Article recommendation black box model optimization method and device, equipment, medium and product

Abstract

The embodiment of the invention discloses an article recommendation black box model optimization method and device, equipment, a medium and a product. A specific embodiment of the method comprises the following steps: sampling article identifiers from an article identifier set to form a sampling set; executing an iteration step: determining actual response ranking information of each article identifier in the sampling set based on an article recommendation black box model; training the initial fitting model according to the article identifier set, the sampling set, the actual response ranking information and the initial model parameter information; through the updated initial model parameter information, screening an article identifier meeting a preset optimization condition from the article identifier set; adding the screened article identifiers to a sampling set; and in response to the fact that the execution frequency is equal to the preset iteration frequency, determining the at least one article identifier meeting the preset ranking condition as a recommended article identifier set. The implementation mode is related to big data, and the practicability of optimizing the article recommendation black box model and the accuracy of the article recommendation black box model are improved.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and specifically to methods, apparatus, devices, media, and products for optimizing black-box models of item recommendations. Background Technology

[0002] With the gradual development of online product information platforms, the number of online products is increasing. Therefore, product recommendations need to be tailored to users. When making product recommendations, black-box models can be used, and stochastic process models can be employed to optimize these models. Currently, the common approach to optimizing black-box models for product recommendations is to use a Gaussian process as a surrogate model to predict the response of the black-box model. The Gaussian process provides the specific value of the response (e.g., a product recommendation rating).

[0003] However, the inventors discovered that when using the above method, the following technical problems often exist: the Gaussian process gives a specific value of its response, which is an absolute response. The absolute response value may be difficult to obtain in practice, resulting in poor practicality of the optimized item recommendation black box model. Furthermore, the absolute response method is highly sensitive to noise, resulting in low accuracy of the final optimized item recommendation black box model.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure provide methods, apparatus, electronic devices, computer-readable media, and computer program products for optimizing black-box models of item recommendations to solve one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a method for optimizing a black-box model of item recommendation. The method includes: sampling a preset number of item identifiers from a set of item identifiers to form a sampling set; and performing the following iterative steps based on the sampling set and initial model parameter information: determining the actual response ranking information of each item identifier corresponding to the target user information in the sampling set based on the black-box model of item recommendation corresponding to the target user information, thereby obtaining each actual response ranking information; training an initial fitted model according to the item identifier set, the sampling set, each actual response ranking information, and the initial model parameter information to update the initial model parameter information; selecting item identifiers that meet preset optimization conditions from the item identifier set using the updated initial model parameter information; adding the selected item identifiers to the sampling set to update the sampling set; and, in response to determining that the number of executions of the iterative steps equals the preset number of iterations, determining at least one item identifier in the sampling set that meets the preset ranking conditions as a set of recommended item identifiers to optimize the aforementioned black-box model of item recommendation.

[0008] Optionally, the above iterative steps further include: in response to the determination that the number of executions of the iterative steps is less than the preset number of iterations, the above iterative steps are executed again based on the updated sample set, the updated initial model parameter information, and the trained initial fitted model.

[0009] Optionally, the above-mentioned black-box model for item recommendation based on corresponding target user information determines the actual response ranking information of each item identifier in the sampling set corresponding to the target user information, including: determining the recommendation value of each item identifier in the sampling set corresponding to the target user information based on the above-mentioned black-box model for item recommendation; for each item identifier in the sampling set, determining the number of each item identifier in the sampling set whose corresponding recommendation value satisfies the preset optimization condition as a metric value, wherein the above-mentioned preset optimization condition corresponds to the recommendation value of the above-mentioned item identifier; and determining the actual response ranking information corresponding to each item identifier in the sampling set based on the determined metric value corresponding to each item identifier in the sampling set.

[0010] Optionally, determining the actual response ranking information corresponding to each item identifier in the sample set based on the measure values ​​corresponding to each item identifier in the sample set includes: sorting each item identifier in the sample set according to the measure values ​​corresponding to each item identifier in the sample set to obtain an item identifier sequence; and for each item identifier in the sample set, determining the position of the item identifier in the item identifier sequence as the actual response ranking information corresponding to the item identifier.

[0011] Optionally, the above-mentioned training of the initial fitting model based on the item identifier set, the sampling set, the actual response ranking information, and the initial model parameter information to update the initial model parameter information includes: for each item identifier in the sampling set, determining the response ranking probability information corresponding to the actual response ranking information of the item identifier based on the actual response ranking information of the item identifier, the item identifier set, and the initial model parameter information; and training the initial fitting model based on the determined response ranking probability information to update the initial model parameter information.

[0012] Optionally, the above-mentioned filtering of item identifiers that meet preset optimization conditions from the item identifier set using updated initial model parameter information includes: combining item identifiers in the item identifier set that are different from each item identifier in the sampling set into a set of item identifiers to be filtered; for each item identifier to be filtered in the set of item identifiers to be filtered, generating a sampled value corresponding to the item identifier to be filtered based on the updated initial model parameter information; and determining the item identifiers in the set of item identifiers to be filtered whose corresponding sampled values ​​meet the preset minimum value condition as item identifiers that meet the preset optimization conditions.

[0013] Optionally, generating the collected value corresponding to the item identifier to be filtered based on the updated initial model parameter information includes: generating a first comparison value based on the sampling set and a first preset coefficient; generating a second comparison value based on the item identifier to be filtered and the updated initial model parameter information; in response to determining that the second comparison value is less than the first comparison value, generating the collected value corresponding to the item identifier to be filtered based on a preset collection function; and in response to determining that the second comparison value is greater than or equal to the first comparison value, selecting a collected value from a preset collection value range as the collected value corresponding to the item identifier to be filtered.

[0014] Optionally, the above-mentioned generation of the collected value corresponding to the item identifier to be filtered based on the updated initial model parameter information includes: based on the initial response ranking, performing the following iterative steps: generating response ranking probability information of the initial response ranking corresponding to the item identifier to be filtered according to the updated initial model parameter information; determining the difference between the preset response ranking and the initial response ranking as the ranking difference; determining the ranking improvement expectation as the product of the ranking difference and the response ranking probability information; performing auto-increment processing on the initial response ranking to update the initial response ranking; in response to determining that the updated initial response ranking is less than the preset response ranking, performing the above iterative steps again; in response to determining that the updated initial response ranking is equal to the preset response ranking, determining the sum of the obtained ranking improvement expectations as the collected value corresponding to the item identifier to be filtered.

