Method and device for training prediction model and predicting click rate

Abstract

The embodiment of the invention discloses a method and a device for training a prediction model and predicting a click rate. The specific implementation mode of the method comprises the steps that samples and an initial prediction model are obtained, the samples comprise user features, commodity features, context features, scene data and expected values of at least one target, and the prediction model comprises a feature input layer, an embedded layer, a progressive hierarchical extraction layer, a star network layer, an output layer and an auxiliary layer; inputting the user features, the commodity features and the context features in the sample into an embedded layer to obtain fusion features; inputting the fusion features into a progressive hierarchical extraction layer, and outputting common information; inputting scene data in the sample into the auxiliary layer, and outputting scene features; inputting the common information and the scene features into a star network layer, and outputting a predicted value of each target; and adjusting the network parameters of the prediction model based on the difference between the predicted value and the expected value of each target. According to the embodiment, the influence of interaction of multiple scenes can be solved, and the accuracy of model prediction is improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and particularly to methods and apparatuses for training a prediction model and predicting click-through rates. Background Art

[0002] With the rapid development of the Internet, the amount of information faced by users has shown an explosive growth, resulting in users having to invest a large amount of time and energy in finding content of interest. On the other hand, the interests, needs, and preferences of different users vary, making personalized recommendation an urgent problem to be solved. E-commerce recommendation systems perform personalized recommendations by predicting the click-through rates of users on candidate products through a model.

[0003] In Internet products, there are often multiple scenarios. The differences between these scenarios mainly lie in the differences in data distribution, the composition of users (new and old users), the content of page display forms, the sparsity of data, etc.

[0004] At the same time, each scenario hopes to use a model for personalized recommendation to improve business results. In this case, there are several problems: 1. Modeling each scenario separately will increase the modeling cost and maintenance cost. 2. Maintaining a model for scenarios with a large data volume and scenarios with a small data volume will cause the scenarios with a small data volume to be insufficiently learned and vulnerable to the influence of the data distribution of large scenarios. Summary of the Invention

[0005] Embodiments of the present disclosure propose methods and apparatuses for training a prediction model and predicting click-through rates.

[0006] In a first aspect, an embodiment of the present disclosure provides a method for training a prediction model, including: obtaining a sample and an initial prediction model, where the sample includes user features, product features, context features, scenario data, and the expected value of at least one target, and the prediction model includes a feature input layer, an embedding layer, a progressive hierarchical extraction layer, a star network layer, an output layer, and an auxiliary layer; after inputting the user features, product features, and context features in the sample into the feature input layer and summarizing through the embedding layer, obtaining a fused feature; inputting the fused feature into the progressive hierarchical extraction layer to output common information; inputting the scenario data in the sample into the auxiliary layer to output scenario features; inputting the common information and the scenario features into the star network layer to learn the interaction effects between the networks, and outputting the predicted value of each target through the output layer; adjusting the network parameters of the prediction model based on the difference between the predicted value of each target and the expected value.

[0007] In some embodiments, the user features include interest preference features; and the method further includes: obtaining the browsing page records of the user within a predetermined time; training the embedding vectors of each page in the browsing page records by means of deep walk; and performing average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user.

[0008] In some embodiments, before training the embedding vectors of each page in the browsing page records by means of deep walk, the method further includes: filtering out the pages entered for the first time from the browsing page records.

[0009] In some embodiments, the performing average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user includes: performing weighted average pooling on the embedding vectors of all pages in the browsing page records according to preset weights to obtain the interest preference features of the user, where the weights are inversely proportional to the time elapsed since browsing.

[0010] In some embodiments, before inputting the scenario data in the sample into the auxiliary layer to output scenario features, the method further includes: normalizing the data of different scenarios respectively.

[0011] In a second aspect, an embodiment of the present disclosure provides a method for predicting click-through rate, including: obtaining user features, product features of candidate products, and scenario data; inputting the user features, product features of candidate products, and scenario data into a prediction model trained by the method according to any one of the first aspect, and outputting the predicted click-through rate and / or conversion rate.

