Methods, apparatus, equipment, and computer storage media for multi-target parameter tuning

Abstract

This application provides a parameter tuning method, apparatus, device, and computer storage medium for multiple objectives. The method includes: collecting historical behavior information of multiple users regarding recommended materials; determining the revenue parameter vector corresponding to each of the N sets of fusion parameter vectors in the fusion network based on the historical behavior information; determining the correlation coefficient between each set of fusion parameter vectors and the revenue parameter vector corresponding to each set of fusion parameter vectors; determining the increment vector between the first revenue parameter vector and each second revenue parameter vector in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors; updating the mean vector of the fusion parameter vector corresponding to each second revenue parameter vector based on each set of increment vectors; and updating the N sets of fusion parameter vectors based on the updated mean vector and the correlation coefficient. This application improves the efficiency and speed of fusion parameter vector optimization, achieving more accurate material recommendations.

Description

Technical Field

[0001] This application relates to the field of computer processing technology, and in particular to a method, apparatus, device, and computer storage medium for parameter tuning of multiple targets. Background Technology

[0002] Currently, recommendation systems aim to increase user growth by recommending materials that users are interested in. During the material recommendation process, the system predicts the click-through rate, interaction rate, and viewing time of candidate materials. These predictions are then weighted and summed based on a fusion parameter vector to obtain a fusion prediction score for each candidate material. Material recommendations are then made to users based on this score. To better explore user interests, the fusion parameter vector is updated based on the user's historical behavior information during each material recommendation process.

[0003] Currently, the new fusion parameter vector generated by the parameter tuning algorithm of the recommendation system may get stuck in a local optimum, making it impossible to find a better fusion parameter vector, which affects the recommendation accuracy of the recommendation system. Summary of the Invention

[0004] This application provides a parameter tuning method, apparatus, device, and computer storage medium for multiple objectives, which can improve the efficiency and speed of fusion parameter vector optimization and achieve more accurate material recommendation.

[0005] The technical solution of this application is implemented as follows:

[0006] This application provides a parameter tuning method for multiple objectives. The method includes: collecting historical behavior information of multiple users towards recommended materials; determining the revenue parameter vector corresponding to each of the N sets of fusion parameter vectors in the fusion network based on the historical behavior information, wherein the fusion network fuses the predicted values ​​of x objectives of the candidate materials based on the fusion parameter vectors, and the revenue parameter vector is used to represent the user's interest in the recommended materials, where N and x are integers greater than or equal to 2; determining the correlation coefficient between each set of fusion parameter vectors and the revenue parameter vector corresponding to each set of fusion parameter vectors; determining the increment vector between the first revenue parameter vector and each set of second revenue parameter vectors in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors, wherein the first revenue parameter vector is the optimal revenue parameter vector in the N sets of revenue parameter vectors, and the second revenue parameter vectors are the other sets of revenue parameter vectors besides the optimal revenue parameter vector; updating the mean vector of the fusion parameter vector corresponding to each set of second revenue parameter vectors based on each set of increment vectors; and updating the N sets of fusion parameter vectors based on the updated mean vector and the correlation coefficient.

[0007] In some possible implementations, determining the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector includes: calculating the Pearson correlation coefficient for x parameters in each set of fusion parameter vectors and x parameters in the corresponding revenue parameter vector to obtain x correlation arrays, wherein each of the x correlation arrays includes x correlation coefficients.

[0008] In some possible implementations, determining the increment vector between the first revenue parameter vector and each group of second revenue parameter vectors in the N groups of revenue parameter vectors corresponding to the N groups of fusion parameter vectors includes: for any group of second revenue parameter vectors, calculating the increment of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector to obtain the increment vector between the first revenue parameter vector and the second revenue parameter vector.

[0009] In some possible implementations, for any set of second revenue parameter vectors, by calculating the increments of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector, the increment vector between the first and second revenue parameter vectors is obtained, which satisfies the following expression (1):

[0010]

[0011] In the formula, Δr i Let r be the i-th increment vector, 1≤i≤N-1. ai ,r bi ,…,r xi Let x be the parameters in the second revenue parameter vector of the i-th group, and r be the parameters in the second revenue parameter vector. a0 ,r b0 ,…,r x0 Let x be the parameters in the first benefit parameter vector, [Δr ai ,Δr bi ,…,Δr xi ] represents the x parameters in the i-th group of increment vectors.

[0012] In some possible implementations, based on each set of incremental vectors, the mean vector of the fusion parameter vector corresponding to each set of second benefit parameter vectors is updated, including: for any set of incremental vectors, collecting the values ​​of x parameters in the incremental vector; subtracting the minimum value from the maximum value among the x parameters in the incremental vector to obtain the updated mean vector.

[0013] In some possible implementations, the first revenue parameter vector corresponds to the first fusion parameter vector in N sets of fusion parameter vectors, and each set of second revenue parameter vectors corresponds to each set of second fusion parameter vectors in N sets of fusion parameter vectors other than the first fusion parameter vector; updating the N sets of fusion parameter vectors according to the updated mean vector and correlation coefficients includes: calculating the values ​​of x parameters in each set of updated second fusion parameter vectors according to the updated mean vector and x correlation arrays; obtaining the updated N sets of fusion parameter vectors based on the updated sets of second fusion parameter vectors and the first fusion parameter vector.

