Recommendation model training method, device, electronic device and storage medium

Abstract

The present disclosure provides a recommendation model training method that can be used in the fields of artificial intelligence and finance. The method includes: obtaining training data, wherein the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes; determining a label sequence of the training data based on the user behavior sequence and the transaction attributes; and using the training data and the label sequence of the training data to train an initial model to obtain a target recommendation model. In addition, the present disclosure also provides a recommendation model training device, an electronic device, a readable storage medium, and a computer program product.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence and finance, and more specifically, to a recommendation model training method, a recommendation model training device, an electronic device, a readable storage medium, and a computer program product. Background Art

[0002] With the rapid development of the Internet, the online business of related institutions has gradually increased, and online marketing activities have also become one of the important means to attract users.

[0003] In the process of realizing the concept of the present disclosure, the inventors found that displaying content on a display position is one of the important ways to recommend content. Summary of the invention

[0004] In view of this, the present disclosure provides a recommendation model training method, a recommendation model training device, an electronic device, a readable storage medium and a computer program product.

[0005] One aspect of the present disclosure provides a recommendation model training method, comprising: acquiring training data, wherein the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes; determining a label sequence of the training data based on the user behavior sequence and the transaction attributes; and training an initial model using the training data and the label sequence of the training data to obtain a target recommendation model.

[0006] According to an embodiment of the present disclosure, the number of bits of the above-mentioned user behavior sequence is equal to the number of content nodes, and the above-mentioned user behavior sequence corresponds one-to-one to the above-mentioned content nodes; the above-mentioned method also includes: determining the above-mentioned user behavior sequence according to the user's access to each of the above-mentioned content nodes within the above-mentioned preset time range; wherein, for each of the above-mentioned content nodes, when the above-mentioned user accesses the above-mentioned content node, the value corresponding to the above-mentioned content node in the above-mentioned user behavior list is determined to be 1; when the above-mentioned user does not access the above-mentioned content node, the value corresponding to the above-mentioned content node in the above-mentioned user behavior list is determined to be 0.

[0007] According to an embodiment of the present disclosure, the above-mentioned determining the label sequence of the above-mentioned training data based on the above-mentioned user behavior sequence and the above-mentioned transaction attributes includes: constructing a first score sequence of the above-mentioned training data based on the above-mentioned user behavior sequence; constructing a second score sequence of the above-mentioned training data based on the above-mentioned transaction attributes; performing weighted summation on the above-mentioned user behavior sequence, the above-mentioned first score sequence and the above-mentioned second score sequence to obtain a third score sequence of the above-mentioned training data; and processing the above-mentioned third score sequence according to preset rules to obtain the label sequence of the above-mentioned training data.

[0008] According to an embodiment of the present disclosure, the above-mentioned content nodes include N preset content nodes, wherein the above-mentioned N is a positive integer; wherein the above-mentioned construction of the first score sequence of the above-mentioned training data based on the above-mentioned user behavior sequence includes: based on the above-mentioned user behavior sequence, determining the number M of preset content nodes in the content nodes visited by the above-mentioned user, wherein the above-mentioned M is a positive integer less than or equal to the above-mentioned N; generating an initial sequence with the same length as the length of the above-mentioned user behavior sequence; and using the ratio of the above-mentioned M to the above-mentioned N to replace the numerical value in the above-mentioned initial sequence to obtain the above-mentioned first score sequence.

[0009] According to an embodiment of the present disclosure, the above-mentioned transaction attributes include the transaction values ​​of the above-mentioned users at each of the above-mentioned content nodes within the above-mentioned preset time range; wherein, the above-mentioned second score sequence of the above-mentioned training data constructed based on the above-mentioned transaction attributes includes: determining the intermediate income of each of the above-mentioned content nodes based on the transaction values ​​of each of the above-mentioned content nodes and the yield rate of each of the above-mentioned content nodes; generating an initial sequence with the same length as the length of the above-mentioned user behavior sequence; and using the ratio of the above-mentioned intermediate income to the preset income to replace the value in the above-mentioned initial sequence to obtain the above-mentioned second score sequence.

