CN115357796A

CN115357796A - Interest classification model training method, interest classification method and device

Info

Publication number: CN115357796A
Application number: CN202211049323.0A
Authority: CN
Inventors: 徐靖宇; 刘昊骋; 徐世界; 王天祺
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-08-30
Filing date: 2022-08-30
Publication date: 2022-11-18

Abstract

The disclosure provides an interest classification model training method, an interest classification method and an interest classification device, and relates to the technical field of artificial intelligence, in particular to the technical fields of data processing, deep learning and the like. The specific implementation scheme is as follows: acquiring a positive sample user and an unlabelled user, wherein the positive sample user is a user with an interest label, and the unlabelled user is a user without the interest label; determining spyware users from the positive sample users; determining reliable negative sample users from the unlabeled users based on the spy users; and training the target model to be trained based on the positive sample user and the reliable negative sample user to obtain a trained target interest classification model. The accuracy of the interest classification model can be improved by the implementation mode.

Description

Interest classification model training method, interest classification method and device

Technical Field

The disclosure relates to the technical field of artificial intelligence, in particular to the technical fields of data processing, deep learning and the like.

Background

At present, in an application scenario of interest classification of a user, an interest classification model is often required to be trained in advance, and an interest tag interested by the user is determined based on the interest classification model. For example, based on a pre-trained interest classification model, it is determined whether the user is interested in the car.

In practice, it is found that it is generally difficult to determine negative examples by the current interest classification model training method. For example, if a user frequently visits a website related to a car, it may be determined that the user is interested in the car, but if a user does not frequently visit a website related to a car, it may not be said that the user is not interested in the car. Based on this, since the negative sample is difficult to determine, the accuracy of the trained interest classification model is poor.

Disclosure of Invention

The disclosure provides an interest classification model training method, an interest classification method and an interest classification device.

According to an aspect of the present disclosure, there is provided an interest classification model training method, including: acquiring a positive sample user and an unlabelled user, wherein the positive sample user is a user with an interest label, and the unlabelled user is a user without the interest label; determining spyware users from the positive sample users; determining reliable negative sample users from the unlabeled users based on the spy users; and training the target model to be trained based on the positive sample user and the reliable negative sample user to obtain a trained target interest classification model.

According to another aspect of the present disclosure, there is provided an interest classification method including: acquiring a target user; based on the target interest classification model, determining an interest tag corresponding to a target user; and classifying the interest of the target user according to the interest label.

According to another aspect of the present disclosure, there is provided an interest classification model training apparatus, including: the system comprises a sample acquisition unit and a sample analysis unit, wherein the sample acquisition unit is configured to acquire a positive sample user and an unlabelled user, the positive sample user is a user with an interest label, and the unlabelled user is a user without the interest label; a spy determining unit configured to determine a spy user from the positive sample users; the negative sample determining unit is configured to determine reliable negative sample users from the unlabeled users based on the spy users; and the model training unit is configured to train the target model to be trained on the basis of the positive sample user and the reliable negative sample user to obtain a trained target interest classification model.

According to another aspect of the present disclosure, there is provided an interest classification apparatus including: a user acquisition unit configured to acquire a target user; the label determining unit is configured to determine an interest label corresponding to the target user based on the target interest classification model; and the interest classification unit is configured to perform interest classification on the target user according to the interest labels.

According to another aspect of the present disclosure, there is provided an electronic device including: one or more processors; a memory for storing one or more programs; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any one of the above methods for interest classification model training or interest classification.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform any one of the interest classification model training method or the interest classification method as described above.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements an interest classification model training method or an interest classification method as any one of the above.

According to the technology disclosed by the invention, the interest classification model training method or the interest classification method is provided, and the accuracy of the interest classification model can be improved.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is an exemplary system architecture diagram in which one embodiment of the present disclosure may be applied;

FIG. 2 is a flow diagram of one embodiment of a method of interest classification model training according to the present disclosure;

FIG. 3 is a schematic diagram of an application scenario of an interest classification model training method according to the present disclosure;

FIG. 4 is a flow diagram of another embodiment of an interest classification model training method according to the present disclosure;

FIG. 5 is a flow diagram for one embodiment of a method of interest classification according to the present disclosure;