[0014] In some possible implementations, based on the updated mean vector and x correlation arrays, the values ​​of x parameters in each updated set of second fusion parameter vectors are calculated to satisfy the following expression (2):

[0015]

[0016] In the formula, a i '、b i '、…、x i ' represents the values ​​of x parameters in the i-th group of the updated N-1 groups of second fusion parameter vectors, 1≤i≤N-1, a i b i ... x i Let x be the values ​​of the i-th parameter in the N-1 sets of second fusion parameter vectors, and μ be the value of the parameter in the i-th set of second fusion parameter vectors. ai ',μ bi ',…,μ xi ' represents the values ​​of x parameters in the updated i-th group of mean vectors, ρ (a,r) For each set of fusion parameter vectors, ρ represents the correlation coefficient set of the first parameter and the x parameters in the corresponding return parameter vector. (b,r) For each set of fusion parameter vectors, ρ represents the correlation coefficient set of the second parameter and the x parameters in the corresponding return parameter vector. (x,r) Let r be the correlation coefficient set between the x-th parameter in each set of fusion parameter vectors and the x-th parameters in the corresponding profit parameter vector, where r min r is the minimum value among the x parameters of the revenue parameter vector. max It is the maximum value among the x parameters of the revenue parameter vector.

[0017] This application provides a parameter tuning device for multiple objectives, installed in a recommendation system. The device includes: a data acquisition module for collecting historical behavior information of multiple users towards recommended materials; a data processing module for determining, based on the historical behavior information, the revenue parameter vector corresponding to each of the N sets of fusion parameter vectors in a fusion network, wherein the fusion network fuses the predicted values ​​of x objectives of the candidate material based on the fusion parameter vectors, and the revenue parameter vector represents the user's degree of interest in the recommended material, where N and x are integers greater than or equal to 2; determining the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector; determining the increment vector between the first revenue parameter vector and each second revenue parameter vector in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors, where the first revenue parameter vector is the optimal revenue parameter vector in the N sets of revenue parameter vectors, and the second revenue parameter vectors are the other sets of revenue parameter vectors besides the optimal one; and an update module for updating the mean vector of the fusion parameter vector corresponding to each second revenue parameter vector based on each set of increment vectors; and updating the N sets of fusion parameter vectors based on the updated mean vector and the correlation coefficient.

[0018] In some possible implementations, the data processing module is further configured to calculate the Pearson correlation coefficient for x parameters in each set of fusion parameter vectors and x parameters in the corresponding revenue parameter vectors of each set of fusion parameter vectors, to obtain x correlation arrays, wherein each of the x correlation arrays includes x correlation coefficients.

[0019] In some possible implementations, the data processing module is further configured to, for any set of second revenue parameter vectors, calculate the increment vector between the first revenue parameter vector and the second revenue parameter vector by calculating the increment of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector.

[0020] In some possible implementations, the increment vector is obtained by calculating the increments of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector, satisfying expression (1).

[0021] In some possible implementations, the data acquisition module is further configured to acquire the values ​​of x parameters in any set of incremental vectors; the data processing module is further configured to subtract the minimum value from the maximum value among the x parameters in the incremental vector to obtain the updated mean vector.

[0022] In some possible implementations, the first revenue parameter vector corresponds to the first fusion parameter vector in the N sets of fusion parameter vectors, and each set of second revenue parameter vectors corresponds to each set of second fusion parameter vectors in the N sets of fusion parameter vectors other than the first fusion parameter vector; the data processing module is further configured to calculate the values ​​of x parameters in each set of updated second fusion parameter vectors based on the updated mean vector and x correlation arrays; the update module is further configured to obtain the updated N sets of fusion parameter vectors based on the updated sets of second fusion parameter vectors and the first fusion parameter vector.

[0023] In some possible implementations, the values ​​of x parameters in the updated N-1 sets of second fusion parameter vectors are calculated to satisfy expression (2) based on the updated mean vector and x correlation arrays.

[0024] This application provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided in this application.

[0025] This application provides a computer storage medium storing executable instructions, which are executed by a processor to implement the method provided in this application.

[0026] This application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the method provided in this application.

[0027] This application has the following beneficial effects:

[0028] This application updates the mean vector by calculating the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector, and by calculating the increment vectors of the first and second revenue parameter vectors among the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors. This can improve the target type with the smallest data increase and suppress the target type with the largest data increase, thus achieving certainty in the exploration direction. In addition, based on the updated mean vector and correlation coefficient, the N sets of fusion parameter vectors are updated, so that the update step size of the fusion parameters depends on the updated mean vector and correlation coefficient. The update step size of the fusion parameters can be flexibly adjusted to avoid the parameter group getting stuck in a local optimum. As a result, the updated fusion parameter vectors can better express the user's interest tendencies and achieve more accurate material recommendations. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of an architecture of the recommendation system provided in an embodiment of this application;

[0030] Figure 2This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0031] Figure 3 This is a schematic diagram of the information flow recommendation process of the recommendation system provided in the embodiments of this application;

[0032] Figure 4 This is a flowchart illustrating a parameter tuning method for multiple targets provided in an embodiment of this application.

[0033] Figure 5 This is an optional flowchart illustrating a parameter tuning method for multiple targets provided in an embodiment of this application;

[0034] Figure 6 This is an optional flowchart illustrating a parameter tuning method for multiple targets provided in an embodiment of this application;

[0035] Figure 7 This is a flowchart illustrating the parameter tuning method provided in this application embodiment applied to social application software. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0038] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0040] In an information flow recommendation system, the expression behaviors of users' interest preference degrees for materials include browsing, clicking, interacting, complaining, etc. The recommendation system needs to predict users' potential interest preference tendencies and weigh the possibilities of various expression behaviors, so as to comprehensively rank the materials, improve the quality of the user experience, and enhance the accuracy of material recommendation. Common recommendation goals include, but are not limited to, the probability that a user clicks on a material (hereinafter referred to as the click-through rate), the probability of an interaction behavior (hereinafter referred to as the interaction rate), and the consumption duration of the user (hereinafter referred to as the viewing duration). The recommendation system will respectively predict the click-through rate, interaction rate, and consumption duration of users for materials, and based on a set of fixed parameters, weight and sum the respective prediction scores to obtain the total material evaluation score, and rank the candidate materials based on the total score.