[0015] Optionally, the method further includes: sending the item information corresponding to each recommended item identifier in the above-mentioned recommended item identifier set to at least one terminal device corresponding to the above-mentioned target user information; selecting a preset number of recommended item identifiers from the above-mentioned recommended item identifier set as a set of item identifiers to be replenished; for each item identifier to be replenished in the above-mentioned set of item identifiers to be replenished, performing the following steps: determining the replenishment quantity corresponding to the above-mentioned item identifier to be replenished; in response to determining that the above-mentioned replenishment quantity meets the preset replenishment conditions, controlling the item scheduling device corresponding to the above-mentioned item identifier to be replenished to perform an item replenishment operation according to the above-mentioned replenishment quantity.

[0016] Secondly, some embodiments of this disclosure provide an optimization apparatus for a black-box model of item recommendation. The apparatus includes: a sampling unit configured to sample a preset number of item identifiers from a set of item identifiers to form a sampling set; and an execution unit configured to perform the following iterative steps based on the sampling set and initial model parameter information: determining the actual response ranking information of each item identifier corresponding to the target user information in the sampling set based on the black-box model of item recommendation corresponding to the target user information, and obtaining each actual response ranking information; training an initial fitting model according to the set of item identifiers, the sampling set, each actual response ranking information, and the initial model parameter information to update the initial model parameter information; selecting item identifiers that meet preset optimization conditions from the set of item identifiers using the updated initial model parameter information; adding the selected item identifiers to the sampling set to update the sampling set; and, in response to determining that the number of executions of the iterative steps is equal to the preset number of iterations, determining at least one item identifier in the sampling set that meets the preset ranking conditions as a set of recommended item identifiers to optimize the aforementioned black-box model of item recommendation.

[0017] Optionally, the above iterative steps further include: in response to the determination that the number of executions of the iterative steps is less than the preset number of iterations, the above iterative steps are executed again based on the updated sample set, the updated initial model parameter information, and the trained initial fitted model.

[0018] Optionally, the execution unit is further configured to: determine the recommended value of the target user information corresponding to each item identifier in the sampling set based on the above-mentioned black-box model for item recommendation; for each item identifier in the sampling set, determine the number of each item identifier in the sampling set whose recommended value satisfies the preset optimization condition as a metric value, wherein the above-mentioned preset optimization condition corresponds to the recommended value of the above-mentioned item identifier; and determine the actual response ranking information corresponding to each item identifier in the sampling set based on the determined metric value corresponding to each item identifier in the sampling set.

[0019] Optionally, the execution unit is further configured to: sort each item identifier in the sample set according to the measure value corresponding to each item identifier in the determined sample set to obtain an item identifier sequence; and for each item identifier in the sample set, determine the order of the item identifier in the item identifier sequence as the actual response ranking information corresponding to the item identifier.

[0020] Optionally, the execution unit is further configured to: for each item identifier in the sampling set, determine the response ranking probability information corresponding to the actual response ranking information of the item identifier based on the actual response ranking information of the item identifier, the item identifier set, and the initial model parameter information; and train the initial fitting model based on the determined response ranking probability information to update the initial model parameter information.

[0021] Optionally, the execution unit is further configured to: combine the item identifiers in the item identifier set that are different from the item identifiers in the sampling set into a set of item identifiers to be filtered; for each item identifier to be filtered in the set of item identifiers to be filtered, generate the collected value corresponding to the item identifier to be filtered based on the updated initial model parameter information; and determine the item identifiers in the set of item identifiers to be filtered whose corresponding collected values ​​satisfy the preset minimum value condition as item identifiers that satisfy the preset optimization condition.

[0022] Optionally, the execution unit is further configured to: generate a first comparison value based on the sampling set and a first preset coefficient; generate a second comparison value based on the item identifier to be filtered and the updated initial model parameter information; in response to determining that the second comparison value is less than the first comparison value, generate a sampled value corresponding to the item identifier to be filtered based on a preset sampling function; and in response to determining that the second comparison value is greater than or equal to the first comparison value, select a sampled value from a preset sampled value range as the sampled value corresponding to the item identifier to be filtered.

[0023] Optionally, the execution unit is further configured to: based on the initial response ranking, perform the following iterative steps: generate response ranking probability information for the initial response ranking corresponding to the above-mentioned item identifier to be filtered, according to the updated initial model parameter information; determine the difference between the preset response ranking and the initial response ranking as the ranking difference; determine the ranking improvement expectation by multiplying the ranking difference and the response ranking probability information; perform auto-increment processing on the initial response ranking to update the initial response ranking; in response to determining that the updated initial response ranking is less than the above-mentioned preset response ranking, perform the above iterative steps again; in response to determining that the updated initial response ranking is equal to the above-mentioned preset response ranking, determine the sum of the obtained ranking improvement expectations as the collected value corresponding to the above-mentioned item identifier to be filtered.

[0024] Optionally, the item recommendation black-box model optimization device further includes: a sending unit, a selection unit, and a step execution unit. The sending unit is configured to send the item information corresponding to each recommended item identifier in the aforementioned set of recommended item identifiers to at least one terminal device corresponding to the aforementioned target user information. The selection unit is configured to select a preset number of recommended item identifiers from the aforementioned set of recommended item identifiers as a set of item identifiers to be replenished. The step execution unit is configured to, for each item identifier to be replenished in the aforementioned set of item identifiers to be replenished, perform the following steps: determine the replenishment quantity corresponding to the item identifier to be replenished; in response to determining that the replenishment quantity meets preset replenishment conditions, control the item scheduling device corresponding to the item identifier to be replenished to perform an item replenishment operation according to the replenishment quantity.