[0010] According to an embodiment of the present disclosure, the third score sequence is processed according to a preset rule to obtain the label sequence of the training data, including: for each sequence value of the third score sequence, when the sequence value is greater than a preset threshold, setting the sequence value to 1; when the sequence value is less than or equal to the preset threshold, setting the sequence value to 0; and traversing all sequence values ​​in the third score sequence to obtain the label sequence.

[0011] According to an embodiment of the present disclosure, the above-mentioned use of the above-mentioned training data and the label sequence of the above-mentioned training data to train the initial model includes: classifying the above-mentioned training data into a positive sample set and a negative sample set according to the above-mentioned label sequence, wherein the above-mentioned negative sample set includes the above-mentioned training data in which all numerical values ​​in the above-mentioned label sequence are 0; adjusting the number of samples in the above-mentioned positive sample set or the above-mentioned negative sample set so that the number of samples in the adjusted positive sample set and the number of samples in the adjusted negative sample set reach a preset ratio; and alternately inputting the samples in the above-mentioned adjusted positive sample set and the samples in the above-mentioned adjusted negative sample set into the above-mentioned initial model to train the above-mentioned initial model.

[0012] Another aspect of the present disclosure provides a recommendation model training device, including an acquisition module, a labeling module and a training module. The acquisition module is used to acquire training data, wherein the training data includes user attributes, user behavior sequences within a preset time range and transaction attributes; the labeling module is used to determine the label sequence of the training data based on the user behavior sequence and the transaction attributes; and the training module is used to train an initial model using the training data and the label sequence of the training data to obtain a target recommendation model.

[0013] Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more instructions, wherein when the one or more instructions are executed by the one or more processors, the one or more processors implement the method as described above.

[0014] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the above method when executed.

[0015] Another aspect of the present disclosure provides a computer program product, which includes computer executable instructions, and when the instructions are executed, are used to implement the method as described above.

[0016] According to the embodiments of the present disclosure, the acquired user historical behavior data and transaction data are used as training data, and the training data are annotated according to the user historical behavior data and transaction data, and then the annotated training data is used to train the recommendation model. By annotating the training data with data of different dimensions, the technical problem of not being able to optimize the value of user traffic in non-direct bidding scenarios is at least partially overcome, effectively improving the user experience and ensuring the intermediate income of the platform provider. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0018] Figure 1 An exemplary system architecture 100 to which a recommendation model training method can be applied according to an embodiment of the present disclosure is schematically shown;

[0019] Figure 2 A flowchart of a recommendation model training method according to an embodiment of the present disclosure is schematically shown;

[0020] Figure 3A A schematic diagram of the structure of training data according to an embodiment of the present disclosure is schematically shown;

[0021] Figure 3BA schematic diagram of the structure of a label sequence of training data according to an embodiment of the present disclosure is schematically shown;

[0022] Figure 4 A schematic diagram schematically shows an initial model training process according to an embodiment of the present disclosure;

[0023] Figure 5 A block diagram of a recommendation model training device according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 6 A block diagram of an electronic device suitable for implementing a recommendation model training method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0025] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0026] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0027] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0028] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0029] With the development of the Internet, the number of services that users can choose when handling business online has also increased. Therefore, in order to increase business volume, different business providers need to promote their business content online, which leads to the phenomenon that the same booth is occupied by different demand parties at the same time.

[0030] In related technologies, the Real-Time Bidding (RTB) strategy is often used. After evaluating a user's display behavior, a target service provider is selected from a number of service providers, and the service content of the service provider with the highest bid among the target service providers is added to the booth for promotion. This strategy reduces the operating costs of service providers, increases the profits of related institutions, and ensures the user experience.

[0031] However, when the service provider is an internal department of a relevant organization, the bid of the service provider when bidding for the booth is generally equal to 0, and the use of a real-time bidding strategy cannot select the most optimized service provider.

