, which is characterized in that the first user is any user in the multiple user, Any object that first object is the known object or the unknown object includes；

First determining module is used for:

According to first user accessed in the preset duration first object number and/or access it is described first pair The duration of elephant determines the access information that first object is accessed by first user.

7. device according to claim 5, which is characterized in that second generation module is used for:

By the label matrix M multiplied by the behavioural matrix A, to obtain the user tag matrix W of the known object；

By the user tag matrix W multiplied by the transposed matrix of the behavioural matrix B, to obtain the label square of the unknown object Battle array N.

8. device according to claim 5, which is characterized in that second generation module is used for:

It is raw according to the label matrix M of the known object, the behavioural matrix A, the behavioural matrix B and penalty F (C) At the label matrix N of the unknown object, whereinC is the number that object is accessed, and n is pre- If frequency threshold value.