[0012] In a third aspect, an embodiment of the present disclosure provides an apparatus for training a prediction model, including: an obtaining unit configured to obtain a sample and an initial prediction model, where the sample includes user features, product features, context features, scenario data, and expected values of at least one target, and the prediction model includes a feature input layer, an embedding layer, a progressive hierarchical extraction layer, a star network layer, an output layer, and an auxiliary layer; a fusion unit configured to input the user features, product features, and context features in the sample into the feature input layer and, after summarization through the embedding layer, obtain fusion features; a common information extraction unit configured to input the fusion features into the progressive hierarchical extraction layer and output common information; a scenario feature extraction unit configured to input the scenario data in the sample into the auxiliary layer and output scenario features; a prediction unit configured to input the common information and the scenario features into the star network layer to learn the interaction effects between the networks and output the predicted values of each target through the output layer; and an adjustment unit configured to adjust the network parameters of the prediction model based on the difference between the predicted values and the expected values of each target.

[0013] In some embodiments, the user feature includes an interest preference feature; and the apparatus further includes a pre-training unit configured to: obtain the browsing page records of the user within a predetermined time; train the embedding vectors of each page in the browsing page records by means of deep walk; perform average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference feature of the user.

[0014] In some embodiments, the pre-training unit is further configured to: filter out the pages entered for the first time from the browsing page records before training the embedding vectors of each page in the browsing page records by means of deep walk.

[0015] In some embodiments, the pre-training unit is further configured to: perform weighted average pooling on the embedding vectors of all pages in the browsing page records according to preset weights to obtain the interest preference feature of the user, where the weights are inversely proportional to the time elapsed since browsing.

[0016] In some embodiments, the apparatus further includes a normalization unit configured to: normalize the data of different scenarios respectively before inputting the scenario data in the sample into the auxiliary layer and outputting the scenario features.

[0017] In a fourth aspect, an embodiment of the present disclosure provides an apparatus for predicting click-through rate, including: an acquisition unit configured to acquire user features, product features of candidate products, and scenario data; a prediction unit configured to input the user features, product features of candidate products, and scenario data into a prediction model trained according to the method described in any one of the first aspect, and output the predicted click-through rate and / or conversion rate.

[0018] In a fifth aspect, an embodiment of the present disclosure provides an electronic device, including: one or more processors; a storage device storing one or more computer programs thereon, and when the one or more computer programs are executed by the one or more processors, the one or more processors implement the method described in any one of the first aspect or the second aspect.

[0019] In a sixth aspect, an embodiment of the present disclosure provides a computer-readable medium storing a computer program thereon, where the computer program, when executed by a processor, implements the method described in any one of the first aspect or the second aspect.

[0020] The training and prediction model, method, and apparatus for predicting click-through rate provided by the embodiments of the present disclosure aim to solve the problem of the prediction accuracy of the prediction model in the e-commerce system in multiple scenarios. Using one model to solve multiple scenarios with interacting effects, through this network structure, the content of the mutual influence or commonality of the scenarios is learned, and at the same time, the unique characteristics of the scenario individuality are also learned.

[0021] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understandable through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present disclosure will become more apparent:

[0023] Figure 1 is an exemplary system architecture diagram to which an embodiment of the present disclosure can be applied;

[0024] Figure 2 is a flowchart of an embodiment of the method for training a prediction model according to the present disclosure;

[0025] Figure 3 is a schematic diagram of the network structure of the prediction model according to the present disclosure;

[0026] Figure 4 is a flowchart of an embodiment of the method for predicting click-through rate according to the present disclosure;

[0027] Figure 5 is a schematic structural diagram of an embodiment of the apparatus for training a prediction model according to the present disclosure;

[0028] Figure 6 is a schematic structural diagram of an embodiment of the apparatus for predicting click-through rate according to the present disclosure;

[0029] Figure 7 is a schematic structural diagram of a computer system of an electronic device suitable for implementing the embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The present disclosure will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than to limit the invention. Additionally, it should be noted that for the sake of convenience of description, only the parts related to the relevant invention are shown in the drawings.

[0031] It should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments may be combined with each other. The following will detail the present disclosure with reference to the accompanying drawings and in conjunction with the embodiments.

[0032] Figure 1 An exemplary system architecture 100 is shown that can apply the method for training a prediction model, the apparatus for training a prediction model, the method for predicting click-through rate, or the apparatus for predicting click-through rate according to the embodiments of the present disclosure.