[0041] Currently, the fusion network in the recommendation system generally sets a fixed fusion parameter vector W=(a, b, c) for the three recommendation goals of click-through rate, interaction rate, and viewing duration. a is the fusion parameter for the click-through rate, b is the fusion parameter for the interaction rate, and c is the fusion parameter for the viewing duration. Assuming that the above fusion parameters follow a Gaussian distribution, two vectors are set, namely the mean vector μ=(μ a , μ b , μ c ) and the variance vector σ=(σ a , σ b , σ c ). The specific process of updating the fusion parameter vector is as follows: (1). The recommendation system performs Gaussian random sampling on the fusion parameters based on W, μ, and σ to obtain n groups of fusion parameter vectors (which can also be called exploration vectors): W1, W2, W3,..., W n ; (2). The recommendation system randomly maps each user to a certain group of exploration vectors, and combines the prediction scores of the user for the recommendation goals such as the click-through rate, interaction rate, and viewing duration of the material, and obtains the total material evaluation score S in a weighted summation manner, S = a × click-through rate prediction value + b × interaction rate prediction value + c × consumption duration prediction value; (3). The recommendation system collects the n groups of fusion parameter vectors distributed by the recommendation system and the user behavior logs, calculates the benefits brought by each group of fusion parameter vectors respectively, and selects the k groups of benefit parameter vectors corresponding to the k groups of fusion parameter vectors with the highest benefits (k < n), and calculates the new mean vector μ' and variance vector σ' respectively based on the k groups of benefit parameter vectors; (4). Update W in the first step to the group of fusion parameter vectors with the highest benefit among W1, W2, W3,..., Wn, and update μ and σ to μ' and σ' respectively; (5) Repeat steps (1) to (4). Through the above steps, online adjustment of multi-objective fusion parameters can be achieved.

[0042] In the aforementioned parameter tuning process, new parameter vector sets are explored through random sampling. For example, based on the initial fusion parameter vector W = (a, b, c), random sampling yields a new parameter vector W1 = (a1, b1, c1). Here, a1 may be greater than a (hereinafter referred to as the positive direction) or less than a (hereinafter referred to as the negative direction), with a probability of 0.5 in both cases. This indicates that the exploration direction is uncertain. Furthermore, a fusion parameter vector containing a, b, and c can form eight different combinations of sampling directions. Theoretically, there exists an optimal parameter vector W0 = [a0, b0, c0] with the highest return, where the directions of each parameter are fixed relative to W. However, as shown by W1 above, the exploration vector generated through the aforementioned parameter tuning process has only a one-eighth probability of sharing the same direction as the optimal parameter vector, reducing the effectiveness of the exploration. In addition, the distance between each parameter in the exploration parameter vector and the initial fusion parameter vector (i.e., the update step size γ = (γa, γb, γc), γa = a1-a, γb = b1-b, γc = c1-c) is related to σ. σ is calculated based on the k sets of fusion parameter vectors with the highest returns. These k sets of fusion parameter vectors may get stuck in local optima, resulting in a limited update step size and an inability to escape local optima. This makes it impossible for the parameter tuning algorithm to find better sets of fusion parameter vectors, thus affecting the recommendation accuracy of the recommendation system.

[0043] To address the aforementioned problems, embodiments of this application provide a method, apparatus, device, and computer storage medium for multi-objective parameter tuning, which can improve the efficiency and speed of fused parameter vector optimization and achieve more accurate material recommendation. The following describes exemplary applications of the electronic device provided in this application. The device provided in this application can be implemented as various types of user terminals such as laptops, tablets, desktop computers, and mobile devices (e.g., mobile phones, wearable smartwatches, dedicated messaging devices), or as a server. Exemplary applications when the electronic device is implemented as a server will be described below.

[0044] See Figure 1 , Figure 1 This is a schematic diagram of an architecture of a recommendation system provided in the embodiments of this application. In order to implement the parameter tuning method described in the embodiments of this application, electronic devices (electronic devices 400-1 and 400-2 are shown as examples) are connected to server 200 through network 300. Network 300 can be a wide area network or a local area network, or a combination of the two.

[0045] In some possible implementations, user A can publish material information A through electronic device 400-1, and user B can publish material information B through electronic device 400-2. Material information A and material information B are uploaded to server 200 via network 300. Server 200 can store material information A and material information B in database 500. To achieve accurate recommendations for users, server 200 can process the information using a neural network to recommend material information that best matches the user's interests. The recommended material information can be displayed on the graphical interface 410 of electronic device 400 (graphical interfaces 410-1 and 410-2 are shown as examples).

[0046] In some embodiments, server 200 may be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms. Electronic device 400 may be a smartphone, tablet, laptop, desktop computer, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment of the invention.

[0047] See Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 2 The illustrated electronic device 400 includes at least one processor 410, a memory 450, at least one network interface 420, and a user interface 430. The various components in the electronic device 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to implement communication between these components. In addition to a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general designated all buses as Bus System 440.

[0048] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0049] User interface 430 includes one or more output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. User interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0050] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices physically located away from the processor 410.

[0051] The memory 450 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 450 described in this application embodiment is intended to include any suitable type of memory.

[0052] In some embodiments, memory 450 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0053] Operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, for implementing various basic business functions and handling hardware-based tasks.

[0054] The network communication module 452 is used to reach other computing devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including Bluetooth, WiFi, and Universal Serial Bus (USB).

[0055] Presentation module 453 enables the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 associated with user interface 430 (e.g., a display screen, a speaker, etc.).

