WO2024051707A1 - Recommendation model training method and apparatus, and resource recommendation method and apparatus - Google Patents

Abstract

Embodiments of the present disclosure provide a recommendation model training method and apparatus, and a resource recommendation method and apparatus. The recommendation model training method may comprise obtaining a first set of training data comprising a real conversion rate of a first-type user for a target resource, wherein the first-type user provides the real conversion rate. The method may further comprise obtaining a second set of training data comprising a predicted conversion rate of a second-type user for the target resource, wherein the second-type user does not provide a real conversion rate. In addition, the method may further comprise training a recommendation model using the first set of training data and the second set of training data. The recommendation model obtained according to the training mode of the present disclosure can accurately recommend resources to users, thereby improving user experience.

Description

Method for training recommendation model, method for recommending resources and device thereof

This application requests the priority of the Chinese patent application submitted to the State Intellectual Property Office of China on September 8, 2022, with the application number 202211098044.3 and the invention title "Method for training recommendation model, method for recommending resources and device", all of which The contents are incorporated into this application by reference.

Technical field

Embodiments of the present disclosure relate to the field of data processing, and more specifically, to methods of training recommendation models, methods, devices, electronic devices, and computer-readable storage media for recommending resources.

Background technique

With the rapid development of the Internet, the information people receive is also growing explosively. Recommendation systems need to recommend resources that users are interested in under the condition of information overload, thereby improving user experience and improving resource distribution efficiency. When the recommendation system faces the recommendation distribution of massive resources, it can apply various types of models to recommend interesting resources to users from tens of millions of resource libraries in milliseconds. Therefore, a recommendation model is needed to achieve accurate recommendation of resources.

Contents of the invention

Embodiments of the present disclosure provide a recommendation model training solution.

In a first aspect of the present disclosure, a method of training a recommendation model is provided. The method may include obtaining a first set of training data including a real conversion rate of a first type of user to the target resource, the first type of user providing the real conversion rate. The method may further include obtaining a second set of training data including a predicted conversion rate of a second type of user to the target resource, the second type of user not providing a true conversion rate. Additionally, the method may further include training the recommendation model using the first set of training data and the second set of training data.

In a second aspect of the present disclosure, a method of training a recommendation model is provided. The method may include determining an initial conversion rate for the target resource by multiple users, with a single user not providing a true conversion rate. The method may further include determining a correction factor based on the total conversion rate and the initial conversion rate of the plurality of users to the target resource. The method may further include determining a predicted conversion rate of the user to the target resource based on the correction factor and the initial conversion rate. Additionally, the method may further include training a recommendation model based at least on the predicted conversion rate.

In a third aspect of the present disclosure, a method of recommending resources is provided. The method may include obtaining user characteristics of the user and resource characteristics of the resource. The method may further include using the conversion rate model trained according to the methods of the first aspect and the second aspect to determine the user's conversion rate to the resource based on the user characteristics and the resource characteristics. In addition, the method may further include recommending resources to the user based on the conversion rate.

In a fourth aspect of the present disclosure, a device for training a recommendation model is provided. The device may include: a first training data acquisition module configured to acquire a first data including a real conversion rate of a first type of user to a target resource. A set of training data, the first type of user provides a true conversion rate; the second training data acquisition module is configured to obtain a second set of training data including the predicted conversion rate of the second type of user to the target resource, the second type of user does not provide a true conversion rate conversion rate; and a first model training module configured to train the recommendation model using the first set of training data and the second set of training data.

In a fifth aspect of the present disclosure, a device for training a recommendation model is provided. The device may include: an initial conversion rate determination module configured to determine the initial conversion rate of multiple users to the target resource, and a single user does not provide a real conversion. rate; the correction factor determination module is configured to determine the correction factor based on the total conversion rate and the initial conversion rate of multiple users to the target resource; the predicted conversion rate determination module is configured to determine the user's conversion rate based on the correction factor and the initial conversion rate. a predicted conversion rate of the target resource; and a second model training module configured to train the recommendation model based on at least the predicted conversion rate.

In a sixth aspect of the present disclosure, a device for recommending resources is provided. The device may include: a feature acquisition module configured to acquire user features of the user and resource features of the resource; a conversion rate determination module configured to utilize The conversion rate model trained by the method of the first aspect or the second aspect determines the user's response to the resource based on the user characteristics and resource characteristics. the conversion rate of the source; and the recommendation module is configured to recommend resources to users based on the conversion rate.

In a seventh aspect of the present disclosure, an electronic device is provided, including: a processor; and a memory coupled to the processor, the memory having instructions stored therein, the instructions when executed by the processor, cause the electronic device to perform according to the first Any step of the method of the first, second or third aspect.

In an eighth aspect of the present disclosure, there is provided a computer-readable storage medium having a computer program stored thereon, which when executed by a processor implements any steps of the method according to the first, second or third aspect.

This Content is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This summary is not intended to identify key features or key features of the disclosure or to limit the scope of the disclosure.

Description of the drawings

The above and other objects, features and advantages of the present disclosure will become more apparent by describing the exemplary embodiments of the present disclosure in more detail with reference to the accompanying drawings, in which the same or similar reference numerals are used in the exemplary embodiments of the present disclosure. Usually represents the same or similar parts. In the attached picture:

1 illustrates a schematic diagram of an example environment in which various embodiments of the present disclosure can be implemented;

2 illustrates a schematic diagram of a detailed example environment for training and applying models in accordance with embodiments of the present disclosure;

3 illustrates a flowchart of a process for training a recommendation model according to one embodiment of the present disclosure;

4 illustrates a flowchart of a process for training a recommendation model according to another embodiment of the present disclosure;

Figure 5 shows an overall flow chart of a scheme for training a recommendation model according to an embodiment of the present disclosure;

Figure 6 illustrates an apparatus for training a recommendation model according to one embodiment of the present disclosure. Schematic diagram of the installation;

Figure 7 shows a schematic diagram of an apparatus for training a recommendation model according to another embodiment of the present disclosure; and

Figure 8 shows a schematic block diagram of an example device that may be used to implement embodiments of the present disclosure.