[0025] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0026] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0027] Fifthly, some embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0028] The above embodiments of this disclosure have the following beneficial effects: the item recommendation black box model optimization method of some embodiments of this disclosure improves the practicality and accuracy of the optimized item recommendation black box model. Specifically, the reason for the poor practicality and low accuracy of the optimized item recommendation black box model is that the Gaussian process gives a specific value of its response, which is an absolute response. The absolute response value may be difficult to obtain in practice, resulting in poor practicality of the optimized item recommendation black box model. Moreover, the absolute response method is highly sensitive to noise, resulting in low accuracy of the final optimized item recommendation black box model. Based on this, the item recommendation black box model optimization method of some embodiments of this disclosure first samples a preset number of item identifiers from the item identifier set to form a sampling set. This avoids using a large number of item identifiers to train the initial fitted model at the beginning. Then, based on the sampling set and the initial model parameter information, the following iterative steps are performed: First, based on the item recommendation black box model corresponding to the target user information, the actual response ranking information of each item identifier corresponding to the target user information in the sampling set is determined, and the actual response ranking information is obtained. Therefore, the relative true response ranking of each point in the sampling points output by the black-box model can be obtained in advance. The second step involves training the initial fitted model based on the item identifier set, the sampling set, the actual response ranking information, and the initial model parameter information, thereby updating the initial model parameter information. This allows training the initial fitted model from the reference dimension of relative ranking. The third step involves selecting item identifiers that meet preset optimization conditions from the item identifier set using the updated initial model parameter information. This allows selecting the optimal query point (i.e., item identifier) ​​to add to the sampling set to optimize the item recommendation black-box model. The fourth step involves adding the selected item identifiers to the sampling set to update the sampling set. The fifth step, in response to the determination that the number of iteration steps equals the preset number of iterations, determines at least one item identifier in the sampling set that meets the preset ranking conditions as the recommended item identifier set to optimize the aforementioned item recommendation black-box model. Thus, after iteration, at least one item identifier in the sampling set that meets the preset ranking conditions can be determined as the final output of the item recommendation black-box model. Because the optimization of the item black-box model did not use absolute response as a reference dimension, but instead used relative ranking to train the initial fitted model, relative rankings are easier to obtain than absolute item recommendation values, thus improving the practicality of the optimized item recommendation black-box model. Furthermore, compared to absolute item recommendation values, the relative ranking relationships between items are less susceptible to noise corruption; therefore, relative rankings are more robust to noise and less sensitive to it, thereby improving the accuracy of the final optimized item recommendation black-box model. Attached Figure Description

[0029] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0030] Figure 1 This is an architecture diagram of an exemplary system to which some embodiments of this disclosure can be applied;

[0031] Figure 2 This is a flowchart of some embodiments of the item recommendation black-box model optimization method according to this disclosure;

[0032] Figure 3 These are flowcharts of some other embodiments of the item recommendation black-box model optimization method according to this disclosure;

[0033] Figure 4 This is a schematic diagram of the structure of some embodiments of the item recommendation black box model optimization device according to the present disclosure;

[0034] Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0035] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0036] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0037] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0038] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0039] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0040] Before performing any of the operations involving the collection, storage, or use of user personal information (such as target user information) disclosed in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, informing personal information subjects, and obtaining prior authorization and consent from personal information subjects.

[0041] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] Figure 1 An exemplary system architecture 100 is shown that can be applied to an item recommendation black-box model optimization method or an item recommendation black-box model optimization apparatus according to some embodiments of the present disclosure.

[0043] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0044] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, online item information applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0045] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with a display screen and supporting web browsing, including but not limited to smartphones, tablets, e-book readers, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are imposed here.

[0046] Server 105 can be a server that provides various services, such as a backend server that supports the information displayed on terminal devices 101, 102, and 103. The backend server can analyze and process received requests and other data, and then feed the processing results back to the terminal devices.

[0047] It should be noted that the item recommendation black-box model optimization method provided in the embodiments of this disclosure can be executed by the server 105. Accordingly, the item recommendation black-box model optimization device can be set in the server 105.

[0048] It should be noted that a server can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0049] Continue to refer to Figure 2 The flowchart 200 illustrates some embodiments of the item recommendation black-box model optimization method according to the present disclosure. The item recommendation black-box model optimization method includes the following steps:

[0050] Step 201: Sample a preset number of item identifiers from the item identifier set to form a sampling set.

[0051] In some embodiments, the execution entity of the item recommendation black-box model optimization method (e.g.) Figure 1 The server shown can sample a preset number of item identifiers from the item identifier set to form a sample set. The item identifier set can represent an item pool. Each item identifier can be a unique code for an item. The preset number can be a pre-defined quantity. In practice, the executing entity can randomly sample a preset number of item identifiers from the item identifier set to form a sample set. The sample set includes the sampled preset number of item identifiers.

[0052] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.

[0053] Step 202: Based on the sample set and initial model parameter information, perform the following iterative steps:

[0054] Step 2021: Based on the black-box model of item recommendation corresponding to the target user information, determine the actual response ranking information of each item identifier corresponding to the target user information in the sampling set, and obtain the actual response ranking information.

[0055] In some embodiments, the execution entity can determine the actual response ranking information of each item identifier corresponding to the target user information in the sampling set based on the item recommendation black-box model corresponding to the target user information, thereby obtaining each actual response ranking information. During the first iteration, the initial model parameter information can be the initialized model parameter information. In subsequent iterations, the initial model parameter information can be the updated initial model parameter information. The model parameter information can be the model parameters of the initially fitted model. For example, the initially fitted model can be a multilayer perceptron. The model parameter information can include the weights of each network layer. The item recommendation black-box model corresponding to the target user information can be a pre-obtained model with an unknown model structure used to generate item recommendation information between item identifiers and the target user information. The item recommendation information can represent the degree of recommendation of the item corresponding to the item identifier to the user corresponding to the target user information, or the degree of interest of the user corresponding to the target user information in the item corresponding to the item identifier. For example, the item recommendation information can be a recommendation score. The target user information can represent a single user or a group of users. Correspondingly, the target user information can include at least one user identifier. In practice, firstly, for each item identifier in the aforementioned sampling set, the executing entity can input the item identifier into the item recommendation black-box model to obtain item recommendation information based on the item identifier and the target user information. Then, the item identifiers in the sampling set can be sorted in descending order according to the corresponding item recommendation information. Finally, the ranking of each sorted item identifier can be determined as the actual ranking information.