[0033] As Figure 1 shown, the system architecture 100 may include terminals 101, 102, a network 103, a database server 104, and a server 105. The network 103 is used to provide a medium for communication links between the terminals 101, 102, the database server 104, and the server 105. The network 103 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0034] The user 110 may use the terminals 101, 102 to interact with the server 105 through the network 103 to receive or send messages, etc. Various client applications may be installed on the terminals 101, 102, such as model training applications, click-through rate prediction applications, shopping applications, payment applications, web browsers, and instant messaging tools, etc.

[0035] The terminals 101, 102 here may be hardware or software. When the terminals 101, 102 are hardware, they may be various electronic devices with a display screen, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), laptop portable computers, and desktop computers, etc. When the terminals 101, 102 are software, they may be installed in the above-listed electronic devices. It may be implemented as multiple software or software modules (for example, to provide distributed services), or it may be implemented as a single software or software module. No specific limitation is made here.

[0036] The database server 104 may be a database server that provides various services. For example, a sample set may be stored in the database server. The sample set contains a large number of samples. Among them, the samples may include user features, commodity features, context features, scenario data, and the expected values of at least one target. In this way, the user 110 may also select samples from the sample set stored in the database server 104 through the terminals 101, 102.

[0037] Server 105 can also be a server that provides various services, such as a background server that supports various applications displayed on terminals 101 and 102. The background server can use the samples in the sample sets sent by terminals 101 and 102 to train the initial model, and can use the training results (such as the generated prediction model) for information recommendation. In this way, the website service provider can apply the prediction model to predict the click-through rate of commodity information or the conversion rate, so as to recommend high-click-through-rate or high-conversion-rate recommendation information to users, thereby improving the hit rate and conversion rate of the recommendation information.

[0038] Here, the database server 104 and the server 105 can also be hardware or software. When they are hardware, they can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When they are software, they can be implemented as multiple software or software modules (such as those used to provide distributed services), or as a single software or software module. No specific limitation is made here. The database server 104 and the server 105 can also be servers of a distributed system, or servers combined with blockchain. The database server 104 and the server 105 can also be cloud servers, or intelligent cloud computing servers or intelligent cloud hosts with artificial intelligence technology.

[0039] It should be noted that the method for training a prediction model or the method for predicting the click-through rate provided by the embodiments of the present disclosure is generally executed by the server 105. Correspondingly, the device for training a prediction model or the device for predicting the click-through rate is generally also set in the server 105.

[0040] It should be pointed out that in the case where the server 105 can implement the relevant functions of the database server 104, the database server 104 may not be provided in the system architecture 100.

[0041] It should be understood that Figure 1 the numbers of terminals, networks, database servers, and servers in

[0042] are merely illustrative. According to the implementation requirements, there can be any number of terminals, networks, database servers, and servers. Figure 2 Continuing to refer to

[0043] Step 201, obtain samples and an initial prediction model.

[0044] In this embodiment, the execution subject of the method for training a prediction model (for example, Figure 1The server shown can obtain samples. Among them, the samples include user features, product features, context features, scenario data, and the expected values of at least one target. The user features, product features, and context features can be respectively extracted from user information, product information, and context information through LSTM. The user information can include information such as user ID, gender, age, geographical location, etc. The product information can include information such as the price, brand, category, attributes, etc. of the product to be recommended. The context information can include information related to the environment such as time, weather, promotions, etc. The scenario data is used to distinguish the scenario to which the product belongs (for example, page display location, new and old user accounts, age distribution of the purchasing group, geographical location distribution, etc.). These features can be used to describe factors such as users, products, and the environment to improve the recommendation effect. There can be multiple targets, including click-through rate and conversion rate. The expected values of the targets are used to identify whether the product is clicked (the expected value of the click-through rate is 1), and whether it is purchased after being clicked (the expected value of the conversion rate is 1).

[0045] As Figure 3 shown, the prediction model may include an embedding layer, a progressive hierarchical extraction layer, a star network layer, an output layer, and an auxiliary layer.

[0046] Step 202: Input the user features, product features, and context features in the sample into the feature input layer and summarize them through the embedding layer to obtain fused features.