[0032] In view of this, an embodiment of the present disclosure provides a method for training a recommendation model for business content recommendation. The trained recommendation model is used to predict user behavior, which can display appropriate business content to users, improve user experience, and ensure the profits of related organizations.

[0033] Specifically, the embodiments of the present disclosure provide a recommendation model training method, a recommendation model training device, an electronic device, a readable storage medium, and a computer program product. The method includes: obtaining training data, wherein the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes; determining a label sequence of the training data based on the user behavior sequence and the transaction attributes; and using the training data and the label sequence of the training data to train an initial model to obtain a target recommendation model.

[0034] In the technical solution disclosed in the present invention, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations, necessary confidentiality measures are taken, and do not violate public order and good morals.

[0035] The recommendation model training method and device provided by the embodiments of the present disclosure can be used in the field of artificial intelligence or finance. For example, in online banking business, the recommendation model training method and device provided by the embodiments of the present disclosure can be used to select appropriate content and place it in each booth in the program interface to maximize the intermediate income of the bank. In addition, the recommendation model training method and device provided by the embodiments of the present disclosure can also be used in any other field besides the field of artificial intelligence and finance, and the application field of the embodiments of the present disclosure is not limited.

[0036] Figure 1 The exemplary system architecture 100 to which the recommendation model training method can be applied according to an embodiment of the present disclosure is schematically shown. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0037] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0038] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc.

[0039] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0040] The server 105 may be a server that provides various services, such as a background management server that provides support for data processing requests sent by users using the terminal devices 101, 102, and 103. The background management server may extract the data to be processed from the corresponding database based on the received data processing request, process the data, and then feed back the processing results to the terminal device.

[0041] It should be noted that the recommendation model training method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the recommendation model training device provided in the embodiment of the present disclosure can generally be set in the server 105. The recommendation model training method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Correspondingly, the recommendation model training device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Alternatively, the recommendation model training method provided in the embodiment of the present disclosure can also be executed by the terminal devices 101, 102, or 103, or can also be executed by other terminal devices different from the terminal devices 101, 102, or 103. Correspondingly, the recommendation model training device provided in the embodiment of the present disclosure may also be arranged in the terminal device 101 , 102 , or 103 , or in other terminal devices different from the terminal device 101 , 102 , or 103 .

[0042] For example, the user log generated after the user session ends is generally stored in a database, which can be accessed by the server 105 or any one of the terminal devices 101, 102, 103 through a network link. After obtaining the training data from the database, the server 105 or any one of the terminal devices 101, 102, 103 can locally execute the recommendation model training method provided by the embodiment of the present disclosure, or can send the training data to other terminal devices, servers or server clusters, and the other terminal devices, servers or server clusters that receive the training data can execute the recommendation model training method provided by the embodiment of the present disclosure.

[0043] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0044] Figure 2 The flowchart of the recommendation model training method according to an embodiment of the present disclosure is schematically shown.

[0045] like Figure 2As shown, the method includes operations S210 to S230.

[0046] It should be noted that, unless it is explicitly stated that there is a sequence of execution between different operations shown in the flowchart in the embodiments of the present disclosure, or there is a sequence of execution between different operations in technical implementation, otherwise, the execution order of multiple operations may not be prioritized, and multiple operations may also be executed simultaneously.

[0047] In operation S210, training data is acquired, where the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes.

[0048] In operation S220 , a label sequence of training data is determined based on the user behavior sequence and the transaction attributes.

[0049] In operation S230 , the initial model is trained using the training data and the label sequence of the training data to obtain a target recommendation model.

[0050] According to an embodiment of the present disclosure, training data may be extracted from user logs. In order to improve the reliability of the model, the training data may be extracted from user logs of multiple users within the past week.

[0051] According to an embodiment of the present disclosure, user attributes may include relevant information pre-stored when the user registers.