[0056] In some optional implementations of certain embodiments, the aforementioned execution entity can determine the actual response ranking information corresponding to the target user information for each item identifier in the sampling set through the following steps based on the item recommendation black-box model corresponding to the target user information:

[0057] The first step is to determine the recommended value for each item identifier in the sample set based on the black-box model of item recommendation corresponding to the target user information. Here, the recommended value can be the item recommendation information output by the black-box model of the item identifier and the target user information. In practice, the executing entity can input the item identifier into the black-box model to obtain the recommended value of the target user information corresponding to the item identifier.

[0058] The second step involves determining, for each item identifier in the sampling set, the number of item identifiers whose recommended values ​​satisfy a preset optimization condition as a metric. This preset optimization condition corresponds to the recommended value of the item identifier. The preset optimization condition can be that the recommended value is greater than the recommended value corresponding to the item identifier. Therefore, the metric values ​​of each item identifier in the sampling set that are superior to a given item identifier can be determined.

[0059] The third step is to determine the actual response ranking information for each item identifier in the sample set based on the measure values ​​corresponding to each item identifier. Therefore, the actual response ranking information for each item identifier can be determined using the measure values ​​of all item identifiers that are superior to the first item identifier.

[0060] In some optional implementations of certain embodiments, the aforementioned execution entity can determine the actual response ranking information corresponding to each item identifier in the sample set by following these steps based on the metric values ​​corresponding to each item identifier in the determined sample set:

[0061] The first step is to sort the item identifiers in the sample set according to the measure values ​​corresponding to each item identifier, thus obtaining an item identifier sequence. In practice, the aforementioned execution entity can sort the item identifiers in the sample set in ascending order based on the measure values ​​corresponding to each item identifier, thereby obtaining an item identifier sequence.

[0062] The second step is to determine the actual response ranking information corresponding to each item identifier in the sample set, based on its position in the item identifier sequence. For example, the initial value of the position can be 0.

[0063] Step 2022: Train the initial fitted model based on the item identifier set, sampling set, ranking information of each actual response, and initial model parameter information, so as to update the initial model parameter information.

[0064] In some embodiments, the execution entity can train an initial fitting model based on the item identifier set, the sampling set, the ranking information of each actual response, and the initial model parameter information to update the initial model parameter information. The initial fitting model can be a multilayer perceptron to be trained. In practice, the execution entity can construct a stochastic process model for response ranking using the item identifier set, the sampling set, the ranking information of each actual response, and the initial model parameter information. For example, the stochastic process model can be a Gaussian process model. Through the stochastic process model, the probability of the ranking of each sampling point in the sampling set can be determined. The probability can be determined by the initial fitting model. Specifically, the initial fitting model can be trained using an optimizer with the objective function of minimizing the log-likelihood.

[0065] In some optional implementations of certain embodiments, the aforementioned execution entity may train an initial fitted model based on the item identifier set, the sampling set, the ranking information of each actual response, and the initial model parameter information through the following steps, in order to update the initial model parameter information:

[0066] The first step is to determine the response ranking probability information of each item identifier in the sampling set based on the actual response ranking information corresponding to the item identifier, the item identifier set, and the initial model parameters. In practice, the executing entity can use a response ranking probability model based on a Poisson process to determine the response ranking probability information of the item identifier corresponding to the actual response ranking information. The response ranking of an item identifier in the sampling set can follow a right-truncated non-homogeneous Poisson process. Under this Poisson process modeling, the probability that an item identifier ranks according to the corresponding actual response ranking information in the sampling set is the response ranking probability information. As an example, the response ranking probability information can be expressed as the following formula:

[0067]

[0068]

[0069] Where x represents the item identifier and k represents the response ranking. Let λ represent the sampling set. λ(s, x) represents the intensity value in space s associated with x. X represents the set of item identifiers. ξ Z(x) represents the transformation of λ(s, x). Z(x) is the normalization term for the right-truncated Poisson process. ξ (x) can be further expressed as λ ξ (x; θ). A multilayer perceptron can be used to analyze λ. ξ The model is fitted using (x; θ). θ can be the model parameters of the multilayer perceptron. The model parameters can include the weights of each network layer. This represents the number of each item identifier in the sampling set, excluding item identifier x.

[0070] The second step involves training the initial fitted model based on the determined ranking probabilities of each response to update the initial model parameters. In practice, the aforementioned execution entity can use maximum likelihood estimation to train θ and update the initial model parameters. That is, log-likelihood can be used as the objective function, and then an optimizer can be used to train θ. For example, the optimizer could be the ADAM optimizer.

[0071] Step 2023: Using the updated initial model parameter information, filter out item identifiers that meet the preset optimization conditions from the item identifier set.

[0072] In some embodiments, the execution entity can filter item identifiers that meet preset optimization conditions from the item identifier set using updated initial model parameter information. The updated initial model parameter information can be used to determine the response ranking probability of the corresponding predicted response ranking for item identifiers outside the sampling set on the item identifier set. In practice, for each item identifier in the item identifier set other than the sampling set, the execution entity can use the updated initial model parameter information to determine the response ranking probability for each predicted response ranking for that item identifier. Then, the variance and standard deviation of each response ranking probability can be determined. Next, the variance and standard deviation can be input into a preset acquisition function to obtain the acquisition value corresponding to the item identifier. The preset acquisition function can be a confidence lower bound acquisition function. Then, the execution entity can filter item identifiers that meet preset optimization conditions from each item identifier to be filtered. The preset optimization condition can be minimizing the acquisition value corresponding to the item identifier.

[0073] Step 2024: Add the selected item identifiers to the sample set to update the sample set.

[0074] In some embodiments, the aforementioned implementing entity may add the selected item identifiers to the sampling set to update the sampling set.

[0075] Step 2025: In response to the determination that the number of iteration steps is equal to the preset number of iterations, at least one item identifier in the sample set that satisfies the preset ranking conditions is determined as the recommended item identifier set, so as to optimize the item recommendation black box model.