Detailed ways

It can be understood that before using the technical solutions disclosed in the embodiments of this disclosure, users should be informed of the type, scope of use, usage scenarios, etc. of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations and obtain the user's authorization. .

For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly remind the user that the operation requested will require the acquisition and use of the user's personal information. Therefore, users can autonomously choose whether to provide personal information to software or hardware such as electronic devices, applications, servers or storage media that perform the operations of the technical solution of the present disclosure based on the prompt information.

As an optional but non-limiting implementation method, in response to receiving the user's active request, the method of sending prompt information to the user may be, for example, a pop-up window, and the prompt information may be presented in the form of text in the pop-up window. In addition, the pop-up window can also contain a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

It can be understood that the above process of notifying and obtaining user authorization is only illustrative and does not limit the implementation of the present disclosure. Other methods that satisfy relevant laws and regulations can also be applied to the implementation of the present disclosure.

It can be understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of corresponding laws, regulations and related regulations.

The principles of the present disclosure will be described below with reference to several example embodiments illustrated in the accompanying drawings. In the description of embodiments of the present disclosure, the term "including" and similar expressions shall be understood as an open inclusion, that is, "including but not limited to." The term "based on" shall be understood to mean "Based at least in part on". The terms "one embodiment" or "the embodiment" should be understood to mean "at least one embodiment". The terms "first", "second", etc. may refer to different or the same object. Other explicit and implicit definitions may be included below.

In the description of the embodiments of the present disclosure, the term "model" can learn the association between the corresponding input and the output from the training data, so that after the training is completed, the given input is processed based on the parameter set obtained by the training. Generate the corresponding output. A "model" may also sometimes be called a "neural network", "learning model", "learning network" or "network". These terms are used interchangeably herein.

The term "feature" refers to a vector representation of a resource or user. The nature of this feature vector makes objects corresponding to vectors with similar distances have similar meanings. For example, if two resources, cars and digital products, are both technological items, then the feature vectors of the car and the feature vectors of the digital products are relatively close in space. For another example, if user A and user B select entertainment information as tags of interest at the same time, then the characteristics of user A and user B are relatively close in space. The concept of "features" can be used to encode objects with vectors and retain the characteristics of their meaning, which is very suitable for deep learning. The term "recommendation" refers to the action of presenting or exposing various resources or content to users in various appropriate forms.

As described above, a recommendation model is needed to achieve accurate recommendation of resources. Traditionally, whether to recommend a user pair is usually determined based on the user's conversion rate of the resource. However, depending on the requirements of different operating systems and user settings, it is often impossible to obtain the true conversion rate as a true value label to train the recommendation model. In normal operation, some initial conversion rates can be used to train the recommendation model. However, due to the large difference between the initial conversion rate and the real conversion rate, the training effect is poor and an accurate recommendation model cannot be obtained. On the one hand, this results in a poor user experience, and on the other hand, the recommendation cost invested by the recommender is too high.

According to embodiments of the present disclosure, a recommendation model training scheme is proposed. This scheme divides the training data into two parts. For the first type of user who provides a real conversion rate, a first set of training data including the real conversion rate is obtained for recommendation model training. For the second type of users who do not provide a true conversion rate, a second set of training data including predicted conversion rates is obtained for recommendation model training. From this, data related to different types of users Used as different training data for recommendation model training, it can improve the generalization of the model. In addition, training the recommendation model based on the predicted conversion rate that is closest to the true conversion rate can improve the accuracy of the recommendation model in predicting the conversion rate, thereby solving the above problems and/or other potential problems.

Various embodiments of the present disclosure will be described in detail below in conjunction with example scenarios. It should be understood that this is for illustrative purposes only and is not intended to limit the scope of the present disclosure in any way.

Figure 1 illustrates a block diagram of an example system 100 for training a recommendation model in accordance with an embodiment of the present disclosure. It should be understood that the system 100 shown in FIG. 1 is only an example in which embodiments of the present disclosure can be implemented, and is not intended to limit the scope of the present disclosure. Embodiments of the present disclosure are equally applicable to other systems or architectures.

As shown in FIG. 1 , system 100 may include computing device 120 . Computing device 120 may be configured to receive input 110, which may include data associated with users and resources, such as user characteristics and resource characteristics. The computing device 120 generates a user conversion rate 130 for the resource based on the input 110 . Specifically, the computing device 120 may generate the conversion rate 130 through the recommendation model 140 disposed therein.

The user can be a user of various types of applications, and the application can be an application including a recommendation system, including but not limited to shopping applications, short video applications, music applications, dating applications, news applications, forum applications, cloud disk storage applications, Search applications, etc. This disclosure is not limited here.

Resources can be products, live broadcast rooms, short videos, pictures, music, character information, etc. in the above applications including recommendation systems. Users receive recommended videos, pictures, texts, voices or combinations thereof that are associated with resources in the above-mentioned applications. For example, after a user enters a news application, he or she receives recommended news cover images, news headline text information, or video information in the display interface. In this article, "resources", "content", "objects", etc. all refer to entities or virtual items that may need to be presented or exposed to users, and this disclosure is not limited here.