FIG. 6 is a schematic diagram of an embodiment of an interest classification model training apparatus according to the present disclosure;

FIG. 7 is a schematic block diagram of one embodiment of an interest classification apparatus according to the present disclosure;

fig. 8 is a block diagram of an electronic device for implementing an interest classification model training method or an interest classification method according to an embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of embodiments of the present disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

It should be noted that, in the present disclosure, the embodiments and the features of the embodiments may be combined with each other without conflict. The present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

As shown in fig. 1, the system architecture 100 may include

terminal devices

101, 102, 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

The user may use the

terminal devices

101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like. The

terminal devices

101, 102, and 103 may obtain positive sample users and unlabeled users for training the interest classification model, and send the positive sample users and the unlabeled users to the server 105 through the network 104, so that the server 105 determines spy users from the positive sample users, determines reliable negative sample users from the unlabeled users based on the spy users, and trains a target model to be trained based on the positive sample users and the reliable negative sample users to obtain a target interest classification model. After that, the

terminal devices

101, 102, and 103 may obtain the target interest classification model, and determine information such as an interest tag and an interest classification of each user based on the target interest classification model.

The

terminal apparatuses

101, 102, and 103 may be hardware or software. When the

terminal devices

101, 102, 103 are hardware, they may be various electronic devices including, but not limited to, mobile phones, computers, tablets, and the like. When the

terminal apparatuses

101, 102, 103 are software, they can be installed in the electronic apparatuses listed above. It may be implemented as a plurality of software or software modules (for example to provide distributed services) or as a single software or software module. And is not particularly limited herein.

The server 105 may be a server providing various services, for example, the server 105 may receive positive sample users and unlabeled users sent by the

terminal devices

101, 102, 103 through the network 104, determine spyware from the positive sample users, determine reliable negative sample users from the unlabeled users based on the spyware, and train the target model to be trained based on the positive sample users and the reliable negative sample users to obtain the target interest classification model. The server 105 may then send the target interest classification model to the

terminal devices

101, 102, 103 via the network 104.

The server 105 may be hardware or software. When the server 105 is hardware, it may be implemented as a distributed server cluster composed of a plurality of servers, or may be implemented as a single server. When the server 105 is software, it may be implemented as multiple pieces of software or software modules (e.g., to provide distributed services), or as a single piece of software or software module. And is not particularly limited herein.

It should be noted that the interest classification model training method or the interest classification method provided in the embodiment of the present disclosure may be executed by the

terminal devices

101, 102, and 103, or may be executed by the server 105, and the interest classification model training apparatus or the interest classification apparatus may be disposed in the

terminal devices

101, 102, and 103, or may be disposed in the server 105, which is not limited in the embodiment of the present disclosure.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

With continued reference to FIG. 2, a flow 200 of one embodiment of a method of interest classification model training in accordance with the present disclosure is shown. The interest classification model training method of the embodiment comprises the following steps:

step 201, a positive sample user and an unlabeled user are obtained, wherein the positive sample user is a user with an interest tag, and the unlabeled user is a user without the interest tag.

In this embodiment, the executing entity (such as the server 105 or the

terminal devices

101, 102, 103 in fig. 1) may obtain the positive sample users and the unmarked users for model training from the locally stored or pre-connected electronic devices. The positive sample user may be a user with an interest tag, and the unlabeled user may be a user without an interest tag. The interest tag may be a tag corresponding to an object in which the user is interested, for example, the interest tag may be "car", "food", "travel", and the like, which is not limited in this embodiment. The type of the interest tag may be one, or may be at least two, which is not limited in this embodiment. For the case that the category of the interest label is one, the trained interest classification model is a binary classification model, that is, whether the user has a specific interest label is determined, for example, whether the user is interested in a car is determined. For the case that the types of the interest labels are at least two, the trained interest classification model may be a multi-classification model, that is, it is determined what the interest labels of the user are, for example, it is determined whether the user has at least one of the interest labels of "car", "food", "travel", and the like.

Optionally, the execution subject may count historical access data of each user in advance, and generate an interest tag of each user based on the historical access data. Specifically, the execution subject may determine, for each type of interest tag, each website access data corresponding to the type of interest tag; the website access data comprises the frequency of accessing the website corresponding to the interest tag by each user; and determining the interest tag corresponding to the website frequently visited by the user as the interest tag of the user for each user based on analysis of the website visiting data. It will be appreciated that for each interest tag, there may be a positive sample user with the interest tag and an unlabeled user without the interest tag.