[0047] In this embodiment, for the numerical type features in the user features, product features, and context features, they can be directly passed through to the embedding layer (Embedding layer), or be converted into categorical features through binning features in the input layer and then input into the embedding layer. The categorical features need embedding lookup (embedding vector query) processing, and the embedding dimension is set, which is generally used as a hyperparameter. The input embedding layer summarizes all the input features to obtain fused features.

[0048] Step 203: Input the fused features into the progressive hierarchical extraction layer and output common information.

[0049] In this embodiment, the progressive hierarchical extraction layer (PLE, Progressive Layered Extraction) includes two types of expert networks. One type of network is a shared network, and the second type is a separate network according to different targets. The shared network obtains information from the shared network and other expert networks. The main purpose is to obtain the common information of each expert network, including the hidden common information between each scenario and the common information between each target. The expert network can obtain information from its own network and the shared network, that is, obtain the unique information and common information of each scenario.

[0050] Step 204: Input the scenario data in the sample into the auxiliary layer and output scenario features.

[0051] In this embodiment, the auxiliary layer is a deep learning network. The main function of the auxiliary layer is to obtain data of scene features. A separate network is built instead of placing it in the feature input layer because if it is placed in the feature input layer, with the deepening of network learning, the characteristics of scene features will be lost, and the influence of scene features will be very small when output finally. It is combined with the STAR network and serves as the last output layer.

[0052] Step 205: Input the common information and scene features into the star network layer to learn the interaction effects between networks, and output the predicted value of each target through the output layer.

[0053] In this embodiment, the main objective of the star network layer (STAR network) is to obtain information of each scene. At the same time, a shared-weight network interacts with each scene network to learn the interaction effects between networks. In the star network, for each FC (fully connected) layer, there are central shared parameters and scene-specific private parameters, and the final parameters of each scene are obtained through element-wise product of the two. Through this implementation method, the shared parameters of STAR are updated by the gradients of all scene samples to learn scene commonalities, while the scene-specific private parameters are only updated by samples within the scene to model scene differences. This application uses one model to solve the problem of the interaction effects among multiple scenes. Through this network structure, the content of scene interaction or commonalities can be learned, and at the same time, the unique characteristics of scene individuality can also be learned. Therefore, this application solves the problem that maintaining one model for scenes with a large data volume and scenes with a small data volume will result in insufficient learning for scenes with a small data volume.

[0054] Another difference from the existing STAR network is that the PN layer is not used. One reason is that the data of each batch in the PN layer needs to come from one network, which is more troublesome to process. The data of each scene cannot enter the network simultaneously, that is, they cannot be trained simultaneously, and the time consumption will also increase. An adjustment of this network is to transfer this part of the work to the normalization part, which can reduce the working cost. At the same time, the data of each scene can be trained simultaneously, improving the training speed.

[0055] The output layer can set one or more targets according to the different targets, and can also be adjusted according to classification tasks or regression tasks. Figure 3 As shown, there are two targets. For example, click-through rate prediction and conversion rate prediction.

[0056] Step 206: Adjust the network parameters of the prediction model based on the difference between the predicted value and the expected value of each target.

[0057] In this embodiment, the model training process of the present application is supervised training. The expected value of the target is used as the supervision signal, the loss value is calculated according to the difference between the predicted value and the expected value of each target, and the network parameters of the prediction model are adjusted by the method of gradient descent.

[0058] The method provided in the above embodiment of the present disclosure performs single-scene separate processing in data processing, replacing the method of processing by scene in the neural network structure, reducing the model training overhead. A unified model for multiple scenes is established, reducing the separate modeling cost and maintenance cost of each scene, and at the same time obtaining the interaction effects between scenes, thereby improving the model effects of each scene.

[0059] In some optional implementation manners of this embodiment, the user features include interest preference features; and the method further includes: obtaining the browsing page records of the user within a predetermined time; training the embedding vectors of each page in the browsing page records by the deep walk method; performing average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user. The interest preference features are graph embedding, and the interest preference features are obtained by pre-training: selecting the browsing page records of the user in the recent 3 months, excluding the pages such as the home page that are entered for the first time, and training the embedding of each page by the deep walk method. The embedding of each user is obtained by the avg pooling of the browsing pages in the recent 3 months. The representation of the embedding of the user browsing path is increased to depict the user's intention in detail.