[0052] According to an embodiment of the present disclosure, the preset time range may include a period of time after the user starts a session and accepts content display. For example, the preset time range may be one hour after the user accepts the relevant advertisement display. In the case where the user's session duration is less than the preset time range, the preset time range may be set to the user's session duration. In the case where the user's session duration is greater than the preset time range, the session is segmented, and multiple training data may be obtained from the user log of the session.

[0053] According to an embodiment of the present disclosure, the user behavior sequence may be an array, each element in the array corresponds to a specific service content, and changes in each element in the array may represent the user's access to the specific service.

[0054] According to an embodiment of the present disclosure, a transaction attribute may be an array, each element in the array corresponds to a specific business content, and the value of each element in the array may represent the transaction value of the user under the specific business.

[0055] According to an embodiment of the present disclosure, the label sequence of the training data can be an array composed of "0" and "1", each element in the array corresponds to a specific business content, and the value of each element in the array can be a comprehensive evaluation value based on the user's actual selection and intermediate benefits.

[0056] According to an embodiment of the present disclosure, the initial model may adopt the structure of an existing model such as a click-through rate model (DeepFm).

[0057] According to an embodiment of the present disclosure, loss functions such as Log loss and L2 loss may be used when training the initial model, and the iteration of the model may be performed by methods such as stochastic gradient descent.

[0058] According to the embodiments of the present disclosure, the acquired user historical behavior data and transaction data are used as training data, and the training data are annotated according to the user historical behavior data and transaction data, and then the annotated training data is used to train the recommendation model. By annotating the training data with data of different dimensions, the technical problem of not being able to optimize the value of user traffic in non-direct bidding scenarios is at least partially overcome, effectively improving the user experience and ensuring the intermediate income of the platform provider.

[0059] Reference below Figure 3A , Figure 3B and Figure 4 , combined with specific embodiments Figure 2 The method shown is further explained.

[0060] Figure 3A The structural diagram of the training data according to the embodiment of the present disclosure is schematically shown.

[0061] like Figure 3A As shown, the training data 300 consists of user attributes 310 , user behavior sequence 320 , and transaction attributes 330 .

[0062] According to an embodiment of the present disclosure, the meaning of specific fields in the user attributes 310 may include the user's age, assets, gender, education level, etc. By abstracting part of these user attribute information into discrete numerical values, the user attributes in the training data used for training the model can be obtained.

[0063] For example, for the attribute of age, the real age may not be used as training data. Instead, a list of age groups may be constructed in advance, such as 0-18, 19-24, 25-29, 30-34, etc., and then the ordinal number of the age group to which the real age belongs is selected as the specific value; for example, if the real age is between 0 and 18 years old, the specific value selected is 1; if the real age is between 25 and 29, the specific value selected is 3.

[0064] For another example, for the attribute of assets, when used as training data, the logarithm of the user's actual assets can be taken, and then the logarithm is rounded as the actual value used.

[0065] According to an embodiment of the present disclosure, each element in the user behavior sequence 320 may indicate whether the user has accessed the corresponding content node within a preset time range.

[0066] For example, in a bank's application, multiple content nodes such as "financial management", "insurance", and "credit card" can be pre-buried, and after accessing the content node, you can jump to the corresponding business page; the user behavior sequence 320 can be an array containing only elements "0" and "1", and the user behavior sequence 320 can correspond to multiple content nodes one by one. After the user opens the session of the application, the user's behavior path can be fully tracked to obtain the user's access to each content node within a preset time range. If it is determined that the user has accessed the content node, the value corresponding to the content node in the user behavior sequence 320 can be set to 1; if it is determined that the user has not accessed the content node, the value corresponding to the content node in the user behavior sequence 320 can be set to 0.

[0067] According to an embodiment of the present disclosure, each element in the transaction attribute 330 may represent a transaction value of a user in a business page corresponding to a content node.

[0068] For example, the transaction attribute 330 may also be an array, which corresponds one-to-one to multiple content nodes; for a node with a transaction value, the logarithm of the transaction value is taken and rounded to an integer to obtain the value of the corresponding element in the array.