The user will go through the following process in the above application: first, the user sees the resource, and then the user may select (for example, click) the resource of interest. Then the user may perform further operations on the resource, such as purchasing, collecting, adding to shopping cart, and downloading. Uploading, forwarding, etc. This behavior is called conversion. Predicting the probability that the user will convert after the resource is shown to the user is called conversion rate estimation.

It is understandable that after the user selects the resource, he will leave the application interface and enter the resource-related page, and the user's actions thereafter will not be visible. Depending on the user settings, the type of operating system, and the settings of the resource parties, the real conversion rate of some users for the resource will be provided to the application (hereinafter, we call it the first type of user), while the real conversion rate of some users will be provided to the application party. The real conversion rate of this resource will not be provided to the application side (hereinafter, we call it the second type of user).

In the present disclosure, the recommendation model 140 may be designed to perform recommendation tasks. Examples of recommended models include, but are not limited to, various types of deep neural networks (DNN), convolutional neural networks (CNN), support vector machines (SVM), decision trees, random forest models, etc. In implementations of the present disclosure, a recommendation model may also be referred to as a "neural network," "learning model," "learning network," "model," and "network" interchangeably.

In some embodiments, computing device 120 may include, but is not limited to, a personal computer, a server computer, a handheld or laptop device, a mobile device (such as a mobile phone, personal digital assistant (PDA), media player, etc.), consumer electronics, small form factor device, etc. Computers, mainframe computers, cloud computing resources, etc.

It should be understood that the devices and/or elements within the devices included in the system 100 are exemplary only and are not intended to limit the scope of the present disclosure. It should be understood that system 100 may also include additional devices and/or units not shown. For example, in some embodiments, the computing device 120 of the system 100 may further include a storage unit (not shown) for storing pre-input hyperparameters and the like.

Training and use of the model in computing device 120 is described below with reference to FIG. 2 . Figure 2 shows a schematic diagram of a detailed example environment 200 in accordance with embodiments of the present disclosure. Similar to FIG. 1 , example environment 200 may include a computing device 220 , an input 210 into the computing device 220 , and a conversion rate 230 output from the computing device 220 . The difference is that the example environment 200 may generally include a model training system 260 and a model application system 270 . As examples, model training system 260 and/or model application system 270 may be implemented in computing device 120 as shown in FIG. 1 or computing device 220 as shown in FIG. 2 . Should be taken for granted It is understood that the structure and functionality of example environment 200 are described for illustrative purposes only and are not intended to limit the scope of the subject matter described herein. The subject matter described herein may be implemented in different structures and/or functions.

As mentioned before, the process of processing the input of the model to determine the user's conversion rate 130 of the resource can be divided into two stages: the model training stage and the model application stage. As an example, as shown in Figure 2, in the model training phase, model training system 260 may utilize training data 250 to train model 240. It should be understood that the training data 250 may be a triplet of (user characteristics; resource characteristics; user-to-resource conversion rate). In some embodiments, training data 250 may include a first set of training data 252 associated with a first type of user, in which the user's true conversion rate for the resource is provided. In some other embodiments, training data 250 may include a second set of training data 254 associated with a second type of user, where the user's true conversion rate for the resource is not provided. At this time, the true conversion rate needs to be predicted for training the model 240, and the prediction process will be described in detail below. Alternatively, in some embodiments, training data 250 may include both first set of training data 252 and second set of training data 242 .

In the model application phase, the model application system 270 may receive the trained model 240 . Thus, the model 240 loaded into the computing device 220 of the model application system 270 can determine the conversion rate 230 based on the input 210 .

In other embodiments, model 240 may be constructed as a learning network. In some embodiments, the learning network may include multiple networks, where each network may be a multi-layer neural network, which may be composed of a large number of neurons. Through the training process, the corresponding parameters of each neuron in the network can be determined. The parameters of the neurons in these networks are collectively referred to as the parameters of the model 240 .

The training process of the model 240 may be performed in an iterative manner until at least some of the parameters of the model 240 converge or until a predetermined number of iterations is reached, thereby obtaining final model parameters.

The technical solutions described above are only used as examples and do not limit the present disclosure. It should be understood that each network can also be arranged in other ways and connection relationships. In order to explain the principle of the above solution more clearly, the following will describe the recommended model training in more detail with reference to Figure 3. process.

Figure 3 illustrates a flow diagram of a process 300 for training a recommendation model in accordance with an embodiment of the present disclosure. In some embodiments, process 300 may be implemented in computing device 120 in FIG. 1 and computing device 220 in FIG. 2 . A process 300 of training a recommendation model according to an embodiment of the present disclosure is now described with reference to FIG. 3 . For ease of understanding, the specific examples mentioned in the following description are illustrative and are not intended to limit the scope of the present disclosure.

At 302, the computing device 120 may obtain a first set of training data including a true conversion rate of a first type of user to the target resource, the first type of user providing the true conversion rate. For example, according to the type of operating system, the user's settings, and the resource party's settings, the computing device 120 may obtain the true conversion rate of the first type of user to the target resource. Computing device 120 may then train the model based on this true conversion rate.

In some embodiments, the computing device 120 may determine the user characteristics of the first type of user and the resource characteristics of the target resource, and then obtain the real conversion rate of the first type of user to the target resource. For example, the training data for user i can be expressed as (x _i , y _i , z _i ). Among them, x _i represents user characteristics and resource characteristics; yi _i uses different values to indicate whether the user has selected (clicked) the target resource displayed to the user. yi _i can be a value between 0 and 1, for example, 0 means no selection. , 1 represents selection; z _i represents the conversion rate of whether the user has made the choice, that is, whether the target behavior has been implemented. The target behavior is the behavior that is considered to be converted by the user in the corresponding scenario. z _i can be a value between 0 and 1. For example, 0 means conversion has occurred and 1 means no conversion has occurred. Please note that the above training data is only exemplary, and different forms of training data may also exist, and the disclosure is not limited here.