In step 202, a spyware is determined from the positive sample users.

In this embodiment, the execution subject may extract a part of the positive sample users from the positive sample users as spyware. The spy user and the unmarked users are used as negative samples to conduct model training at the same time, and reliable negative sample users are determined from the unmarked users based on model prediction results in the model training process.

Optionally, the execution subject may randomly extract spyware users in a preset proportion from the positive sample users, and may also randomly extract a preset number of spyware users from the positive sample users.

And step 203, determining reliable negative sample users from the unlabeled users based on the spyware.

In this embodiment, after determining that the spyware is obtained, the execution subject may perform model training by using the spyware and the unmarked users together as negative examples. In the process of model training, the execution subject can predict the interest classification of the unmarked user and the spy user simultaneously by using the model obtained by current training.

For an application scenario where the number of interest tags is one, the prediction result output by the interest classification model to be trained may be a numerical value between 0 and 1, where the closer the numerical value of the prediction result is to 0, the user performing prediction does not have a specific interest tag, and the closer the numerical value of the prediction result is to 1, the user performing prediction has a specific interest tag.

For the spy user, the essence of the spy user belongs to a positive sample user, but the spy user is used as a negative sample to perform model training in a model training stage, so that the score distribution of the prediction result of the spy user can reflect the reliability of the negative sample. Thus, the executive may determine a score threshold for unannotated users based on the score distribution of the spyware. Thereby determining reliable negative examples from among the unlabeled users based on the score threshold and the actual predicted scores for each of the unlabeled users.

And step 204, training the target model to be trained based on the positive sample user and the reliable negative sample user to obtain a trained target interest classification model.

In this embodiment, the executing subject may train the model to be trained based on the positive sample user and the reliable negative sample user after obtaining the reliable negative sample user, so as to obtain the trained target interest classification model. The target interest classification model can output corresponding interest label information based on the input user information.

Specifically, in the process of training a target model to be trained, the execution subject can simultaneously score a spy user and an unlabeled user by using a current model in each iteration process, re-determine a score threshold value based on the score distribution of the spy user, re-select a newly added reliable negative sample user from the unlabeled users based on the score threshold value, update the reliable negative sample user based on the newly added reliable negative sample user, and perform model training by using the updated reliable negative sample user.

With continued reference to FIG. 3, a schematic diagram of one application scenario of the interest classification model training method according to the present disclosure is shown. In the application scenario of fig. 3, the executing agent may obtain a positive sample user 301 and an unlabeled user 302 for training the interest classification model, and then determine a spyware from the positive sample user 301. Thereafter, the executing agent may perform a first stage of model training based on the positive samples 303 and the negative samples 304, with the spyware-removed positive sample users as positive samples 303 of the model training, and the unlabeled users and the spyware as negative samples 304 of the model training. Based on the training results of the first stage of model training, reliable negative examples are determined 305 from the unlabeled users. For the specific manner of the reliable negative sample 305, please refer to the above description, and the detailed description thereof is omitted here. After obtaining the reliable negative examples 305, the executing entity may perform a second stage of model training based on the positive examples 303 and the reliable negative examples 305, and obtain a trained target interest classification model 306.

According to the interest classification model training method provided by the embodiment of the disclosure, the positive sample users with interest labels and the unmarked users without interest labels can be obtained, the reliable negative sample users are determined from the unmarked users based on the spy users extracted from the positive sample users and the spy users in the model training process, so that the model training is performed based on the positive sample users and the reliable negative sample users, and the accuracy of the interest classification model can be improved.

With continued reference to FIG. 4, a flow 400 of another embodiment of an interest classification model training method according to the present disclosure is shown. As shown in fig. 4, the interest classification model training method of this embodiment may include the following steps:

step 401, a positive sample user and an unlabeled user are obtained, wherein the positive sample user is a user with an interest tag, and the unlabeled user is a user without an interest tag.

In this embodiment, please refer to the detailed description of step 201 for the detailed description of step 401, which is not repeated herein.

A spyware is determined from the positive sample users, step 402.