[0060] In some optional implementation manners of this embodiment, before training the embedding vectors of each page in the browsing page records by the deep walk method, the method further includes: filtering out the pages that are entered for the first time from the browsing page records. Filter out the pages that are frequently entered but cannot reflect the user's preference to prevent misdetection.

[0061] In some optional implementation manners of this embodiment, the performing average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user includes: performing weighted average pooling on the embedding vectors of all pages in the browsing page records according to the preset weights to obtain the interest preference features of the user, where the weights are inversely proportional to the time since browsing. When calculating the avg pooling, time weighting is added, that is, the weights of the pages with a longer time since browsing are smaller, and the pages browsed recently can better reflect the user's intention or interest.

[0062] In some alternative implementation manners of this embodiment, before inputting the scene data in the sample into the auxiliary layer and outputting scene features, the method further includes: normalizing the data of different scenes separately. Since the composition of new and old users, the display positions, or the content forms of different scenes are different, resulting in different click-through rates and conversion rates for each scene, there may be a large difference in the data of each scene. Normalize each scene separately. Different from conventional normalization, the conventional approach is to perform the same normalization on the data of multiple scenes. Normalizing each scene separately can reduce the impact caused by different distributions between scenes. The minimum-maximum normalization or mean normalization method can be used to normalize each scene separately.

[0063] The minimum-maximum normalization uses the following formula to map the original value to the interval [0, 1]:

[0064] Y = (X - MIN) / (MAX - MIN)

[0065] where X is the original data, Y is the normalized data, and MIN and MAX are the minimum and maximum values of the original data, respectively.

[0066] The mean normalization uses the following formula to map the original value to the interval [-1, 1]:

[0067] Y = (X - AVG) / (MAX - MIN)

[0068] where X is the original data, Y is the normalized data, MIN and MAX are the minimum and maximum values of the original data, respectively, and AVG is the average value of the original data.

[0069] The following takes the book and fresh food scenes as examples to explain the difference between scene-based normalization and unified normalization:

[0070] The age distribution of the population in the book scene is mainly in the range of [36, 45], and the age distribution of the population in the fresh food scene is mainly in the range of [25, 35].

[0071] If unified normalization is performed, then after the minimum-maximum normalization, the value of 30 years old is (30 - 25) / (45 - 25) = 0.25, and the value of 40 years old after the minimum-maximum normalization is (40 - 25) / (45 - 25) = 0.75.

[0072] If scene-based normalization is performed, then after the minimum-maximum normalization, the value of 30 years old is (30 - 25) / (35 - 25) = 0.5, and the value of 40 years old after the minimum-maximum normalization is (40 - 36) / (45 - 36) = 0.44.

[0073] It can be seen that there is a huge difference between the results of normalizing each scenario separately and unified normalization. By normalizing each scenario separately in this application, the influence caused by different distributions between scenarios can be reduced.

[0074] Further referring to Figure 4 , which shows a flowchart 400 of an embodiment of a method for predicting click-through rate. The flowchart 400 of the method for predicting click-through rate includes the following steps:

[0075] Step 401, obtain user features, product features of candidate products, and scenario data.

[0076] In this embodiment, user features, scenario features, product features, and statistical features of each candidate product that have been extracted in advance through networks such as LSTM can be directly obtained. These features are the same as those in step 201. Candidate products are materials recalled according to user search terms, etc.

[0077] Step 402, input the user features, product features of candidate products, and scenario data into a prediction model, and output the predicted click-through rate and / or conversion rate.

[0078] In this embodiment, a large number of recalled materials are sorted according to the click-through rate and / or conversion rate to improve the personalized recommendation ability of the recommendation system. The prediction model of this application adopts a prediction model trained according to the method described in process 200, which can not only predict the click-through rate but also predict the conversion rate.

[0079] In the implementation manner of this application, one model is used to solve the influence of the interaction between multiple scenarios. Through this network structure, the content of the mutual influence or commonality of scenarios is learned, and at the same time, the unique characteristics of the individuality of scenarios are learned.

[0080] Further referring to Figure 5 , as an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of an apparatus for training a prediction model. This apparatus embodiment corresponds to Figure 2 the method embodiment shown, and this apparatus can be specifically applied to various electronic devices.