[0069] Figure 3B A schematic diagram of the structure of a label sequence of training data according to an embodiment of the present disclosure is schematically shown.

[0070] like Figure 3B As shown, the tag sequence 340 can be calculated from the user behavior sequence 320 , the first score sequence 341 , and the second score sequence 342 .

[0071] According to an embodiment of the present disclosure, the first score sequence 341 may be constructed according to the proportion of preset content nodes among the content nodes visited by the user.

[0072] For example, in the content nodes of a banking application, it may include content nodes that do not involve specific services, such as "login", "registration", and "query", and may also include N preset content nodes related to specific services, such as "financial management", "insurance", and "credit card". In a session, based on the user behavior sequence 320 of the session, the number M of preset content nodes in the content nodes visited by the user can be determined, and then As the first score value, an array is constructed according to the first score value, and the constructed array is the first score sequence 341.

[0073] In some embodiments, preset content nodes can be added or deleted arbitrarily.

[0074] For example, in a banking application, the originally set preset content nodes may be the three nodes of "financial management", "insurance", and "credit card". According to the needs of subsequent business, "deposit" in other content nodes may be added to the preset content node to obtain a new preset content node.

[0075] According to an embodiment of the present disclosure, the second score sequence 342 may be constructed according to the transaction value.

[0076] For example, in a bank's application, the transaction value of a specific business corresponding to a content node of a user is α, and the rate of return of the specific business in the application is b. Correspondingly, the intermediate profit that the transaction value can bring to the bank is a×b; thereafter, the intermediate profit can be converted into a value between 0 and 1 according to a preset profit value, and the value is used as the second score value; finally, an array is constructed according to the second score value, and the constructed array is the second score sequence 342.

[0077] According to an embodiment of the present disclosure, the user behavior sequence 320 , the first score sequence 341 , and the second score sequence 342 may be weighted and summed using weights w1 , w2 , and w3 to obtain a third score sequence 343 .

[0078] According to an embodiment of the present disclosure, when selecting weights, they can be set according to the constraint conditions shown in formula (1).

[0079]

[0080] According to an embodiment of the present disclosure, each sequence value in the third score sequence 343 may be rounded to obtain the tag sequence 340 .

[0081] For example, for each sequence value of the third score sequence, if the sequence value is greater than a preset threshold, the sequence value may be set to 1; if the sequence value is less than or equal to the preset threshold, the sequence value may be set to 0.

[0082] According to an embodiment of the present disclosure, the preset threshold may be any number between 0.5 and 1. The closer the value of the preset threshold is to 1, the more it can be considered that the recommendation model is more affected by the intermediate profit.

[0083] Figure 4 A schematic diagram of the initial model training process according to an embodiment of the present disclosure is schematically shown.

[0084] like Figure 4 As shown, the initial model training process includes operations S401 to S405.

[0085] In operation S401 , training data is classified into a positive sample set and a negative sample set.

[0086] In operation S402, it is determined whether the number of samples in the positive sample set and the negative sample set reaches a preset ratio. If the determination result is "no", operation S403 is performed; if the determination result is "yes", operation S404 is performed.

[0087] In operation S403 , the number of samples in the positive sample set or the negative sample set is adjusted.

[0088] In operation S404, positive samples and negative samples are alternately input into the initial model.

[0089] In operation S405, it is determined whether the model training is complete. If it is determined that the model has been fully trained, the target recommendation model is obtained and the training is terminated; if it is determined that the model has not been trained, the method of operation S404 is continued, and the input sample has been trained.

[0090] According to an embodiment of the present disclosure, the training data may be classified according to the label sequence. For example, the training data whose values ​​in the label sequence are all 0 may be added to the negative sample set as negative samples.

[0091] According to an embodiment of the present disclosure, the preset ratio may be set to 1, so that positive samples and negative samples are alternately input into the initial model in a 1:1 manner for training.