[0056] The input processing module 454 is used to detect and translate one or more user inputs or interactions from one or more input devices 432.

[0057] In some embodiments, the apparatus provided in this application can be implemented in software. Figure 2 A multi-target parameter tuning device 455, stored in memory 450, is shown. This device can be software in the form of programs or plug-ins. The parameter tuning device 455 includes the following software modules: a data acquisition module 4551, a data processing module 4552, and an update module 4553. These modules are logically linked and can therefore be arbitrarily combined or further separated according to the functions they implement. The functions of each module will be described below.

[0058] In other embodiments, the parameter tuning device provided in this application can be implemented in hardware. As an example, the device provided in this application can be a processor in the form of a hardware decoding processor, which is programmed to execute the parameter tuning method provided in this application. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0059] The parameter tuning method provided in this application will be described below with reference to exemplary applications and implementations of the electronic devices provided in the embodiments of this application.

[0060] Before describing the parameter tuning method for multiple objectives as described in the embodiments of this application, the flow of the parameter tuning method for multiple objectives in the information flow of the recommendation system in the embodiments of this application will be explained first. See Figure 3 , Figure 3 This is a schematic diagram of the information flow recommendation process of the recommendation system provided in the embodiments of this application. The following will be combined with... Figure 3 The recommendation process of the illustrated recommendation system will be explained.

[0061] It should be noted that the targets mentioned in the embodiments of this application are recommended targets.

[0062] S101, the recommendation system receives material recommendation requests from users.

[0063] S102, the recommendation system selects a batch of materials that match the user's interests from a large number of candidate materials based on the user's personal interests and preferences.

[0064] S103, the recommendation system predicts the click-through rate, interaction rate and viewing time of the selected materials, and obtains the predicted values ​​of click-through rate, interaction rate and viewing time.

[0065] S104, the recommendation system obtains the fusion prediction value S by weighted summation based on the predicted click-through rate, predicted interaction rate, predicted viewing time, and the fusion parameter vector (i.e., W = (a, b, c)). That is: S = a × predicted click-through rate + b × predicted interaction rate + c × predicted viewing time.

[0066] S105, the recommendation system sorts each candidate material according to the fusion estimate S, and selects a specified number of recommended materials from the largest to the smallest score.

[0067] S106, the recommendation system exposes the sorted and filtered candidate materials to the user.

[0068] S107, the recommendation system associates and saves information such as the user's behavior towards recommended materials and the user's corresponding fusion parameter vector.

[0069] S108, the recommendation system updates the fusion parameter vector based on the collected logs and according to the parameter tuning method for multiple targets provided in the embodiments of this application, and calculates a new fusion parameter vector for use in the next target fusion.

[0070] It should be noted that the aforementioned candidate materials are those that have not yet been recommended to users, that is, materials that are still in the candidate pool and will be recommended to users soon; while recommended materials are those that have already been recommended to users and have generated user behavior.

[0071] See also some possible implementations. Figure 4 , Figure 4 This is a flowchart illustrating a multi-target parameter tuning method (hereinafter referred to as the parameter tuning method) provided in an embodiment of this application. Figure 4 The parameter tuning method for multiple targets shown is as follows: Figure 3 S108 in the recommendation process of the recommendation system in the image will be discussed below. Figure 4 Steps S201 to S203 are described below.

[0072] S201 collects historical behavior information of multiple users regarding recommended materials.

[0073] In some embodiments, when adjusting the parameters of the fusion network (i.e., the fusion parameter vector), the parameter tuning device collects historical behavior information of multiple users regarding recommended materials from the behavior logs of these users. In one embodiment, the historical behavior information may include, but is not limited to, clicks, interactions, and viewing duration. Each user corresponds to N sets of fusion parameter vectors; that is, each user has a corresponding fusion parameter vector. Because the number of users is much larger, the mapping relationship between multiple users and N sets of fusion parameter vectors is typically many-to-one.

[0074] In one embodiment, the fusion network in the recommendation system fuses the predicted values ​​of x targets of the candidate material based on the above-mentioned fusion parameter vector to obtain the fusion prediction value of the candidate material, which is stored in the database of the server. When the ranking model ranks the candidate materials, it directly obtains the above-mentioned fusion prediction value from the database and recommends materials to the user based on the fusion prediction value.

[0075] In some possible implementations, before step S201 is executed, the parameter tuning device may also obtain N sets of fusion parameter vectors based on preset mean vectors and variance vectors.

[0076] In some embodiments, the parameter tuning device can perform Gaussian random sampling on preset mean vectors and variance vectors to obtain N sets of fused parameter vectors. For a detailed description, please refer to step (1) in the parameter tuning process described above.

[0077] In one example, suppose the recommendation system fuses three objectives (x = 3) for candidate items, where the objectives are click-through rate, engagement rate, and viewing time. The mean vector is initialized to μ = (μ a ,μ b ,μ c ), where μ a =2,μ b =2,μ c =2; the initial value of the variance vector is σ = (σ a ,σ b ,σ c ), where σ a =0.04, σ b =0.04, σ c =0.04, perform Gaussian random sampling on the mean vector and variance vector to obtain N sets of fusion parameter vectors, where the fusion parameter vectors are W = (a, b, c). Then, the N sets of fusion parameter vectors are W1, W2, W3, ..., W... n In the above vector, a corresponds to the click-through rate, b corresponds to the interaction rate, and c corresponds to the viewing time.

[0078] In some possible implementations, the parameter tuning device can fuse the predicted values ​​of x targets for candidate materials in a weighted summation manner based on the fusion parameter vector to obtain a fused estimate; wherein, the expression for calculating the fused estimate is as follows:

[0079] S = a × predicted click-through rate + b × predicted interaction rate + c × predicted viewing time (3)

[0080] In the formula, S is the estimated value of the fusion, a is the fusion parameter of the click-through rate, b is the fusion parameter of the interaction rate, and c is the fusion parameter of the viewing time.