For the determination of user characteristics and resource characteristics, in some embodiments, computing device 120 may determine user characteristics and resource characteristics respectively. For example, computing device 120 may characterize the user based on the user's historical selection of resources. Computing device 120 may determine resource characteristics based on one or more of the resource category, the resource publisher, and user characteristics of users who have historically selected the resource. The above method of determining characteristics is only exemplary, and the computing device 120 may determine user characteristics and resource characteristics based on other suitable inherent characteristics of users and resources to accurately represent the complex non-linear relationship between users and resources.

Alternatively, in some embodiments, the user's interaction information with the resource can be used to determine Determine the user characteristics of the user and the resource characteristics of the resource. For example, a node graph can be constructed based on the relationship between users' clicks, sharing, publishing and other operations on resources, where each node represents a user and a resource. Then determine the user characteristics and resource characteristics by walking in the node graph. It can be understood that by accurately representing the characteristics of users and resources, the feature capacity and generalization of the recommendation model to be trained can be improved.

The above describes the acquisition and determination of training data for users whose true prediction rates are known, and the following describes the acquisition and determination of training data for users whose true prediction rates are unknown. At 304, the computing device 120 may obtain a second set of training data including predicted conversion rates for the target resource by a second type of user who does not provide a true conversion rate. It is understandable that for some types of operating systems, dual authorization from users and resource related parties is sometimes required to obtain the true conversion rate of a single user. At this time, the computing device 120 needs to predict the conversion rate based on existing data.

In one example, computing device 120 may first determine user characteristics of the second type of user and resource characteristics of the target resource. Regarding the method of determining user characteristics and resource characteristics, please refer to the above description and will not be repeated here. Computing device 120 may then determine a predicted conversion rate based on user behavior after the second type of user selects the target resource. For example, although the computing device 120 cannot directly obtain the user's true conversion rate, it can predict the conversion rate based on the user's interaction behavior with the target resource.

In some embodiments, the computing device 120 may determine the initial conversion rate as the predicted conversion rate based on the action relationship between the second type user and the target resource. It can be understood that the more operations the user performs on the resource, or the longer it takes for the user to enter other interfaces and then jump back to the application after selecting the resource, it means that the user has a greater probability of implementing conversion behavior. For example, the computing device 120 may determine one or more of whether the user likes the target resource, whether the user forwards the target resource, the time when the user switches back to the resource display interface after clicking on the target resource, and whether the user downloads the target resource. item to determine the initial conversion rate as the predicted conversion rate. The initial conversion rate can be a value between 0 and 1. Conversion rates can be accurately predicted based on the relationship between user actions and resources within the app.

Alternatively, in some embodiments, computing device 120 may provide the above The action relationship between the first type of conversion rate user and the target resource and its true conversion rate are used to train the machine learning model to obtain a trained initial conversion rate model. Computing device 120 may then utilize the trained initial conversion rate model to predict the initial conversion rate for the second type of user. It can be understood that due to the difference between the training data used (i.e., first type users, target resources, and the real conversion rate between them) and the data used for prediction (i.e., second type users, target resources, and the initial conversion rate between them) Conversion rate) are all targeted at the same target resource, which allows the initial conversion rate to be accurately predicted, that is, closer to the true conversion rate.

It is understandable that sometimes it is not accurate enough based only on user interaction behavior. The initial conversion rate can also be corrected based on some other data to obtain a predicted conversion rate that is closer to the real conversion rate.

Additionally or alternatively, in some embodiments, although the conversion rate of an individual user to the target resource is not provided due to various factors, the total conversion rate of all (or a portion) of the users in a plan to the target resource may be given. In this case, the computing device 120 may determine the correction factor based on the total conversion rate of the resource by the first type of users and the second type of users, the true conversion rate, and the initial conversion rate. For example, the correction factor C can be determined according to the following equation (1):
C=(TR)/I Equation (1)

The total conversion rate T is the ratio between the number of converted users among the first type users and the second type users and the total number of the first type users and the second type users, and the real conversion rate R is the ratio among the first type users The ratio between the number of converted users and the total number of first-type users and second-type users. The initial conversion rate I may be the average of the predicted initial conversion rates of all second-type users. By multiplying the numerator and denominator in equation (1) by the total number of first type users and second type users at the same time, the above equation (1) can also be presented in the form of the following equation (2):
C＝(TN-FN)/IN Equation (2)

Among them, TN is the total number of conversions, FN is the number of converted users among the first type of users, and IN is the initial number of conversions corresponding to the initial conversion rate.

Computing device 120 then applies the correction factor to the initial conversion rate to obtain a corrected initial conversion rate as the predicted conversion rate. For example, computing device 120 may multiply the initial conversion rate of each second type user by a correction factor as the predicted conversion rate. It can be understood that correcting the predicted initial conversion rate based on the obtained total conversion rate can make the predicted conversion rate more accurate, thereby improving the prediction accuracy of the subsequently trained recommendation model.

After determining and obtaining training data for different types of users, at 306, the computing device 120 may use the first set of training data and the second set of training data to train the recommendation model. As mentioned above, training data includes user characteristics, resource characteristics, and conversion rates. The computing device 120 may, for example, input user characteristics and resource characteristics into the initial model to obtain a predicted conversion rate. The error between the predicted conversion rate and the conversion rate as the ground-truth label is then determined, and the computing device 120 then propagates the error in the opposite direction (ie, from the output layer to the input layer of the model to be trained). During the backpropagation process, you can rely on the gradient descent algorithm to adjust the values of parameters of each layer in the model to be trained. According to multiple rounds of training, the error between the prediction and the actual value of the model to be trained will become smaller and smaller until the model converges and the training process is completed. From this, the computing device 120 obtains the recommended model.