[0076] In some embodiments, the execution entity may, in response to determining that the number of iterations equals a preset number of iterations, identify at least one item identifier in the sampling set that satisfies a preset ranking condition as the recommended item identifier set, thereby optimizing the item recommendation black-box model. The preset ranking condition may be that the actual response ranking information corresponding to the item identifier is minimized. Alternatively, the preset ranking condition may be that the corresponding actual response ranking information is less than a preset ranking threshold. Here, the smaller the actual response ranking information, the higher the ranking. The specific setting of the preset ranking threshold is not limited. Thus, the initial fitted model can be iteratively trained within a preset number of iterations.

[0077] Optionally, the execution entity may also, in response to the determination that the number of iterations is less than the preset number of iterations, execute the iterations again based on the updated sample set, the updated initial model parameter information, and the trained initial fitted model.

[0078] Optionally, the executing entity may also send the item information corresponding to each recommended item identifier in the aforementioned set of recommended item identifiers to at least one terminal device corresponding to the aforementioned target user information. The item information may include, but is not limited to, at least one of the following: item title, item image, item video, and item details. Then, a predetermined number of recommended item identifiers can be selected from the aforementioned set of recommended item identifiers as the set of item identifiers to be replenished. In practice, the executing entity may select a predetermined number of recommended item identifiers that rank highly in the response ranking information from the aforementioned set of recommended item identifiers as the set of item identifiers to be replenished. Next, for each item identifier to be replenished in the aforementioned set of item identifiers to be replenished, the following steps can be performed:

[0079] The first step is to determine the replenishment quantity corresponding to the aforementioned item identification. In practice, the implementing entity can obtain the replenishment quantity corresponding to the aforementioned item identification from the item replenishment system. The item replenishment system can be a system used to monitor the replenishment status of items.

[0080] The second step involves, in response to the determination that the replenishment quantity meets the preset replenishment conditions, controlling the item dispatching equipment corresponding to the item identification to be replenished to perform an item replenishment operation based on the replenishment quantity. This item dispatching equipment can be an unmanned transport vehicle. The preset replenishment conditions can be that the replenishment quantity is greater than a preset replenishment quantity. For example, the preset replenishment quantity can be 0. The item replenishment operation can be the operation of dispatching the item to its storage location in the warehouse. Therefore, the items to be replenished can be determined using the item recommendation results for the user, allowing these items to be replenished first, thus fulfilling the user's delivery needs as quickly as possible after recommending these items to the user.

[0081] The above embodiments of this disclosure have the following beneficial effects: the item recommendation black box model optimization method of some embodiments of this disclosure improves the practicality and accuracy of the optimized item recommendation black box model. Specifically, the reason for the poor practicality and low accuracy of the optimized item recommendation black box model is that the Gaussian process gives a specific value of its response, which is an absolute response. The absolute response value may be difficult to obtain in practice, resulting in poor practicality of the optimized item recommendation black box model. Moreover, the absolute response method is highly sensitive to noise, resulting in low accuracy of the final optimized item recommendation black box model. Based on this, the item recommendation black box model optimization method of some embodiments of this disclosure first samples a preset number of item identifiers from the item identifier set to form a sampling set. This avoids using a large number of item identifiers to train the initial fitted model at the beginning. Then, based on the sampling set and the initial model parameter information, the following iterative steps are performed: First, based on the item recommendation black box model corresponding to the target user information, the actual response ranking information of each item identifier corresponding to the target user information in the sampling set is determined, and the actual response ranking information is obtained. Therefore, the relative true response ranking of each point in the sampling points output by the black-box model can be obtained in advance. The second step involves training the initial fitted model based on the item identifier set, the sampling set, the actual response ranking information, and the initial model parameter information, thereby updating the initial model parameter information. This allows training the initial fitted model from the reference dimension of relative ranking. The third step involves selecting item identifiers that meet preset optimization conditions from the item identifier set using the updated initial model parameter information. This allows selecting the optimal query point (i.e., item identifier) ​​to add to the sampling set to optimize the item recommendation black-box model. The fourth step involves adding the selected item identifiers to the sampling set to update the sampling set. The fifth step, in response to the determination that the number of iteration steps equals the preset number of iterations, determines at least one item identifier in the sampling set that meets the preset ranking conditions as the recommended item identifier set to optimize the aforementioned item recommendation black-box model. Thus, after iteration, at least one item identifier in the sampling set that meets the preset ranking conditions can be determined as the final output of the item recommendation black-box model. Because the optimization of the item black-box model did not use absolute response as a reference dimension, but instead used relative ranking to train the initial fitted model, relative rankings are easier to obtain than absolute item recommendation values, thus improving the practicality of the optimized item recommendation black-box model. Furthermore, compared to absolute item recommendation values, the relative ranking relationships between items are less susceptible to noise corruption; therefore, relative rankings are more robust to noise and less sensitive to it, thereby improving the accuracy of the final optimized item recommendation black-box model.

[0082] Further reference Figure 3This illustrates a flow 300 of another embodiment of the item recommendation black-box model optimization method. Flow 300 of the item recommendation black-box model optimization method includes the following steps:

[0083] Step 301: Sample a preset number of item identifiers from the item identifier set to form a sampling set.

[0084] Step 302: Based on the sample set and initial model parameter information, perform the following iterative steps:

[0085] Step 3021: Based on the black-box model of item recommendation corresponding to the target user information, determine the actual response ranking information of each item identifier corresponding to the target user information in the sampling set, and obtain the actual response ranking information.

[0086] Step 3022: Train the initial fitted model based on the item identifier set, sampling set, ranking information of each actual response, and initial model parameter information to update the initial model parameter information.

[0087] In some embodiments, the specific implementation of steps 301-3022 and their effects can be found in [reference needed]. Figure 2 Steps 201-2022 in the corresponding embodiments will not be repeated here.

[0088] Step 3023: Combine the item identifiers in the item identifier set that are different from the item identifiers in the sampling set into a set of item identifiers to be screened.

[0089] In some embodiments, the execution entity of the item recommendation black-box model optimization method (e.g.) Figure 1 The server shown can combine item identifiers in the item identifier set that are different from the item identifiers in the sampling set into a set of item identifiers to be filtered.

[0090] Step 3024: For each item identifier in the set of item identifiers to be filtered, generate the collected value corresponding to the item identifier to be filtered based on the updated initial model parameter information.