[0081] It should be noted that the above expression (3) is an example when the target of the candidate material is click-through rate, interaction rate and viewing time, but it is not limited to the above three targets. The expression for calculating the fusion prediction value is to multiply multiple fusion parameters by the predicted value of the target corresponding to the fusion parameter, and then add the results of these multiple multiplications together.

[0082] S202, based on historical behavior information, determine the revenue parameter vector corresponding to each of the N sets of fusion parameter vectors in the fusion network. The fusion network then fuses the predicted values ​​of x targets for candidate materials based on the fusion parameter vectors. Here, the revenue parameter vector represents the user's level of interest in the recommended materials, and N and x are integers greater than or equal to 2.

[0083] In some embodiments, after obtaining the fusion estimate of x targets for candidate materials calculated by the fusion network, the recommendation system recommends materials to the user based on the fusion estimate. At this time, the parameter tuning device can determine the revenue parameter vector corresponding to the fusion parameter vector based on the user's behavioral information regarding the recommended materials (which can be obtained from user behavior logs). In one embodiment, the revenue parameter vector may include, but is not limited to, parameters such as the user's click-through rate, interaction rate, and viewing time for the recommended materials within a preset time period; the revenue parameter vector can represent the user's degree of interest in the recommended materials.

[0084] In one example, assume there are 6 users: A, B, C, D, E, and F. The number of fusion parameter vectors is 3: W1, W2, and W3. Users A and B correspond to W1, users C and D to W2, and users E and F to W3. After the recommendation system makes recommendations to these users based on the fusion parameter vectors, the parameter tuning device in the system can collect user behavior logs for users A, B, C, D, E, and F to obtain their click-through rate, interaction rate, and viewing time for recommended materials within 7 days, thus obtaining the corresponding revenue parameter vectors. The revenue parameter vector corresponding to W1 is r1 = (r... a1 ,rb2 ,r c1 ), r a1 The click-through rate (r) of recommended materials for users A and B within 7 days. b1 r represents the interaction rate between user A and user B with the recommended materials within 7 days. c1 Let W2 be the viewing time of recommended materials by users A and B within 7 days, and the corresponding revenue parameter vector is r2 = (r a2 ,r b2 ,r c2 ), r a2 The click-through rate (r) of recommended materials for users C and D within 7 days. b2 r represents the interaction rate of users C and D with the recommended materials within 7 days. c2 Let W3 be the viewing time of recommended materials by users C and D within 7 days, and the corresponding revenue parameter vector is r3 = (r a3 ,r b3 ,r c3 ), r a3 The click-through rate (r) of recommended materials for users E and F within 7 days. b3 r represents the interaction rate of users E and F with the recommended materials within 7 days. c3 The viewing time of recommended materials for users E and F within 7 days.

[0085] S203, determine the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector.

[0086] In some embodiments, after determining the revenue parameter vector, the parameter tuning device can perform linear correlation calculations on each set of fusion parameter vectors and the revenue parameter vectors corresponding to each set of fusion parameter vectors to determine the correlation coefficient between each set of fusion parameter vectors and the revenue parameter vectors corresponding to each set of fusion parameter vectors.

[0087] In some possible implementations, S203 above may include: calculating the Pearson correlation coefficient for x parameters in each set of fusion parameter vectors and x parameters in the corresponding benefit parameter vectors of each set of fusion parameter vectors to obtain x correlation arrays, wherein each of the x correlation arrays includes x correlation coefficients.

[0088] In one example, the fusion parameter vector is W1 = (a, b, c), and the corresponding profit parameter vector is r1 = (r a1 ,r b1 ,r c1 The parameter tuning device calculates the Pearson correlation coefficient for the parameters in W1 and r1 using the following formula:

[0089]

[0090] In the formula, ρ (X,Y) Let X be the correlation coefficient between W1 and r1, where X takes any parameter from (a, b, c), and Y takes a value from (r1, b, c). a1 ,r b1 ,r c1 Any parameter in ), cov(X,Y) is the covariance of X and Y, σ X σ Y It is the product of the standard deviations of X and Y.

[0091] Thus, three correlation arrays can be obtained through formula (4), where each correlation array contains three correlation coefficients, namely... and

[0092] It should be noted that in this embodiment of the application, the Pearson correlation coefficient is introduced to measure the correlation between the fusion parameter vector and the return parameter vector. The obtained correlation coefficient value is between 1 and -1. The larger the correlation coefficient, the higher the correlation. A correlation coefficient value greater than 0 indicates a positive correlation, and a correlation coefficient value less than 0 indicates a negative correlation.

[0093] S204, determine the incremental vector between the first revenue parameter vector and each second revenue parameter vector in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors.

[0094] Among them, the first revenue parameter vector is the optimal revenue parameter vector among the N groups of revenue parameter vectors, and the second revenue parameter vector is the revenue parameter vector of the other groups besides the optimal revenue parameter vector.

[0095] In some embodiments, the optimal set of revenue parameter vectors among the N sets of revenue parameter vectors can be denoted as the first revenue parameter vector, and the other N-1 sets of revenue parameter vectors can be denoted as the second revenue parameter vectors. After determining the first revenue parameter vector and the second revenue parameter vector, the parameter tuning device calculates the N-1 sets of incremental vectors of the N-1 sets of the second revenue parameter vectors relative to the first revenue parameter vector.

[0096] In some possible implementations, the parameter tuning device can determine the first revenue parameter vector and the second revenue parameter vector based on the difference between the parameters of each of the N revenue parameter vectors.