Through the above-mentioned embodiments, the present disclosure can accurately predict the user's conversion rate through user behavior and the total conversion rate when the conversion rate of a single user for resources cannot be obtained. In addition, the present disclosure uses the accurate conversion rate obtained above to train the recommendation model, which can improve the prediction accuracy and generalization of the recommendation model. Furthermore, the trained recommendation model can be used to accurately recommend resources to users, improving user experience and reducing resource related party costs.

The above describes the solution in which two types of users exist at the same time. The following describes the solution in which only the second type of users exists. 4 illustrates a flowchart of a process for training a recommendation model according to another embodiment of the present disclosure.

At 402, computing device 120 determines multiple users' initial conversion rates to the target resource and that a single user does not provide a true conversion rate. The process of determining the initial conversion rate is similar to the step described in 304 and will not be described again here.

At 404, the computing device 120 based on the total conversion rate of the plurality of users to the target resource and the initial Initial conversion rate, determine the correction factor. Unlike the step described in 304, there is no first type of user. Computing device 120 may determine the correction factor C according to equation (3) below:
C＝T/I Equation (3)

The total conversion rate T is the ratio between the number of converted users among the second type users and the total number of second type users, and the initial conversion rate I can be the average of the initial conversion rates of all second type users predicted above. value.

At 406, computing device 120 determines the user's predicted conversion rate for the target resource based on the correction factor and the initial conversion rate. Computing device 120 may apply the correction factor to the initial conversion rate to obtain a corrected initial conversion rate as a predicted conversion rate. For example, computing device 120 may multiply the initial conversion rate of each second type user by a correction factor as the predicted conversion rate. It can be understood that correcting the predicted initial conversion rate based on the obtained total conversion rate can make the predicted conversion rate more accurate, thereby improving the prediction accuracy of the subsequently trained recommendation model.

At 408, computing device 120 trains a recommendation model based at least on the predicted conversion rate. The process of training the model is similar to the steps described in 306 and will not be described again here.

Through the above process, the present disclosure can accurately predict the user's conversion rate through user behavior and the total conversion rate when only the second type of user exists. In addition, the present disclosure uses the accurate conversion rate obtained above to train the recommendation model, which can improve the prediction accuracy and generalization of the recommendation model. Furthermore, the trained recommendation model can be used to accurately recommend resources to users, improving user experience and reducing resource related party costs.

Figure 5 shows an overall flowchart of a scheme for training a recommendation model according to an embodiment of the present disclosure. First, the training data 510 is divided into training data for the first type of users and training data for the second type of users. The training data for the first type of users includes the true conversion rate of 540. The conversion rate of the training data of the second type of user needs to be determined according to the above-mentioned processes 330 and 400, in which the incident conversion rate 550, the correction factor 570 and the predicted conversion rate 560 are determined respectively. Then, the training data of the first type of user and the training data of the second type of user can be respectively input into the value model 520 for training, and the output of the model 520 is the conversion rate 530. For specific steps, please refer to the above description and will not be repeated here.

The training process of the recommendation model is described in detail above, and the application process of the recommendation model is described below. First, the computing device 120 obtains the user characteristics of the user and the resource characteristics of the resource. For example, the computing device 120 may determine user characteristics and resource characteristics. For the determination process of user characteristics and resource characteristics, refer to the above description and will not be described again here. In some embodiments, the computing device 120 may also call pre-stored user characteristics and resource characteristics from the database according to the user identification and resource identification.

Computing device 120 may then determine the user's conversion rate to the resource based on the user characteristics and the resource characteristics according to the conversion rate model trained in processes 300 and 400 . For example, the computing device 120 may use user characteristics and resource characteristics as inputs to the recommendation model to derive a predicted conversion rate. The computing device 120 then recommends resources to the user based on the conversion rate. For example, the computing device 120 may rank a user's conversion rates for multiple resources and recommend resources that are in front of the predetermined sort order to the user.

The present disclosure also provides a model training device. Specifically, FIG. 6 shows a schematic diagram of an apparatus 600 for training a recommendation model according to an embodiment of the present disclosure. As shown in Figure 6, the device 600 may at least include: a first training data acquisition module 602 configured to acquire a first set of training data including the real conversion rate of a first type of user to the target resource, and the first type of user provides real conversion. rate; the second training data acquisition module 604 is configured to obtain a second set of training data including the predicted conversion rate of a second type of user to the target resource, the second type of user does not provide a true conversion rate; and the first model training module 606 , is configured to train the recommendation model using the first set of training data and the second set of training data.

In some embodiments, the second training data acquisition module 604 may include: a first feature determination module configured to determine user features of the second type of user and resource features of the target resource; and a first prediction module configured to The predicted conversion rate is determined based on user behavior after the second type of user selects the target resource.

In some embodiments, the first prediction module may include: a second prediction module configured to determine the initial conversion rate as the predicted conversion rate based on the action relationship between the second type user and the target resource.

In some embodiments, the action relationship between the second type user and the target resource includes Including at least one of the following: whether the user likes the target resource, whether the user forwards the target resource, the time when the user switches back to the resource display interface after clicking on the target resource, and whether the user downloads the target resource.

In some embodiments, the apparatus 600 may further include: a correction factor module configured to determine the correction factor based on the total conversion rate, the real conversion rate and the initial conversion rate of the resource by the first type of users and the second type of users; and and a correction application module configured to apply the correction factor to the initial conversion rate to obtain a corrected initial conversion rate as the predicted conversion rate.