[0091] In some embodiments, for each item identifier in the set of item identifiers to be filtered, the execution entity may generate the collected value corresponding to the item identifier to be filtered based on the updated initial model parameter information.

[0092] In some optional implementations of certain embodiments, the aforementioned execution entity can generate the collected values ​​corresponding to the item identifiers to be filtered based on the updated initial model parameter information through the following steps:

[0093] The first step is to generate a first comparison value based on the sample set and a first preset coefficient. In practice, the aforementioned executing entity can determine the first comparison value by multiplying the number of item identifiers included in the updated sample set by the first preset coefficient. The first preset coefficient can be a pre-set value. For example, the first preset coefficient can be the Baidu ratio of the response ranking. The value range of the first preset coefficient can be (0, 1), and it can be used to determine the ranking threshold.

[0094] The second step involves generating a second comparison value based on the identified items to be filtered and the updated initial model parameters. In practice, the aforementioned execution entity can use λ... ξ The second comparison value is determined by the product of (x; θ) and the number of each item identifier included in the item identifier set. Here, x represents the item identifier to be filtered, and θ represents the updated initial model parameter information.

[0095] Third, in response to the determination that the second comparison value is less than the first comparison value, a collection value corresponding to the item identifier to be filtered is generated based on a preset collection function. The preset collection function can be a confidence lower bound collection function. When using a confidence lower bound collection function as the collection function, variance and standard deviation can represent the variance and standard deviation of the probability information of each response ranking for all predicted response rankings.

[0096] Fourth, in response to determining that the second comparison value is greater than or equal to the first comparison value, a sampled value is selected from a preset sampled value range as the sampled value corresponding to the item identifier to be filtered. The preset sampled value range can be [0, 1]. In practice, the executing entity can randomly select a sampled value from the preset sampled value range as the sampled value corresponding to the item identifier to be filtered. Thus, all item identifiers with excessively low rankings exceeding the ranking threshold can be considered as having the same ranking, and random uniform sampling can be used for sampling.

[0097] In some optional implementations of certain embodiments, the aforementioned execution entity can generate the collected values ​​corresponding to the item identifiers to be filtered based on the updated initial model parameter information through the following steps:

[0098] Based on the initial response ranking, perform the following iterative steps:

[0099] The first step is to generate the response ranking probability information for the initial response ranking corresponding to the aforementioned item identifiers to be filtered, based on the updated initial model parameter information. The initial response ranking can be 0. In practice, the executing entity can generate the response ranking probability information for the initial response ranking corresponding to the aforementioned item identifiers to be filtered using the following formula:

[0100]

[0101] Where, x *This indicates the item to be filtered.

[0102] The second step is to determine the ranking difference as the difference between the preset response ranking and the initial response ranking. The preset response ranking can be pre-set. For example, the preset response ranking can be 5.

[0103] The third step is to determine the ranking improvement expectation by multiplying the ranking difference by the response ranking probability information.

[0104] The fourth step is to increment the initial response ranking to update it. In practice, the executing entity can increment the initial response ranking by 1 to update it.

[0105] Fifth, in response to the determination that the updated initial response ranking is less than the preset response ranking, the above loop steps are executed again.

[0106] Step 6: In response to the confirmation that the updated initial response ranking is equal to the aforementioned preset response ranking, the sum of the expected increases in each ranking is determined as the collected value corresponding to the item identifier to be filtered. Therefore, when determining the item identifier for the next query, the expected increase in absolute response value can be replaced with the expected increase in relative ranking, thus determining how much the expected increase in all item identifiers to be filtered can be relative to the preset response ranking.

[0107] Step 3025: Identify the item identifiers in the set of item identifiers whose corresponding collected values ​​meet the preset minimum value condition as item identifiers that meet the preset optimization conditions.

[0108] In some embodiments, the executing entity may determine the item identifiers in the set of item identifiers to be filtered that whose corresponding collected values ​​meet a preset minimum value condition as item identifiers that meet preset optimization conditions. The preset minimum value condition may be the minimum collected value.

[0109] Step 3026: Add the selected item identifiers to the sampling set to update the sampling set.

[0110] Step 3027: In response to the determination that the number of iteration steps is equal to the preset number of iterations, at least one item identifier in the sampling set that satisfies the preset ranking conditions is determined as the recommended item identifier set, so as to optimize the item recommendation black box model.

[0111] In some embodiments, the specific implementation of steps 3026-3027 and their effects can be found in [reference needed]. Figure 2 Steps 2024-2025 in the corresponding embodiments will not be repeated here.

[0112] from Figure 3 It can be seen from this that, with Figure 2 Compared to the description of some corresponding embodiments, Figure 3 The flow 300 of the item recommendation black-box model optimization method in some corresponding embodiments embodies the step of expanding the item identifiers for filtering the next query. Therefore, the schemes described in these embodiments can be adapted to the optimization process of a corresponding item recommendation black-box model.

[0113] Further reference Figure 4 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an item recommendation black-box model optimization device, which are similar to... Figure 2 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0114] like Figure 4 As shown, the item recommendation black-box model optimization apparatus 400 in some embodiments includes a sampling unit 401 and an execution unit 402. The sampling unit 401 is configured to sample a preset number of item identifiers from an item identifier set to form a sampling set. The execution unit 402 is configured to perform the following iterative steps based on the sampling set and initial model parameter information: determining the actual response ranking information of each item identifier corresponding to the target user information in the sampling set based on the item recommendation black-box model corresponding to the target user information, thus obtaining each actual response ranking information; training an initial fitted model according to the item identifier set, the sampling set, each actual response ranking information, and the initial model parameter information to update the initial model parameter information; selecting item identifiers that meet preset optimization conditions from the item identifier set using the updated initial model parameter information; adding the selected item identifiers to the sampling set to update the sampling set; and, in response to determining that the number of executions of the iterative steps equals the preset number of iterations, determining at least one item identifier in the sampling set that meets the preset ranking conditions as a set of recommended item identifiers to optimize the aforementioned item recommendation black-box model.

[0115] Optionally, the above iterative steps further include: in response to the determination that the number of executions of the iterative steps is less than the preset number of iterations, the above iterative steps are executed again based on the updated sample set, the updated initial model parameter information, and the trained initial fitted model.