[0081] Such as Figure 5As shown in the figure, the device 500 for training and predicting a model includes: an acquisition unit 501, a fusion unit 502, a common information extraction unit 503, a scenario feature extraction unit 504, a prediction unit 505, and an adjustment unit 506. Among them, the acquisition unit 501 is configured to acquire samples and an initial prediction model, where the samples include user features, commodity features, context features, scenario data, and expected values of at least one target, and the prediction model includes a feature input layer, an embedding layer, a progressive hierarchical extraction layer, a star network layer, an output layer, and an auxiliary layer; the fusion unit 502 is configured to input the user features, commodity features, and context features in the samples into the feature input layer and, after summarization through the embedding layer, obtain fused features; the common information extraction unit 503 is configured to input the fused features into the progressive hierarchical extraction layer and output common information; the scenario feature extraction unit 504 is configured to input the scenario data in the samples into the auxiliary layer and output scenario features; the prediction unit 505 is configured to input the common information and the scenario features into the star network layer to learn the interaction effects between the networks and output prediction values of each target through the output layer; the adjustment unit 506 is configured to adjust the network parameters of the prediction model based on the difference between the prediction values and the expected values of each target.

[0082] In this embodiment, the specific processing of the acquisition unit 501, the fusion unit 502, the common information extraction unit 503, the scenario feature extraction unit 504, the prediction unit 505, and the adjustment unit 506 of the device 500 for training and predicting a model can refer to Figure 2 Steps 201 - 206 in the corresponding embodiment.

[0083] In some optional implementation manners of this embodiment, the user features include interest preference features; and the device further includes a pre-training unit (not shown in the figure), which is configured to: acquire the browsing page records of the user within a predetermined time; train the embedding vectors of each page in the browsing page records by means of deep walk; perform average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user.

[0084] In some optional implementation manners of this embodiment, the pre-training unit is further configured to: before training the embedding vectors of each page in the browsing page records by means of deep walk, filter out the pages that are entered for the first time from the browsing page records.

[0085] In some optional implementation manners of this embodiment, the pre-training unit is further configured to: perform weighted average pooling on the embedding vectors of all pages in the browsing page records according to preset weights to obtain the interest preference features of the user, where the weights are inversely proportional to the time elapsed since browsing.

[0086] In some alternative implementation manners of this embodiment, the apparatus 500 further includes a normalization unit (not shown in the drawings), configured to normalize the data of different scenarios respectively before inputting the scenario data in the sample into the auxiliary layer and outputting scenario features.

[0087] Further referring to Figure 6 , as an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of an apparatus for predicting click-through rate. This apparatus embodiment corresponds to Figure 4 the method embodiment shown, and this apparatus can be specifically applied to various electronic devices.

[0088] As shown in Figure 6 , the apparatus 600 for predicting click-through rate in this embodiment includes: an obtaining unit 601, configured to obtain user features, product features of candidate products, and scenario data; a prediction unit 602, configured to input the user features, product features of candidate products, and scenario data into a prediction model trained by the apparatus 500, and output the predicted click-through rate and / or conversion rate.

[0089] It should be noted that in the technical solution of the present disclosure, in aspects such as the collection, acquisition, update, analysis, processing, use, transmission, and storage of user personal information, it complies with the provisions of relevant laws and regulations, is used for legal purposes, and does not violate public order and good customs. Necessary measures are taken for user personal information to prevent illegal access to user personal information data, and to maintain the security of user personal information, network security, and national security.

[0090] According to the embodiments of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

[0091] An electronic device includes: one or more processors; a storage device, on which one or more computer programs are stored. When the one or more computer programs are executed by the one or more processors, the one or more processors implement the method described in process 200 or 400.

[0092] A computer-readable medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the method described in process 200 or 400.

[0093] Figure 7FIG. shows a schematic block diagram of an exemplary electronic device 700 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0094] As Figure 7 shown, the device 700 includes a computing unit 701 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0095] Multiple components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0096] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 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, etc. The computing unit 701 executes the various methods and processes described above, such as the road area planning method. For example, in some embodiments, the road area planning method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the road area planning method described above can be executed. Alternatively, in other embodiments, the computing unit 701 can be configured to execute the road area planning method in any other suitable manner (e.g., by means of firmware).