In this embodiment, please refer to the detailed description of step 202 for the detailed description of step 402, which is not repeated herein.

And step 403, removing the spy user from the positive sample users to obtain a first positive sample user.

In this embodiment, the number of positive sample users is multiple, the number of spyware users is multiple, and the number of spyware users is smaller than the number of positive sample users. The execution subject may remove other positive sample users of the spyware as the first positive sample user after extracting the spyware from the positive sample users.

Step 404, the users without marks and the spyware are determined as the first negative sample users.

In this embodiment, the execution principal may use the unlabeled user and the spyware as the first negative sample user.

Step 405, based on the first positive sample user and the first negative sample user, performing model training on the initial model to be trained to obtain a trained initial interest classification model.

In this embodiment, the initial model to be trained may be a classification model. The execution subject can perform model training on the initial model to be trained based on the first positive sample user and the first negative sample user until the model meets a preset convergence condition, so as to obtain a trained initial interest classification model.

In some optional implementations of this embodiment, the method further includes: and determining the trained initial interest classification model as a target model to be trained.

In this implementation manner, the execution subject may use the trained initial interest classification model as a target model to be trained, and further train the initial interest classification model based on the positive sample user and the reliable negative sample user to obtain a target interest classification model, thereby further improving the accuracy of the interest classification model.

And 406, scoring the spyware and the unmarked users based on the initial interest classification model to obtain a first spyware scoring result and a first unmarked user scoring result.

In this embodiment, after the execution subject obtains the trained initial interest classification model, the execution subject may score the spy user by using the initial interest classification model to obtain a first spy user scoring result. And scoring the unmarked users by using the initial interest classification model to obtain a first unmarked user scoring result. The first spyware scoring result can be used for indicating an interest tag prediction result of the spyware, and the first unlabeled user scoring result can be used for indicating an interest tag prediction result of the unlabeled user.

Step 407, determining reliable negative sample users from the unmarked users based on the first spyware scoring result and the first unmarked user scoring result.

In this embodiment, the execution subject may compare the scoring results of the spyware and the unlabeled users based on the first scoring result of the spyware and the first scoring result of the unlabeled users, and determine reliable negative sample users from the unlabeled users based on the comparison result.

In some optional implementations of this embodiment, determining reliable negative examples users from among the unlabeled users based on the first spyware scoring result and the first unlabeled user scoring result includes: generating a score threshold value based on the scoring result of the first spy user; and determining the un-annotated user with the score lower than a score threshold value as a reliable negative sample user based on the scoring result of the first un-annotated user.

In this implementation, the executing agent may generate a score threshold based on the first spyware scoring result. Optionally, the execution main body may determine a minimum score in the first spyware scoring result as a score threshold, or determine a score corresponding to 25 deciles in the first spyware scoring result as a score threshold, which is not limited in this embodiment. Wherein the score threshold may be adjusted based on user demand. Then, for each un-labeled user, the execution subject may compare the score of the un-labeled user with a score threshold, and determine the un-labeled user with the score lower than the score threshold as a reliable negative sample user.

Step 408, remove spy users from the positive sample users to obtain a second positive sample user.

In this embodiment, the execution subject may take the positive sample user without the spyware as the positive sample user of the second stage of model training, that is, the second positive sample user, when performing the second stage of model training.

In step 409, the reliable negative example user is determined as the second negative example user.

In this embodiment, the executive subject may use the reliable negative example user as the negative example user of the model training in the second stage, i.e., the second negative example user.

And step 410, training the target model to be trained based on the second positive sample user and the second negative sample user to obtain a trained target interest classification model.

In this embodiment, the executing entity may train the target to-be-trained model based on the second positive sample user and the second negative sample user, that is, train the initial interest classification model to obtain a trained target interest classification model.

In some optional implementations of this embodiment, the method further includes: in each iteration of training the target model to be trained, scoring the spy users based on the target model to be trained to obtain a second spy user scoring result; scoring the unmarked users without the reliable negative sample users based on the target training model to obtain a second unmarked user scoring result; selecting users to be marked from the unmarked users without the reliable negative sample users based on the second spy user marking result and the second unmarked user marking result; and updating the reliable negative sample user based on the user to be marked.