[0116] Optionally, the execution unit 402 can be further configured to: determine the recommended value of the target user information corresponding to each item identifier in the sampling set based on the above-mentioned black-box model for item recommendation; for each item identifier in the sampling set, determine the number of each item identifier in the sampling set whose recommended value satisfies the preset optimization condition as a metric value, wherein the above-mentioned preset optimization condition corresponds to the recommended value of the above-mentioned item identifier; and determine the actual response ranking information corresponding to each item identifier in the sampling set based on the determined metric value corresponding to each item identifier in the sampling set.

[0117] Optionally, the execution unit 402 may be further configured to: sort each item identifier in the sample set according to the measure value corresponding to each item identifier in the determined sample set to obtain an item identifier sequence; and for each item identifier in the sample set, determine the order of the item identifier in the item identifier sequence as the actual response ranking information corresponding to the item identifier.

[0118] Optionally, the execution unit 402 may be further configured to: for each item identifier in the sampling set, determine the response ranking probability information corresponding to the actual response ranking information of the item identifier based on the actual response ranking information of the item identifier, the item identifier set and the initial model parameter information; and train the initial fitting model based on the determined response ranking probability information to update the initial model parameter information.

[0119] Optionally, the execution unit 402 may be further configured to: combine the item identifiers in the item identifier set that are different from the item identifiers in the sampling set into a set of item identifiers to be filtered; for each item identifier to be filtered in the set of item identifiers to be filtered, generate the collected value corresponding to the item identifier to be filtered based on the updated initial model parameter information; and determine the item identifiers in the set of item identifiers to be filtered whose corresponding collected values ​​satisfy the preset minimum value condition as item identifiers that satisfy the preset optimization condition.

[0120] Optionally, the execution unit 402 may be further configured to: generate a first comparison value based on the sampling set and a first preset coefficient; generate a second comparison value based on the item identifier to be filtered and the updated initial model parameter information; in response to determining that the second comparison value is less than the first comparison value, generate a collection value corresponding to the item identifier to be filtered based on a preset collection function; and in response to determining that the second comparison value is greater than or equal to the first comparison value, select a collection value from the preset collection value range as the collection value corresponding to the item identifier to be filtered.

[0121] Optionally, the execution unit 402 may be further configured to: based on the initial response ranking, perform the following iterative steps: generate response ranking probability information for the initial response ranking corresponding to the above-mentioned item identifier to be filtered, according to the updated initial model parameter information; determine the difference between the preset response ranking and the initial response ranking as the ranking difference; determine the ranking improvement expectation by multiplying the ranking difference and the response ranking probability information; perform auto-increment processing on the initial response ranking to update the initial response ranking; in response to determining that the updated initial response ranking is less than the above-mentioned preset response ranking, perform the above iterative steps again; in response to determining that the updated initial response ranking is equal to the above-mentioned preset response ranking, determine the sum of the obtained ranking improvement expectations as the collected value corresponding to the above-mentioned item identifier to be filtered.

[0122] Optionally, the item recommendation black-box model optimization device 400 further includes: a sending unit, a selection unit, and a step execution unit (not shown in the figure). The sending unit is configured to send the item information corresponding to each recommended item identifier in the aforementioned set of recommended item identifiers to at least one terminal device corresponding to the aforementioned target user information. The selection unit is configured to select a preset number of recommended item identifiers from the aforementioned set of recommended item identifiers as a set of item identifiers to be replenished. The step execution unit is configured to, for each item identifier to be replenished in the aforementioned set of item identifiers to be replenished, perform the following steps: determine the replenishment quantity corresponding to the item identifier to be replenished; in response to determining that the replenishment quantity meets preset replenishment conditions, control the item scheduling device corresponding to the item identifier to be replenished to perform an item replenishment operation according to the replenishment quantity.

[0123] It is understandable that the units described in the device 400 are related to the reference. Figure 2 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method also apply to device 400 and the units contained therein, and will not be repeated here.

[0124] The following is for reference. Figure 5 It illustrates an electronic device 500 suitable for implementing some embodiments of the present disclosure (e.g., Figure 1 A schematic diagram of the structure of the server in the diagram. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0125] like Figure 5As shown, the electronic device 500 may include a processing unit 501 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0126] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0127] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 509, or installed from storage device 508, or installed from ROM 502. When the computer program is executed by processing device 501, it performs the functions defined in the methods of some embodiments of this disclosure.

[0128] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0129] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0130] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: sample a predetermined number of item identifiers from the item identifier set to form a sampling set; based on the sampling set and initial model parameter information, perform the following iterative steps: based on the item recommendation black-box model corresponding to the target user information, determine the actual response ranking information of each item identifier in the sampling set corresponding to the target user information, and obtain each actual response ranking information; train the initial fitting model according to the item identifier set, the sampling set, each actual response ranking information, and the initial model parameter information to update the initial model parameter information; filter item identifiers that meet the predetermined optimization conditions from the item identifier set using the updated initial model parameter information; add the selected item identifiers to the sampling set to update the sampling set; in response to determining that the number of executions of the iterative steps equals the predetermined number of iterations, determine at least one item identifier in the sampling set that meets the predetermined ranking conditions as the recommended item identifier set to optimize the aforementioned item recommendation black-box model.

[0131] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0133] The units described in some embodiments of this disclosure can be implemented in software or in hardware. The described units can also be housed in a processor; for example, a processor may be described as including a sampling unit and an execution unit. The names of these units do not necessarily limit the unit itself; for example, a sampling unit may also be described as "a unit that samples a preset number of item identifiers from a set of item identifiers to form a sampling set."

[0134] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0135] Some embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements any of the above-described item recommendation black-box model optimization methods.