[0097] In one example, suppose there are four revenue parameter vectors, namely r1 = (0.85, 0.95, 3), r2 = (0.93, 0.85, 2.5), r3 = (0.89, 0.92, 3), and r4 = (0.82, 0.92, 2). The first parameter in the revenue parameter vector represents the click-through rate of users on the recommended materials within 7 days, the second parameter represents the interaction rate of users on the recommended materials within 7 days, and the third parameter represents the viewing time of users on the recommended materials within 7 days. The parameter tuning device first obtains r1 and r2, and calculates the differences between various parameters in r1 and r2. The differences between r1 and r2 are: click-through rate difference = 0.08, interaction rate difference = 0.1, and viewing duration difference = 0.5. The recommended target with the largest difference is viewing duration, so r1 with the largest viewing duration is used as the first revenue parameter vector. The parameter tuning device continues to calculate the differences between various parameters in r1 and r3, obtaining the following differences: click-through rate difference = 0.04, interaction rate difference = 0.03, and viewing duration difference = 0.5. The difference between r3 and r4 is 0. The recommended target with the largest difference is click-through rate (CTR). Therefore, r3 with the larger CTR is taken as the new first revenue parameter vector. The parameter tuning device continues to calculate the differences between each parameter in r3 and r4. We can obtain the following results for r3 and r4: the difference in CTR is 0.07, the difference in interaction rate is 0, and the difference in viewing time is 1. The recommended target with the largest difference is viewing time. Therefore, r3 with the larger viewing time is still the first revenue parameter vector. The differences between r1, r2, and r4, which are smaller than the differences between r3, are all second revenue parameter vectors.

[0098] In some possible implementations, S204 may include: for any set of second revenue parameter vectors, calculating the increments of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector to obtain an increment vector between the first and second revenue parameter vectors.

[0099] For example, the increment vector can be obtained by the following expression (1):

[0100]

[0101] In the formula, Δr i Let r be the i-th increment vector, 1≤i≤N-1. ai ,r bi ,…,r xi Let x be the parameters in the i-th second benefit parameter vector, and r a0 ,r b0 ,…,r x0 Let x be the parameters in the first benefit parameter vector, [Δr ai ,Δr bi ,…,Δr xi ] represents the x parameters in the i-th increment vector.

[0102] S205, based on each set of incremental vectors, update the mean vector of the fusion parameter vector corresponding to each set of second benefit parameter vectors.

[0103] In some embodiments, after calculating N-1 sets of incremental vectors, the parameter tuning device can update the N-1 sets of mean vectors based on these N-1 sets of incremental vectors. The N-1 sets of mean vectors correspond to the N-1 sets of fusion parameter vectors corresponding to the second revenue parameter vector. The i-th mean vector in the updated N-1 sets of mean vectors can be represented as μ. i '=(μ' ai ,μ' bi ,…,μ' xi ).

[0104] In some possible implementations, S205 may include S301 to S302, see [link to implementation details]. Figure 5 , Figure 5 This is an optional flowchart illustrating a multi-target parameter tuning method provided in an embodiment of this application. The following will be combined with... Figure 5 The steps shown are explained.

[0105] S301: For any set of incremental vectors, collect the values ​​of x parameters in the incremental vector.

[0106] In some embodiments, after obtaining N-1 sets of incremental vectors, the parameter tuning device can collect the values ​​of x parameters in each of the N-1 sets of incremental vectors.

[0107] S302, subtract the minimum value from the maximum value among the x parameters in the increment vector to obtain the updated mean vector.

[0108] In some embodiments, after collecting the values ​​of x parameters in each set of incremental vectors, the parameter tuning device first determines the maximum and minimum values ​​of the x parameters in each set of incremental vectors, and subtracts the minimum value from the maximum value to obtain the difference, which is the value of the x parameters in the updated mean vector.

[0109] In some possible implementations, the values ​​of the x parameters in the updated mean vector can be calculated using expression (5):

[0110] μ' i =(μ' ai ,μ' bi ,…,μ' xi )=(Δr imax -Δr imax ,Δr imax -Δr imax ,…,Δr imax -Δr imax(5)

[0111] In the formula, μ' i Let μ' be the mean vector of the i-th group in the updated N-1 groups of mean vectors, where 1 ≤ i ≤ N-1. ai ,μ' bi ,…,μ' xi Let x be the values ​​of the updated i-th group of mean vectors, and Δr be the value of each parameter. imax For Δr i The maximum value in, Δr imin For Δr i The minimum value in.

[0112] In this embodiment, determining the minimum value in the incremental vector is equivalent to determining the target type with the smallest data increase, and determining the maximum value is equivalent to determining the target type with the largest data increase. The target type can be, but is not limited to, any of the following: click-through rate, interaction rate, and viewing time. Currently, the principle for updating the fusion parameter vector is to maximize the target type with the smallest data increase and minimize the target type with the largest data increase. Updating the mean vector based on the maximum and minimum values ​​in the incremental vector can satisfy the above principle for updating the fusion parameter vector as much as possible.

[0113] S206, Update N sets of fusion parameter vectors based on the updated mean vector and correlation coefficient.

[0114] In some embodiments, the parameter tuning device can adjust the fusion parameters in the N sets of fusion parameter vectors (i.e., update the N sets of fusion parameter vectors) based on the updated mean vector and correlation coefficient.

[0115] In some possible implementations, the first revenue parameter vector corresponds to the first fusion parameter vector in N sets of fusion parameter vectors, and each set of second revenue parameter vectors corresponds to the other sets of second fusion parameter vectors in N sets of fusion parameter vectors besides the first fusion parameter vector; S206 may include S401 to S402, see [link to relevant documentation]. Figure 6 , Figure 6 This is an optional flowchart illustrating a multi-target parameter tuning method provided in an embodiment of this application. The following will be combined with... Figure 6 The steps shown are explained.