In some embodiments, the first training data acquisition module 602 may include: a second characteristic determination module configured to determine user characteristics of the first type of user and resource characteristics of the target resource; and a conversion rate acquisition module configured to Get the true conversion rate.

In some embodiments, the first feature determination module and the second feature determination module may include: a user feature determination module configured to determine user features based on the user's historical selection of resources.

In some embodiments, the first feature determination module and the second feature determination module may include: a resource feature determination module configured to be based on at least one of resource categories, resource publishers, and user features of users who have historically selected the resource. Items determine resource characteristics.

FIG. 7 shows a schematic diagram of an apparatus 700 for training a recommendation model according to another embodiment of the present disclosure. As shown in Figure 7, the device 700 may at least include: an initial conversion rate determination module 702, configured to determine the initial conversion rate of multiple users to the target resource, and a single user does not provide a true conversion rate; a correction factor determination module 704, configured to Determine a correction factor based on the total conversion rate and the initial conversion rate of multiple users to the target resource; the predicted conversion rate determination module 706 is configured to determine the predicted conversion rate of the user to the target resource based on the correction factor and the initial conversion rate; and The second model training module 708 is configured to train the recommendation model based on at least the predicted conversion rate.

In addition, although not shown, the present disclosure also provides a device for recommending resources, which may include: a feature acquisition module configured to acquire user features of the user and resource features of the resource; a conversion rate determination module configured to utilize According to the conversion rate model trained by the process 300 and 400 methods, based on user characteristics and resource characteristics, determine the user's response to the resource conversion rate; and a recommendation module configured to recommend resources to users based on the conversion rate.

Figure 8 shows a schematic block diagram of an example device 800 that may be used to implement embodiments of the present disclosure. For example, computing device 120 shown in FIG. 1 and computing device 220 shown in FIG. 2 may be implemented by device 800. As shown, the device 800 includes a central processing unit (CPU) 801 that can operate on a computer in accordance with computer program instructions stored in a read-only memory (ROM) 802 or loaded from a storage unit 808 into a random access memory (RAM) 803 Program instructions to perform various appropriate actions and processes. In the RAM 803, various programs and data required for the operation of the device 900 can also be stored. CPU 801, ROM 802 and RAM 803 are connected to each other via bus 804. An input/output (I/O) interface 805 is also connected to bus 804.

Multiple components in the device 800 are connected to the I/O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a magnetic disk, optical disk, etc. ; and 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 through computer networks such as the Internet and/or various telecommunications networks. It should be understood that the present disclosure can use the output unit 807 to display real-time dynamic change information of user satisfaction, key factor identification information of satisfied group users or individual users, optimization strategy information, and strategy implementation effect evaluation information, etc.

The processing unit 801 may be implemented by one or more processing circuits. The processing unit 801 may be configured to perform the various processes and processes described above, such as processes 300, 400, and 500. For example, in some embodiments, processes 300, 400, and 500 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 may be loaded and/or installed onto device 800 via ROM 802 and/or communication unit 809. When the computer program is loaded into RAM 803 and executed by CPU 801, one or more steps in processes 300, 400, and 500 described above may be performed.

The present disclosure may be a system, method, and/or computer program product. A computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for performing various aspects of the present disclosure.

Computer-readable storage media may be tangible devices that can retain and store instructions for use by an instruction execution device. The computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or Flash memory), Static Random Access Memory (SRAM), Compact Disk Read Only Memory (CD-ROM), Digital Versatile Disk (DVD), Memory Stick, Floppy Disk, Mechanical Coding Device, such as a printer with instructions stored on it. Protruding structures in hole cards or grooves, and any suitable combination of the above. As used herein, computer-readable storage media are not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or through electrical wires. transmitted electrical signals.

Computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to various computing/processing devices, or to an external computer or external storage device over a network, such as the Internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage on a computer-readable storage medium in the respective computing/processing device .

Computer program instructions for performing operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or instructions in one or more programming languages. Source code or object code written in any combination of object-oriented programming languages - such as Smalltalk, C++, etc., and conventional procedural programming languages - such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer. be executed on a computer or server. In situations involving remote computers, the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as an Internet service provider through the Internet). connect). In some embodiments, by utilizing state information of computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can Computer readable program instructions are executed to implement various aspects of the disclosure.

Aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, thereby producing a machine such that the instructions, when executed by a processing unit of the computer or other programmable data processing apparatus, , resulting in an apparatus that implements the functions/actions specified in one or more blocks in the flowchart and/or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium. These instructions cause the computer, programmable data processing device and/or other equipment to work in a specific manner. Therefore, the computer-readable medium storing the instructions includes An article of manufacture that includes instructions that implement aspects of the functions/acts specified in one or more blocks of the flowcharts and/or block diagrams.

Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other equipment, causing a series of operating steps to be performed on the computer, other programmable data processing apparatus, or other equipment to produce a computer-implemented process , thereby causing instructions executed on a computer, other programmable data processing apparatus, or other equipment to implement the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams.

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment Or a part of an instruction. The module, program segment or part of the instruction contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two consecutive blocks may actually execute substantially in parallel, or they may sometimes execute in the reverse order, depending on the functionality involved. It will also be noted that each block of the block diagram and/or flowchart illustration, and combinations of blocks in the block diagram and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts. , or can be implemented using a combination of specialized hardware and computer instructions.