[0097] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0098] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program code is executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0099] In the context of this disclosure, a machine-readable medium can 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 can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the 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 portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0100] In order to provide interaction with a user, the systems and techniques described herein 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 the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the 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 input, voice input, or tactile input).

[0101] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend 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: a local area network (LAN), a wide area network (WAN), and the Internet.

[0102] A computer system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is created by computer programs running on respective computers and having a client-server relationship with each other. The server may be a server of a distributed system or a server incorporating a blockchain. The server may also be a cloud server or an intelligent cloud computing server or an intelligent cloud host with artificial intelligence technology. The server may be a server of a distributed system or a server incorporating a blockchain. The server may also be a cloud server or an intelligent cloud computing server or an intelligent cloud host with artificial intelligence technology.

[0103] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added or deleted. For example, the steps recited in this disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this is not limited herein.

[0104] The above specific embodiments do not constitute a limitation on the protection scope of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A method for training a prediction model, comprising: obtaining a sample and an initial prediction model, wherein the sample includes user features, product features, context features, scenario data, and the expected value of at least one target, and the prediction model includes a feature input layer, an embedding layer, a progressive hierarchical extraction layer, a star network layer, an output layer, and an auxiliary layer; inputting the user features, product features, and context features in the sample into the feature input layer and obtaining a fused feature after summarization through the embedding layer; inputting the fused feature into the progressive hierarchical extraction layer and outputting common information; inputting the scenario data in the sample into the auxiliary layer and outputting scenario features; inputting the common information and the scenario features into the star network layer to learn the interaction effects between the networks, and outputting the predicted value of each target through the output layer; adjusting the network parameters of the prediction model based on the difference between the predicted value and the expected value of each target.

2. The method according to claim 1, wherein, the user features include interest preference features; and the method further comprises: obtaining the browsing page records of the user within a predetermined time; training the embedding vectors of each page in the browsing page records by means of deep walk; performing average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user.

3. The method according to claim 2, wherein, before training the embedding vectors of each page in the browsing page records by means of deep walk, the method further comprises: filtering out the pages entered for the first time from the browsing page records.

4. The method according to claim 2, wherein, performing average pooling on the embedding vectors of all pages in the browsing page records to obtain the interest preference features of the user, comprising: performing weighted average pooling on the embedding vectors of all pages in the browsing page records according to preset weights to obtain the interest preference features of the user, wherein the weights are inversely proportional to the time elapsed since browsing.

5. The method according to claim 1, wherein, before inputting the scenario data in the sample into the auxiliary layer and outputting scenario features, the method further comprises: normalizing the data of different scenarios respectively.

6. A method for predicting click-through rate, comprising: obtaining user features, product features of candidate products, and scenario data; inputting the user features, product features of candidate products, and scenario data into the prediction model trained by the method according to any one of claims 1-5, and outputting the predicted click-through rate and / or conversion rate.

7. An apparatus for training a prediction model, comprising: an obtaining unit configured to obtain a sample and an initial prediction model, wherein the sample includes user features, product features, context features, scenario data, and the expected value of at least one target, and the prediction model includes a feature input layer, an embedding layer, a progressive hierarchical extraction layer, a star network layer, an output layer, and an auxiliary layer; a fusion unit configured to input the user features, product features, and context features in the sample into the feature input layer and obtain a fused feature after summarization through the embedding layer; A common information extraction unit, configured to input the fusion features into the progressive hierarchical extraction layer and output common information; A scene feature extraction unit, configured to input the scene data in the sample into the auxiliary layer and output scene features; A prediction unit, configured to input the common information and the scene features into the star network layer to learn the interaction effects between the networks, and output the predicted values of each target through the output layer; An adjustment unit, configured to adjust the network parameters of the prediction model based on the difference between the predicted value and the expected value of each target.

8. A device for predicting click-through rate, comprising: An acquisition unit, configured to acquire user features, product features of candidate products, and scene data; A prediction unit, configured to input the user features, product features of candidate products, and scene data into a prediction model trained by the method according to any one of claims 1-5, and output the predicted click-through rate and / or conversion rate.

9. An electronic device, comprising: One or more processors; A storage device, on which one or more computer programs are stored, When the one or more computer programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1-6.

10. A computer-readable medium, on which a computer program is stored, wherein, When the computer program is executed by a processor, the method according to any one of claims 1-6 is implemented.

Images