[0092] According to an embodiment of the present disclosure, when performing initial model training, hyperparameters such as training rounds, the amount of data for each batch of training data, and a loss value threshold can be set to control the model training process; it can be set that when the current training round reaches the set training round or the loss value is less than the loss value threshold, the model is considered to have been trained.

[0093] According to an embodiment of the present disclosure, by adopting a method of alternately inputting positive samples and negative samples for training, the robustness of the trained model can be effectively improved.

[0094] Figure 5 A block diagram of a recommendation model training device according to an embodiment of the present disclosure is schematically shown.

[0095] like Figure 5 As shown, the recommendation model training device includes an acquisition module 510, a labeling module 520 and a training module 530.

[0096] The acquisition module 510 is used to acquire training data, wherein the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes.

[0097] The labeling module 520 is used to determine the label sequence of the training data based on the user behavior sequence and transaction attributes.

[0098] The training module 530 is used to train the initial model using the training data and the label sequence of the training data to obtain the target recommendation model.

[0099] According to the embodiments of the present disclosure, the acquired user historical behavior data and transaction data are used as training data, and the training data are annotated according to the user historical behavior data and transaction data, and then the annotated training data is used to train the recommendation model. By annotating the training data with data of different dimensions, the technical problem of not being able to optimize the value of user traffic in non-direct bidding scenarios is at least partially overcome, effectively improving the user experience and ensuring the intermediate income of the platform provider.

[0100] According to an embodiment of the present disclosure, the number of bits of the user behavior sequence is equal to the number of content nodes, and the user behavior sequence corresponds one-to-one to the content nodes.

[0101] According to an embodiment of the present disclosure, the recommendation model training device also includes a determination module.

[0102] The determination module is used to determine the user behavior sequence according to the user's access to each content node within a preset time range. For each content node, if the user accesses the content node, the value corresponding to the content node in the user behavior list is determined to be 1; if the user does not access the content node, the value corresponding to the content node in the user behavior list is determined to be 0.

[0103] According to an embodiment of the present disclosure, the labeling module 520 includes a first labeling unit, a second labeling unit, a third labeling unit, and a fourth labeling unit.

[0104] The first labeling unit is used to construct a first score sequence of training data based on the user behavior sequence.

[0105] The second labeling unit is used to construct a second score sequence of the training data based on the transaction attributes.

[0106] The third labeling unit is used to perform weighted summation on the user behavior sequence, the first score sequence, and the second score sequence to obtain a third score sequence of the training data.

[0107] The fourth labeling unit is used to process the third score sequence according to a preset rule to obtain a label sequence of training data.

[0108] According to an embodiment of the present disclosure, the content node includes N preset content nodes, where N is a positive integer.

[0109] According to an embodiment of the present disclosure, the first labeling unit includes a first labeling sub-unit, a second labeling sub-unit and a third labeling sub-unit.

[0110] The first marking subunit is used to determine the number M of preset content nodes in the content nodes visited by the user based on the user behavior sequence, where M is a positive integer less than or equal to N.

[0111] The second labeling subunit is used to generate an initial sequence with the same length as the user behavior sequence.

[0112] The third marking subunit is used to replace the numerical values ​​in the initial sequence with the ratio of M to N to obtain a first score sequence.

[0113] According to an embodiment of the present disclosure, the transaction attribute includes the transaction value of the user at each content node within a preset time range.

[0114] According to an embodiment of the present disclosure, the second labeling unit includes a fourth labeling sub-unit, a fifth labeling sub-unit and a sixth labeling sub-unit.

[0115] The fourth marking subunit is used to determine the intermediate income of each content node based on the transaction value of each content node and the income rate of each content node.

[0116] The fifth labeling subunit is used to generate an initial sequence having the same length as the user behavior sequence.

[0117] The sixth marking subunit is used to replace the numerical values ​​in the initial sequence with the ratio of the intermediate income to the preset income to obtain a second score sequence.