In accordance with one or more embodiments of the present disclosure. Example 1. A method of training a recommendation model, including: obtaining a first set of training data including a first type of user's true conversion rate to a target resource, the first type of user providing the true conversion rate; obtaining a second set of training data including a second type of user's true conversion rate to a target resource. a second set of training data for the predicted conversion rate of a type of user to the target resource, the second type of user not providing the true conversion rate; and using the first set of training data and the second set of training data to Train the recommendation model.

Example 2. The method of Example 1, wherein obtaining the second set of training data includes: determining user characteristics of the second type of user and resource characteristics of the target resource; and selecting based on the second type of user User behavior following the target resource determines the predicted conversion rate.

Example 3. The method according to Example 1-2, wherein determining the predicted conversion rate includes: determining an initial conversion rate as the predicted conversion rate based on an action relationship between the second type user and the target resource.

Example 4. The method according to Example 1-3, wherein the action relationship between the second type user and the target resource includes at least one of the following: whether the user likes the target resource, whether the user forwards the target resource , the time for the user to switch back to the resource display interface after clicking on the target resource, and whether the user downloads the target resource.

Example 5. The method according to examples 1-4, further comprising based on the total conversion rate of the resource by the first type of users and the second type of users, the true conversion rate and the initial conversion rate, determining a correction factor; and applying said correction factor to said The initial conversion rate is used to obtain the corrected initial conversion rate as the predicted conversion rate.

Example 6. The method according to examples 1-5, wherein obtaining the first set of training data includes: determining user characteristics of the first type of user and resource characteristics of the target resource; and obtaining the true conversion rate.

Example 7. The method of examples 1-6, wherein determining the user characteristics includes determining the user characteristics based on the user's historical selection of resources.

Example 8. The method of Examples 1-7, wherein determining the resource characteristics of the target resource includes determining the resource based on at least one of a resource category, a resource publisher, and user characteristics of a user who has historically selected the resource. feature.

Example 9. A method of training a recommendation model, including: determining the initial conversion rate of multiple users to the target resource, and a single user does not provide a real conversion rate; based on the total conversion rate of the multiple users to the target resource and the an initial conversion rate, determining a correction factor; based on the correction factor and the initial conversion rate, determining a user's predicted conversion rate for the target resource; and training the recommendation model based at least on the predicted conversion rate.

Example 10. A method of recommending resources, including: obtaining user characteristics of the user and resource characteristics of the resource; using a conversion rate model trained according to the method described in any one of Examples 1 to 9, based on the user characteristics and the resource characteristics The resource characteristics are used to determine the user's conversion rate of the resource; and based on the conversion rate, the resource is recommended to the user.

Example 11. A device for training a recommendation model, including: a first training data acquisition module configured to acquire a first set of training data including the true conversion rate of a first type of user to a target resource, the first type of user providing The real conversion rate; a second training data acquisition module configured to obtain a second set of training data including the predicted conversion rate of a second type of user to the target resource, the second type of user not providing the real conversion rate; and a first model training module configured to train the recommendation model using the first set of training data and the second set of training data.

Example 12. The apparatus according to Example 11, the obtaining the second training data acquisition module includes: a first feature determination module configured to determine user features of the second type user and resources of the target resource Features; and a first prediction module configured to determine based on user behavior after the second type user selects the target resource. Determine the predicted conversion rate.

Example 13. The apparatus according to Example 11 or 12, the first prediction module includes: a second prediction module configured to determine an initial conversion based on an action relationship between the second type user and the target resource. rate as the predicted conversion rate.

Example 14. The device according to examples 11-13, wherein the action relationship between the second type user and the target resource includes at least one of the following: whether the user likes the target resource, whether the user forwards the target resource , the time for the user to switch back to the resource display interface after clicking on the target resource, and whether the user downloads the target resource.

Example 15. The apparatus according to examples 11-14, the apparatus further comprising: a correction factor module configured to be based on the total conversion rate of the resource by the first type of users and the second type of users, the The true conversion rate and the initial conversion rate are used to determine a correction factor; and a correction application module is configured to apply the correction factor to the initial conversion rate to obtain a corrected initial conversion rate as a predicted conversion rate.

Example 16. The apparatus according to examples 11-15, the first training data acquisition module includes: a second feature determination module configured to determine user features of the first type of user and resource features of the target resource ; and a conversion rate acquisition module configured to obtain the true conversion rate.

Example 17. The apparatus according to examples 11-16, the first feature determination module and the second feature determination module may include: a user feature determination module configured to determine the user feature based on the user's historical selection of resources. .

Example 18. The apparatus according to Examples 11-17, the first feature determination module and the second feature determination module may include: a resource feature determination module configured to select resources based on resource categories, resource publishers, and historical selections of resources. At least one of the user characteristics of the user determines the resource characteristics.

Example 19. A device for training a recommendation model, including: an initial conversion rate determination configured to determine the initial conversion rate of multiple users to a target resource, and a single user does not provide a true conversion rate; a correction factor determination module configured to determine based on the Determine a correction factor based on the total conversion rate of the multiple users to the target resource and the initial conversion rate; the predicted conversion rate determination module is configured to determine the user based on the correction factor and the initial conversion rate. a user's predicted conversion rate for the target resource; and a second model training module configured to train the recommendation model based at least on the predicted conversion rate.

Example 20. A device for recommending resources, including: a feature acquisition module configured to acquire user characteristics of the user and resource characteristics of the resource; a conversion rate determination module configured to utilize the method according to any one of Examples 1 to 10 The conversion rate model trained by the method determines the user's conversion rate of resources based on the user characteristics and the resource characteristics; and a recommendation module is configured to recommend resources to the user based on the conversion rate.