[0116] S401, based on the updated mean vector and x correlation arrays, calculate the values ​​of x parameters in each of the updated N-1 sets of second fusion parameter vectors.

[0117] In some embodiments, the parameter tuning device obtains the updated mean vector in S205, and after obtaining x correlation arrays in S203, it can update the second fusion parameter vector (i.e. update the values ​​of x parameters in each group of the second fusion parameter vector) based on the updated mean vector and the x correlation arrays.

[0118] In some possible implementations, the above S401 can be achieved by expression (2):

[0119]

[0120] In the formula, a i '、b i '、…、x i ' represents the values ​​of x parameters in the i-th group of the updated N-1 groups of second fusion parameter vectors, a i b i ... x i Let x be the values ​​of the i-th parameter in the N-1 sets of second fusion parameter vectors, and μ be the value of the parameter in the i-th set of second fusion parameter vectors. ai ',μ bi ',…,μ xi ' represents the values ​​of x parameters in the updated i-th group of mean vectors, ρ (a,r) For each set of fusion parameter vectors, ρ represents the correlation coefficient set of the first parameter and the x parameters in the corresponding return parameter vector. (b,r) For each set of fusion parameter vectors, ρ represents the correlation coefficient set of the second parameter and the x parameters in the corresponding return parameter vector. (x,r) Let r be the correlation coefficient set between the x-th parameter in each set of fusion parameter vectors and the x-th parameters in the corresponding profit parameter vector, where r min r is the minimum value among the x parameters of the revenue parameter vector. max It is the maximum value among the x parameters of the revenue parameter vector.

[0121] In this embodiment, the direction of updating the fusion parameter vector is determined by the difference between the correlation coefficient between the minimum value of the fusion parameter and the maximum value of the revenue parameter vector in the fusion parameter vector, thus achieving certainty in the exploration direction. For example, assuming the target type with the smallest current data increase is click-through rate (CTR) and the target type with the largest data increase is viewing time, the principle is to increase CTR or decrease viewing time. When the fusion parameter (i.e., 'a') corresponding to CTR in the fusion parameter vector is updated, the correlation array related to 'a' needs to be updated. Assuming the correlation coefficient is greater than 0, the direction for updating the fusion parameter vector is positively correlated; the larger 'a' is, the higher the click-through rate (CTR). Therefore, increasing 'a' will increase the CTR. When considering the correlation coefficient between a and viewing duration (i.e., ρ) (a,rc) When 'a' is used as its direction, assuming the corresponding correlation coefficient is greater than 0, then it is a positive correlation. The larger 'a' is, the longer the consumption time. Decreasing 'a' can decrease the consumption time. When considering both the target type with the smallest data increase and the target type with the largest data increase, then Ultimately, the direction of 'a' update depends on The direction is the difference between the correlation coefficients of 'a' and the target type with the smallest data increase and the target type with the largest data increase.

[0122] In this embodiment, the variables affecting the update step size of the fusion parameter vector are μ' and in, The primary function of μ' is to explore the direction of updates, while μ' is a crucial factor influencing the update step size. According to S302, the value of μ' depends on the maximum and minimum values ​​of the x parameters in the increment vector. The difference between the maximum and minimum values ​​of the x parameters in the increment vector determines a more suitable update step size. A larger difference indicates a greater disparity in the returns generated by the two sets of parameters (i.e., the parameters corresponding to the maximum and minimum values ​​mentioned above), suggesting a significant difference in the distribution of the two sets of parameters, thus requiring a larger update step size. Therefore, the size of each update step size depends on the difference between the maximum and minimum values ​​in the increment vector, and the step size can be flexibly adjusted to avoid the parameter set getting trapped in local optima.

[0123] S402, based on the updated second fusion parameter vectors and the first fusion parameter vectors of each group, obtain the updated N groups of fusion parameter vectors.

[0124] In some embodiments, the parameter tuning device combines the updated N-1 sets of second fusion parameter vectors and the first fusion parameter vectors to obtain updated N sets of fusion parameter vectors. When the recommendation system recommends materials to multiple users in the next instance, it will calculate the fusion estimate of the candidate materials based on the updated N sets of fusion parameter vectors.

[0125] At this point, the parameter tuning device has completed the parameter tuning process for multiple targets.

[0126] The following will describe an exemplary application of the parameter tuning method of this application in a real-world application scenario.

[0127] When the above parameter tuning method is applied to the server of a social application (such as a Weibo server), the implementation process of applying the above parameter tuning method to Weibo is as follows: Figure 7 . Figure 7This is a flowchart illustrating the parameter tuning method provided in this application when applied to social application software. The following will combine... Figure 7 The steps shown are explained.

[0128] In the embodiments of this application, Figure 7 The use case shown is that the recommendation system adjusts three sets of fusion parameter vectors (W1, W2, and W3, respectively, which are mapped to users A, B, C, D, E, and F, respectively, based on the historical behavior information in the behavior logs of users A, B, C, D, E, and F. Where users A and B correspond to W1, users C and D correspond to W2, and users E and F correspond to W3) based on the historical behavior information in the behavior logs of users F. Then, based on these three sets of fusion parameter vectors, it calculates the fusion estimate of the predicted click-through rate, predicted interaction rate, and predicted viewing time of candidate materials, and recommends materials to the above users based on the fusion estimate.

[0129] It should be noted that, Figure 7 The parameter tuning method shown is merely one example of an embodiment of this application and should not be used to limit other implementations of this application. Furthermore, Figure 7 The fusion network, parameter tuning device, and sorting device mentioned in the text are all software modules in the recommendation system.