[0118] According to an embodiment of the present disclosure, the fourth labeling unit includes a seventh labeling sub-unit and an eighth labeling sub-unit.

[0119] The seventh marking subunit is used to set the sequence value of each sequence value of the third score sequence to 1 when the sequence value is greater than a preset threshold value; and to set the sequence value to 0 when the sequence value is less than or equal to the preset threshold value.

[0120] The eighth labeling subunit is used to traverse all sequence values ​​in the third score sequence to obtain a label sequence.

[0121] According to an embodiment of the present disclosure, the training module 530 includes a first training unit, a second training unit, and a third training unit.

[0122] The first training unit is used to classify the training data into a positive sample set and a negative sample set according to the label sequence, wherein the negative sample set includes the training data in which all values ​​in the label sequence are 0.

[0123] The second training unit is used to adjust the number of samples in the positive sample set or the negative sample set so that the number of samples in the adjusted positive sample set and the number of samples in the adjusted negative sample set reach a preset ratio.

[0124] The third training unit is used to alternately input the samples in the adjusted positive sample set and the samples in the adjusted negative sample set into the initial model to train the initial model.

[0125] According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be performed.

[0126] For example, any multiple of the acquisition module 510, the annotation module 520, and the training module 530 can be combined in one module / unit / sub-unit for implementation, or any one of the modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units can be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present disclosure, at least one of the acquisition module 510, the annotation module 520, and the training module 530 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware or in an appropriate combination of any of them. Alternatively, at least one of the acquisition module 510, the labeling module 520, and the training module 530 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be performed.

[0127] It should be noted that the recommended model training device part in the embodiments of the present disclosure corresponds to the recommended model training method part in the embodiments of the present disclosure. The description of the recommended model training device part specifically refers to the recommended model training method part, which will not be repeated here.

[0128] Figure 6 A block diagram of an electronic device suitable for implementing a recommendation model training method according to an embodiment of the present disclosure is schematically shown. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0129] like Figure 6 As shown, the computer electronic device 600 according to the embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 602 or the program loaded from the storage part 608 to the random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include an onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present disclosure.

[0130] In RAM 603, various programs and data required for the operation of electronic device 600 are stored. Processor 601, ROM 602 and RAM 603 are connected to each other via bus 604. Processor 601 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 602 and / or RAM 603. It should be noted that the program can also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0131] According to an embodiment of the present disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to the bus 604. The electronic device 600 may further include one or more of the following components connected to the I / O interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 608 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage portion 608 as needed.

[0132] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0133] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0134] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include, but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0135] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 602 and / or the RAM 603 described above and / or one or more memories other than the ROM 602 and the RAM 603 .

[0136] An embodiment of the present disclosure also includes a computer program product, which includes a computer program, which contains program code for executing the method provided by the embodiment of the present disclosure. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the recommendation model training method provided by the embodiment of the present disclosure.

[0137] When the computer program is executed by the processor 601, the above functions defined in the system / device of the embodiment of the present disclosure are executed. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0138] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0139] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0140] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the 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 may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments and / or claims of the present disclosure can be combined and / or combined in a variety of ways, even if such a combination or combination is not explicitly recorded in the present disclosure. In particular, without departing from the spirit and teaching of the present disclosure, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0141] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A recommendation model training method, comprising: Acquiring training data, wherein the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes; Determining a label sequence of the training data based on the user behavior sequence and the transaction attributes; and Using the training data and the label sequence of the training data to train an initial model to obtain a target recommendation model, wherein the value of each element in the label sequence of the training data is determined based on the actual selection of the user and the comprehensive evaluation value of the intermediate income, and the intermediate income is determined based on the transaction value of each content node and the rate of return of each content node; Wherein, determining the label sequence of the training data based on the user behavior sequence and the transaction attribute includes: Constructing a first score sequence of the training data based on the user behavior sequence; constructing a second score sequence of the training data based on the transaction attributes; Performing a weighted summation on the user behavior sequence, the first score sequence, and the second score sequence to obtain a third score sequence of the training data; Processing the third score sequence according to a preset rule to obtain a label sequence of the training data; The number of bits of the user behavior sequence is equal to the number of content nodes, and the user behavior sequence corresponds to the content nodes one by one; The transaction attribute includes the transaction value of the user at each content node within the preset time range.