In this implementation, the executing agent may score the spy user by using the current target model to be trained for each iteration while performing the second stage model training, and obtain the second spy user scoring result. And scoring the unmarked users without the reliable negative sample users by using the current target model to be trained to obtain a second spy user scoring result and a second unmarked user scoring result. And then, selecting users to be marked from the unmarked users without the reliable negative examples users based on the second spy user marking result and the second unmarked user marking result. And the user to be marked is an unmarked user to be marked as a reliable negative sample user. The executing agent may then update the reliable negative example user based on the user to be annotated, and perform the next iteration based on the updated reliable negative example user.

Optionally, the model convergence condition for training the target model to be trained may be that the number of the users to be labeled is smaller than a threshold or that the model meets a preset convergence condition.

In the interest classification model training method provided by the embodiment of the disclosure, the positive sample user from which the spyware is removed can be used as the first positive sample user, the unmarked user and the spyware can be used as the first negative sample user, and the initial classification model is subjected to the first-stage model training based on the first positive sample user and the first negative sample user. And based on the model training of the first stage, based on the scoring results of the current model for the spy users and the users which are not marked, the reliable negative sample users are determined from the users which are not marked, so that the determination accuracy of the reliable negative sample users is improved. And taking the reliable negative sample user as a second negative sample user, taking the positive sample user without the spy user as a second positive sample user, performing second-stage model training on the initial interest classification model based on the second positive sample user and the second negative sample user, and obtaining a final target interest classification model based on the two-stage model training, thereby further improving the accuracy of the interest classification model. And in the model training process of the second stage, reliable negative sample users can be updated in real time, the reliability of the negative samples is improved, and therefore the accuracy of the interest classification model is improved.

With continued reference to FIG. 5, a flow 500 of one embodiment of a method of interest classification in accordance with the present disclosure is shown. The interest classification method of the embodiment comprises the following steps:

step 501, a target user is obtained.

In this embodiment, the executing entity (such as the server 105 or the

terminal devices

101, 102, 103 in fig. 1) may obtain the target user needing to determine the interest tag from the electronic device which is locally stored or has a connection established in advance.

Step 502, based on the target interest classification model, determining an interest tag corresponding to a target user.

In this embodiment, the executing agent may input the user information of the target user into the target interest classification model obtained by the training of the interest classification model training method, so as to obtain the interest label information output by the target interest classification model. Based on parsing the interest tag information, the executive agent may determine an interest tag corresponding to the target user. Wherein the number of interest tags may be at least one.

And 503, classifying the interest of the target user according to the interest label.

In this embodiment, the execution subject may classify the target user into a specified interest category according to the interest tag, so as to realize interest classification of the target user. Optionally, after the target user is subjected to interest classification according to the interest tags, the execution subject may also push related content to the target user based on the interest classification result. Alternatively, the executive agent may also update the user representation of the target user based on the interest classification results.

According to the interest classification method provided by the embodiment of the disclosure, the target interest classification model obtained by training through the interest classification model training method can be used for performing interest classification on the target user, so that the accuracy of interest classification is improved.

With further reference to fig. 6, as an implementation of the methods shown in the above diagrams, the present disclosure provides an embodiment of an interest classification model training apparatus, where the embodiment of the apparatus corresponds to the embodiment of the method shown in fig. 6, and the apparatus may be specifically applied to electronic devices such as a terminal device and a server.

As shown in fig. 6, the interest classification model training apparatus 600 of the present embodiment includes a sample acquisition unit 601, a spy determination unit 602, a negative sample determination unit 603, and a model training unit 604.

The sample acquiring unit 601 is configured to acquire a positive sample user and an unlabeled user, where the positive sample user is a user with an interest tag, and the unlabeled user is a user without an interest tag.

A spy determining unit 602 configured to determine a spy user from the positive sample users.

A negative sample determination unit 603 configured to determine reliable negative sample users from the unlabeled users based on the spyware.

The model training unit 604 is configured to train the target to-be-trained model based on the positive sample user and the reliable negative sample user, so as to obtain a trained target interest classification model.