Example 21. An electronic device, comprising: a processor; and a memory coupled to the processor, the memory having instructions stored therein that, when executed by the processor, cause the electronic device to perform actions, The actions include: obtaining a first set of training data including a real conversion rate of a first type user to the target resource, the first type user providing the real conversion rate; obtaining a first set of training data including a second type user to the target resource. Predicting a second set of training data for a conversion rate, the second type of user not providing the true conversion rate; and using the first set of training data and the second set of training data to train the recommendation model.

Example 22. An electronic device, comprising: a processor; and a memory coupled to the processor, the memory having instructions stored therein that, when executed by the processor, cause the electronic device to perform actions, The actions include: determining an initial conversion rate of multiple users to the target resource, and a single user does not provide a true conversion rate; determining a correction factor based on the total conversion rate of the multiple users to the target resource and the initial conversion rate; Determine a user's predicted conversion rate for the target resource based on the correction factor and the initial conversion rate; and train the recommendation model based on at least the predicted conversion rate.

Example 23. An electronic device, comprising: a processor; and a memory coupled to the processor, the memory having instructions stored therein that, when executed by the processor, cause the electronic device to perform actions, The actions include: obtaining user characteristics of the user and resource characteristics of the resource; using a conversion rate model trained according to the method described in any one of Examples 1 to 9, based on the user characteristics and the resource characteristics, determining the The user's conversion rate of resources; and based on the conversion rate, recommendations to the user resource.

Example 24. A computer-readable storage medium having a computer program stored thereon, which when executed by a processor implements the method described in any one of Examples 1-10.

The embodiments of the present disclosure have been described above. The above description is illustrative, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims (15)

1. A method of training a recommendation model, including:

   Obtaining a first set of training data including a true conversion rate of a first type of user to the target resource, the first type of user providing the true conversion rate;

   Obtaining a second set of training data including a predicted conversion rate of a second type of user to the target resource, the second type of user not providing the true conversion rate; and

   The recommendation model is trained using the first set of training data and the second set of training data.
2. The method of claim 1, wherein obtaining the second set of training data includes:

   Determining user characteristics of the second type user and resource characteristics of the target resource; and

   The predicted conversion rate is determined based on user behavior after the second type user selects the target resource.
3. The method of claim 2, wherein determining the predicted conversion rate includes:

   Based on the action relationship between the second type user and the target resource, an initial conversion rate is determined as the predicted conversion rate.
4. The method according to claim 3, wherein the action relationship between the second type user and the target resource includes at least one of the following:

   Whether the user likes the target resource, whether the user forwards the target resource, the time when the user switches back to the resource display interface after clicking on the target resource, and whether the user downloads the target resource.
5. The method of claim 3, further comprising

   determining a correction factor based on the total conversion rate of the resource by the first type of users and the second type of users, the true conversion rate and the initial conversion rate; and

   The correction factor is applied to the initial conversion rate to obtain a corrected initial conversion rate as the predicted conversion rate.
6. The method of claim 1, wherein obtaining the first set of training data includes:

   Determining user characteristics of the first type user and resource characteristics of the target resource; and

   Get said true conversion rate.
7. The method of claim 2 or 6, wherein determining the user characteristics includes:

   The user characteristics are determined based on the user's historical selection of resources.
8. The method according to claim 2 or 6, wherein determining the resource characteristics of the target resource includes:

   The resource characteristics are determined based on at least one of a resource category, a resource publisher, and user characteristics of users who have historically selected the resource.
9. A method of training a recommendation model, including:

   Determine the initial conversion rate of multiple users to the target resource, and users among the multiple users do not provide a true conversion rate;

   Determine a correction factor based on the total conversion rate of the multiple users to the target resource and the initial conversion rate;

   determining a user's predicted conversion rate for the target resource based on the correction factor and the initial conversion rate; and

   The recommendation model is trained based on at least the predicted conversion rate.
10. A method of recommending resources including:

    Obtain the user characteristics of the user and the resource characteristics of the resource;

    Using a conversion rate model trained according to the method of any one of claims 1 to 9, based on the user characteristics and the resource characteristics, determining the user's conversion rate of resources; and

    Based on the conversion rate, resources are recommended to the user.
11. A device for training recommendation models, including:

    A first training data acquisition module configured to acquire a first set of training data including a real conversion rate of a first type of user to the target resource, the first type of user providing the real conversion rate;

    The second training data acquisition module is configured to acquire a second set of training data including the predicted conversion rate of a second type of user to the target resource. The second type of user has not mentioned to provide stated true conversion rates; and

    A first model training module configured to train the recommendation model using the first set of training data and the second set of training data.
12. A device for training recommendation models, including:

    Initial conversion rate determination is configured to determine the initial conversion rate of multiple users to the target resource, and a single user does not provide a true conversion rate;

    a correction factor determination module configured to determine a correction factor based on the total conversion rate of the multiple users to the target resource and the initial conversion rate;

    a predicted conversion rate determination module configured to determine a user's predicted conversion rate for the target resource based on the correction factor and the initial conversion rate; and

    A second model training module is configured to train the recommendation model based on at least the predicted conversion rate.
13. A device for recommending resources, including:

    a feature acquisition module configured to acquire user features of the user and resource features of the resource;

    A conversion rate determination module configured to determine the user's conversion rate of resources based on the user characteristics and the resource characteristics using a conversion rate model trained according to the method of any one of claims 1 to 10; as well as

    A recommendation module configured to recommend resources to the user based on the conversion rate.
14. An electronic device including:

    processor; and

    A memory coupled to the processor, the memory having instructions stored therein that when executed by the processor cause the electronic device to perform the method of any one of claims 1-10.
15. A computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method according to any one of claims 1-10 is implemented.