2. The method according to claim 1, wherein: The method further comprises: Determining the user behavior sequence according to the user's access to each of the content nodes within the preset time range; Among them, for each of the content nodes, when the user visits the content node, the value corresponding to the content node in the user behavior list is determined to be 1; when the user does not visit the content node, the value corresponding to the content node in the user behavior list is determined to be 0.

3. The method according to claim 2, wherein: The content nodes include N preset content nodes, where N is a positive integer; Wherein, constructing the first score sequence of the training data based on the user behavior sequence includes: Based on the user behavior sequence, determine the number M of preset content nodes among the content nodes visited by the user, wherein M is a positive integer less than or equal to N; generating an initial sequence having the same length as the user behavior sequence; and The ratio of the M to the N is used to replace the values ​​in the initial sequence to obtain the first score sequence.

4. The method according to claim 2, wherein: The step of constructing a second score sequence of the training data based on the transaction attributes includes: Determining the intermediate income of each of the content nodes based on the transaction value of each of the content nodes and the yield of each of the content nodes; generating an initial sequence having the same length as the user behavior sequence; and The ratio of the intermediate income to the preset income is used to replace the values ​​in the initial sequence to obtain the second score sequence.

5. The method according to claim 2, wherein: The step of processing the third score sequence according to a preset rule to obtain a label sequence of the training data includes: For each sequence value of the third score sequence, if the sequence value is greater than a preset threshold, the sequence value is set to 1; if the sequence value is less than or equal to the preset threshold, the sequence value is set to 0; and All sequence values ​​in the third score sequence are traversed to obtain the label sequence.

6. The method according to claim 1, wherein: The using the training data and the label sequence of the training data to train the initial model includes: Classifying the training data into a positive sample set and a negative sample set according to the label sequence, wherein the negative sample set includes the training data in which all values ​​in the label sequence are 0; Adjusting the number of samples in the positive sample set or the negative sample set so that the number of samples in the adjusted positive sample set and the number of samples in the adjusted negative sample set reach a preset ratio; and The samples in the adjusted positive sample set and the samples in the adjusted negative sample set are alternately input into the initial model to train the initial model.

7. A recommendation model training device, comprising: An acquisition module, used to acquire training data, wherein the training data includes user attributes, user behavior sequences within a preset time range, and transaction attributes; a labeling module, configured to determine a label sequence of the training data based on the user behavior sequence and the transaction attributes; and A training module, used to train an initial model using the training data and a label sequence of the training data to obtain a target recommendation model, wherein the value of each element in the label sequence of the training data is determined based on the actual selection of the user and a comprehensive evaluation value of the intermediate income, and the intermediate income is determined based on the transaction value of each content node and the rate of return of each content node; Wherein, the annotation module is configured as follows: Constructing a first score sequence of the training data based on the user behavior sequence; constructing a second score sequence of the training data based on the transaction attributes; Performing a weighted summation on the user behavior sequence, the first score sequence, and the second score sequence to obtain a third score sequence of the training data; Processing the third score sequence according to a preset rule to obtain a label sequence of the training data; The number of bits of the user behavior sequence is equal to the number of content nodes, and the user behavior sequence corresponds to the content nodes one by one; The transaction attribute includes the transaction value of the user at each content node within the preset time range.

8. An electronic device comprising: one or more processors; a memory for storing one or more instructions, Wherein, when the one or more instructions are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having executable instructions stored thereon, wherein when the executable instructions are executed by a processor, the processor is enabled to implement the method according to any one of claims 1 to 6.

10. A computer program product, comprising computer executable instructions, which are used to implement the method of any one of claims 1 to 6 when executed.

Images