In some optional implementations of this embodiment, the negative example determining unit 603 is further configured to: removing the spy user from the positive sample user to obtain a first positive sample user; determining users who are not marked and spy users as first negative sample users; model training is carried out on the initial model to be trained based on the first positive sample user and the first negative sample user, and an initial interest classification model after training is obtained; scoring the spyware and the unmarked users based on the initial interest classification model to obtain a first spyware scoring result and a first unmarked user scoring result; and determining reliable negative sample users from the unmarked users based on the scoring result of the first spyware and the scoring result of the first unmarked users.

In some optional implementations of this embodiment, the negative example determining unit 603 is further configured to: generating a score threshold value based on the scoring result of the first spy user; and determining the unmarked users with the scores lower than the score threshold value as reliable negative sample users based on the scoring result of the first unmarked users.

In some optional implementations of this embodiment, the method further includes: and the model determining unit is configured to determine the trained initial interest classification model as the target model to be trained.

In some optional implementations of this embodiment, the model training unit 604 is further configured to: removing the spy user from the positive sample user to obtain a second positive sample user; determining a reliable negative example user as a second negative example user; and training the target model to be trained based on the second positive sample user and the second negative sample user to obtain a trained target interest classification model.

In some optional implementations of this embodiment, the model training unit 604 is further configured to: in each iteration of training the target model to be trained, scoring the spy users based on the target model to be trained to obtain a second spy user scoring result; scoring the unmarked users without the reliable negative sample users based on the target training model to obtain a second unmarked user scoring result; selecting users to be marked from the unmarked users without the reliable negative sample users based on the second spy user marking result and the second unmarked user marking result; and updating the reliable negative sample user based on the user to be marked.

It should be understood that units 601 to 604 respectively recited in the interest classification model training apparatus 600 correspond to the respective steps in the method described with reference to fig. 2. Thus, the operations and features described above for the interest classification model training method are also applicable to the apparatus 600 and the units included therein, and are not described herein again.

With further reference to fig. 7, as an implementation of the methods shown in the above-mentioned figures, the present disclosure provides an embodiment of an interest classification apparatus, where the embodiment of the apparatus corresponds to the embodiment of the method shown in fig. 5, and the apparatus may be specifically applied to electronic devices such as a terminal device and a server.

As shown in fig. 7, the interest classification apparatus 700 of the present embodiment includes: a user acquisition unit 701, a label determination unit 702 and an interest classification unit 703.

A user acquisition unit 701 configured to acquire a target user.

A tag determining unit 702 configured to determine an interest tag corresponding to the target user based on the target interest classification model.

An interest classification unit 703 configured to perform interest classification on the target user according to the interest labels.

It should be understood that the units 701 to 703 recited in the interest classification apparatus 700 respectively correspond to the respective steps in the method described with reference to fig. 5. Thus, the operations and features described above for the interest classification method are also applicable to the apparatus 700 and the units included therein, and are not described herein again.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

FIG. 8 shows a schematic block diagram of an example electronic device 800 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 8, the apparatus 800 includes a computing unit 801 which can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 802 or a computer program loaded from a storage unit 808 into a Random Access Memory (RAM) 803. In the RAM803, various programs and data necessary for the operation of the device 800 can also be stored. The calculation unit 801, the ROM 802, and the RAM803 are connected to each other by a bus 804. An input/output (I/O) interface 805 is also connected to bus 804.

A number of components in the device 800 are connected to the I/O interface 805, including: an input unit 806, such as a keyboard, a mouse, or the like; an output unit 807 such as various types of displays, speakers, and the like; a storage unit 808, such as a magnetic disk, optical disk, or the like; and a communication unit 809 such as a network card, modem, wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

Computing unit 801 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 801 performs the respective methods and processes described above, such as an interest classification model training method or an interest classification method. For example, in some embodiments, the interest classification model training method or the interest classification method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and/or installed onto device 800 via ROM 802 and/or communications unit 809. When loaded into RAM803 and executed by the computing unit 801, a computer program may perform one or more steps of the interest classification model training method or the interest classification method described above. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the interest classification model training method or the interest classification method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user may provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server combining a blockchain.

It should be understood that various forms of the flows shown above, reordering, adding or deleting steps, may be used. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. An interest classification model training method comprises the following steps:

acquiring a positive sample user and an unlabelled user, wherein the positive sample user is a user with an interest tag, and the unlabelled user is a user without the interest tag;

determining spyware from the positive sample users;

determining reliable negative sample users from the unmarked users based on the spy users;

and training the target model to be trained on the basis of the positive sample user and the reliable negative sample user to obtain a trained target interest classification model.