[0136] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for optimizing an item recommendation black box model, comprising: sampling a preset number of item identifiers from an item identifier set to form a sampling set; based on the sampling set and initial model parameter information, performing the following iteration steps: determining actual response ranking information of each item identifier in the sampling set corresponding to target user information based on the item recommendation black box model corresponding to the target user information, to obtain respective actual response ranking information; training an initial fitting model based on the item identifier set, the sampling set, the respective actual response ranking information, and the initial model parameter information, to update the initial model parameter information; screening item identifiers from the item identifier set that satisfy a preset optimization condition through the updated initial model parameter information; adding the screened item identifiers to the sampling set to update the sampling set; in response to determining that the number of execution times of the iteration steps is equal to a preset iteration number, determining at least one item identifier in the sampling set that satisfies a preset ranking condition as a recommended item identifier set to optimize the item recommendation black box model.

2. The method of claim 1, wherein, The iteration steps further comprise: in response to determining that the number of execution times of the iteration steps is less than the preset iteration number, based on the updated sampling set, the updated initial model parameter information, and the trained initial fitting model, executing the iteration steps again.

3. The method of claim 1, wherein, The determining of the actual response ranking information of each item identifier in the sampling set corresponding to target user information based on the item recommendation black box model corresponding to the target user information comprises: determining a recommendation value of each item identifier in the sampling set corresponding to the target user information based on the item recommendation black box model; for each item identifier in the sampling set, determining a number of respective item identifiers in the sampling set that satisfy a preset recommendation condition corresponding to the recommendation value of the item identifier as a measure value; determining the actual response ranking information of each item identifier in the sampling set corresponding to the target user information according to the determined measure value of each item identifier in the sampling set.

4. The method of claim 3, wherein, The determining of the actual response ranking information of each item identifier in the sampling set corresponding to the target user information according to the determined measure value of each item identifier in the sampling set comprises: sorting each item identifier in the sampling set according to the determined measure value of each item identifier in the sampling set to obtain an item identifier sequence; for each item identifier in the sampling set, determining a ranking position of the item identifier in the item identifier sequence as the actual response ranking information of the item identifier.

5. The method of claim 1, wherein, The training of the initial fitting model based on the item identifier set, the sampling set, the respective actual response ranking information, and the initial model parameter information to update the initial model parameter information comprises: for each item identifier in the sampling set, determining response ranking probability information corresponding to the actual response ranking information of the item identifier based on the actual response ranking information of the item identifier, the item identifier set, and the initial model parameter information; training the initial fitting model based on the determined respective response ranking probability information to update the initial model parameter information.

6. The method of claim 1, wherein, The filtering of the item identifiers satisfying the preset optimization condition from the item identifier set based on the updated initial model parameter information comprises: combining each item identifier in the item identifier set that is different from each item identifier in the sampling set into a set of to-be-filtered item identifiers; for each to-be-filtered item identifier in the set of to-be-filtered item identifiers, generating a collection value corresponding to the to-be-filtered item identifier based on the updated initial model parameter information; determining, as the item identifiers satisfying the preset optimization condition, the to-be-filtered item identifiers in the set of to-be-filtered item identifiers for which the corresponding collection values satisfy a preset minimum value condition.

7. The method of claim 6, wherein, The generating of the collection value corresponding to the to-be-filtered item identifier based on the updated initial model parameter information comprises: generating a first comparison value according to the sampling set and a first preset coefficient; generating a second comparison value according to the to-be-filtered item identifier and the updated initial model parameter information; in response to determining that the second comparison value is less than the first comparison value, generating the collection value corresponding to the to-be-filtered item identifier based on a preset collection function; in response to determining that the second comparison value is greater than or equal to the first comparison value, selecting a collection value from a preset collection value range as the collection value corresponding to the to-be-filtered item identifier.

8. The method of claim 6, wherein, The generating of the collection value corresponding to the to-be-filtered item identifier based on the updated initial model parameter information comprises: based on the initial response ranking, performing the following loop steps: generating response ranking probability information of the initial response ranking corresponding to the to-be-filtered item identifier according to the updated initial model parameter information; determining a ranking difference as the difference between a preset response ranking and the initial response ranking; determining a ranking promotion expectation as the product of the ranking difference and the response ranking probability information; performing an increment operation on the initial response ranking to update the initial response ranking; in response to determining that the updated initial response ranking is less than the preset response ranking, performing the loop steps again; in response to determining that the updated initial response ranking is equal to the preset response ranking, determining the sum of the obtained ranking promotion expectations as the collection value corresponding to the to-be-filtered item identifier.

9. The method according to one of claims 1-8, wherein, The method further comprises: sending item information corresponding to each recommended item identifier in the set of recommended item identifiers to at least one terminal device corresponding to the target user information; selecting a preset number of recommended item identifiers from the set of recommended item identifiers as a set of to-be-replenished item identifiers; for each to-be-replenished item identifier in the set of to-be-replenished item identifiers, performing the following steps: determining a replenishment quantity corresponding to the to-be-replenished item identifier; in response to determining that the replenishment quantity satisfies a preset replenishment condition, controlling an item scheduling device corresponding to the to-be-replenished item identifier to perform an item replenishment operation according to the replenishment quantity.

10. An item recommendation black box model optimization apparatus, comprising: a sampling unit configured to sample a preset number of item identifiers from a set of item identifiers to form a sampling set; The execution unit is configured to perform the following iterative steps based on the sample set and the initial model parameter information: determining actual response ranking information corresponding to the target user information for each item identifier in the sample set based on the item recommendation black box model of the corresponding target user information, to obtain respective actual response ranking information; training the initial fitting model based on the set of item identifiers, the sample set, the respective actual response ranking information, and the initial model parameter information to update the initial model parameter information; filtering, from the set of item identifiers, item identifiers that satisfy a preset optimization condition by using the updated initial model parameter information; adding the filtered item identifiers to the sample set to update the sample set; and in response to determining that the number of iterations of the iterative steps is equal to a preset number of iterations, determining at least one item identifier in the sample set that satisfies a preset ranking condition as a set of recommended item identifiers to optimize the item recommendation black box model. 11.An electronic device, comprising: one or more processors; a memory device having stored thereon one or more programs, when the one or more programs are executed by the one or more processors, cause the one or more processors to carry out the method of any one of claims 1-9.

12. A computer readable medium having stored thereon a computer program, wherein, The computer program is executed by the processor to implement the method of any one of claims 1-9. 13.A computer program product comprising a computer program which, when executed by a processor, carries out the method according to any one of claims 1-9.

Images