2. The method of claim 1, wherein said determining reliable negative examples users from among the unlabeled users based on the spyware comprises:

removing the spy user from the positive sample user to obtain a first positive sample user;

determining the unmarked user and the spy user as a first negative sample user;

model training is carried out on the initial model to be trained on the basis of the first positive sample user and the first negative sample user, and an initial interest classification model which is trained is obtained;

scoring the spyware and the unmarked users based on the initial interest classification model to obtain a first spyware scoring result and a first unmarked user scoring result;

and determining the reliable negative sample users from the unmarked users based on the first spy user scoring result and the first unmarked user scoring result.

3. The method of claim 2, wherein said determining said reliable negative sample user from said unlabeled users based on said first spyware scoring result and said first unlabeled user scoring result comprises:

generating a score threshold value based on the first spy user scoring result;

and determining the unmarked users with the scores lower than the score threshold value as the reliable negative sample users based on the scoring result of the first unmarked users.

4. The method of claim 2, further comprising:

and determining the trained initial interest classification model as the target model to be trained.

5. The method of claim 1, wherein the training a target model to be trained based on the positive sample user and the reliable negative sample user to obtain a trained target interest classification model comprises:

removing the spy user from the positive sample user to obtain a second positive sample user;

determining the reliable negative example user as a second negative example user;

and training the target model to be trained based on the second positive sample user and the second negative sample user to obtain the trained target interest classification model.

6. The method of claim 5, further comprising:

in each iteration of training the target model to be trained, scoring the spy users based on the target model to be trained to obtain a second spy user scoring result; scoring the unmarked users without the reliable negative sample users based on the target training model to obtain a second unmarked user scoring result;

selecting users to be marked from the unmarked users without the reliable negative sample users based on the second spyware marking result and the second unmarked user marking result;

and updating the reliable negative sample user based on the user to be marked.

7. An interest classification method comprising:

acquiring a target user;

determining an interest tag corresponding to the target user based on the target interest classification model according to any one of claims 1 to 6;

and classifying the interest of the target user according to the interest label.

8. An interest classification model training apparatus comprising:

the sample acquiring unit is configured to acquire a positive sample user and an unlabeled user, wherein the positive sample user is a user with an interest tag, and the unlabeled user is a user without the interest tag;

a spyware determining unit configured to determine a spyware user from the positive sample users;

a negative sample determining unit configured to determine reliable negative sample users from the unmarked users based on the spy users;

and the model training unit is configured to train the target to-be-trained model based on the positive sample user and the reliable negative sample user to obtain a trained target interest classification model.

9. The apparatus of claim 8, wherein the negative examples determination unit is further configured to:

removing the spy user from the positive sample users to obtain a first positive sample user;

determining the unmarked user and the spy user as a first negative sample user;

based on the initial interest classification model, scoring is carried out on the spy users and the unmarked users to obtain a first spy user scoring result and a first unmarked user scoring result;

10. The apparatus of claim 9, wherein the negative examples determination unit is further configured to:

generating a score threshold value based on the first spy user scoring result;

and determining the un-annotated user with the score lower than the score threshold value as the reliable negative sample user based on the first un-annotated user scoring result.

11. The apparatus of claim 9, further comprising:

and the model determining unit is configured to determine the trained initial interest classification model as the target model to be trained.

12. The apparatus of claim 8, wherein the model training unit is further configured to:

13. The apparatus of claim 12, wherein the model training unit is further configured to:

in each iteration of training the target model to be trained, scoring the spy user based on the target model to be trained to obtain a second spy user scoring result; scoring the unmarked users without the reliable negative sample users based on the target training model to obtain a second unmarked user scoring result;

and updating the reliable negative sample user based on the user to be marked.

14. An interest classification apparatus comprising:

a user acquisition unit configured to acquire a target user;

a tag determination unit configured to determine an interest tag corresponding to the target user based on the target interest classification model according to any one of claims 1 to 6;

and the interest classification unit is configured to perform interest classification on the target user according to the interest tag.

15. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

16. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-7.

17. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-7.