[0130] S501, the parameter adjustment device generates three sets of fusion parameter vectors, denoted as W1, W2, and W3, based on the preset mean vector μ and variance vector σ.

[0131] S502, the fusion network calculates the fusion estimate of candidate materials for users A and B, users C and D, and users E and F based on W1, W2 and W3.

[0132] S503, the sorting device recommends materials to the user based on the fusion estimate of the candidate materials.

[0133] S504, the parameter adjustment device collects user logs.

[0134] S505, the parameter tuning device calculates the revenue parameter vectors corresponding to W1, W2, and W3 based on the user behavior information in the user log, denoted as r1, r2, and r3.

[0135] S506, the parameter adjustment device calculates the correlation coefficients between W1 and r1, W2 and r2, and W3 and r3, and obtains three correlation arrays, each containing three correlation coefficients.

[0136] S507, the parameter tuning device determines r1 as the optimal benefit parameter vector, and calculates the incremental vectors of r2 and r3 relative to r1, where W1 corresponding to r1 is the optimal fusion parameter vector.

[0137] S508, the parameter tuning device updates the mean vector based on the increment vector in S507, and the updated mean vector is μ'.

[0138] S509, the parameter adjustment device updates W2 and W3 according to μ' in S508 and the correlation coefficient in S506, to obtain W'2 and W'3.

[0139] S510, the parameter tuning device updates the fusion parameter vector in the fusion network to W1, W'2 and W'3.

[0140] S511, in response to the updated fusion parameter vectors W1, W'2 and W'3, repeat S502 to S510.

[0141] This completes the above-described method for parameter tuning for multiple objectives.

[0142] This application's embodiments update the mean vector by calculating the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector, and by calculating the increment vectors of the first and second revenue parameter vectors among the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors. This can improve the target type with the smallest data increase and suppress the target type with the largest data increase, thus achieving certainty in the exploration direction. In addition, based on the updated mean vector and correlation coefficient, the N sets of fusion parameter vectors are updated, so that the update step size of the fusion parameters depends on the updated mean vector and correlation coefficient. The update step size of the fusion parameters can be flexibly adjusted to avoid the parameter group getting stuck in a local optimum, thereby making the updated fusion parameter vectors better reflect the user's interest tendencies and achieving more accurate material recommendations.

[0143] The following continues to describe the exemplary structure of the multi-target parameter tuning device 455 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 2As shown, the software modules stored in the multi-target parameter tuning device 455 in the memory 450 may include: a data acquisition module 4551, used to collect historical behavior information of multiple users towards recommended materials; a data processing module 4552, used to determine the revenue parameter vector corresponding to each of the N sets of fusion parameter vectors in the fusion network based on the historical behavior information; the fusion network fuses the predicted values ​​of x targets of the candidate materials based on the fusion parameter vectors; the revenue parameter vector is used to represent the user's interest in the recommended materials; N and x are integers greater than or equal to 2; and determine the relationship between each set of fusion parameter vectors and the target target. The correlation coefficient of the revenue parameter vector corresponding to each set of fusion parameter vectors is determined; the increment vector between the first revenue parameter vector and each set of second revenue parameter vectors in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors is determined, where the first revenue parameter vector is the optimal revenue parameter vector in the N sets of revenue parameter vectors, and the second revenue parameter vectors are the revenue parameter vectors of other sets besides the optimal revenue parameter vector; the update module 4553 is used to update the mean vector of the fusion parameter vector corresponding to each set of second revenue parameter vectors based on each set of increment vectors; and the N sets of fusion parameter vectors are updated according to the updated mean vector and correlation coefficient.

[0144] In some possible implementations, the data processing module 4552 is further configured to calculate the Pearson correlation coefficient for x parameters in each set of fusion parameter vectors and x parameters in the corresponding benefit parameter vectors of each set of fusion parameter vectors, to obtain x correlation arrays, wherein each of the x correlation arrays includes x correlation coefficients.

[0145] In some possible implementations, the data processing module 4552 is further configured to, for any set of second revenue parameter vectors, calculate the increment of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector to obtain an increment vector between the first revenue parameter vector and the second revenue parameter vector.

[0146] In some possible implementations, the increment vector is obtained by calculating the increments of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector, satisfying expression (1).

[0147] In some possible implementations, the data acquisition module 4551 is further configured to acquire the values ​​of x parameters in any set of incremental vectors; the data processing module 4552 is further configured to subtract the minimum value from the maximum value among the x parameters in the incremental vector to obtain the updated mean vector.

[0148] In some possible implementations, the first benefit parameter vector corresponds to the first fusion parameter vector in the N sets of fusion parameter vectors, and each set of second benefit parameter vectors corresponds to each set of second fusion parameter vectors in the N sets of fusion parameter vectors other than the first fusion parameter vector; the data processing module 4552 is further configured to calculate the values ​​of x parameters in each set of updated second fusion parameter vectors based on the updated mean vector and x correlation arrays; the update module 4553 is further configured to obtain the updated N sets of fusion parameter vectors based on the updated sets of second fusion parameter vectors and the first fusion parameter vector.

[0149] In some possible implementations, the values ​​of x parameters in the updated N-1 sets of second fusion parameter vectors are calculated to satisfy expression (2) based on the updated mean vector and x correlation arrays.

[0150] This application provides a computer program product or computer program that includes computer instructions stored in a computer storage medium. A processor of a computer device reads the computer instructions from the computer storage medium and executes the computer instructions, causing the computer device to perform the parameter tuning method described above in this application.

[0151] This application provides a computer storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to execute the parameter tuning method for multiple targets provided in this application, for example... Figure 3 The parameter tuning method for multiple objectives is shown.

[0152] In some embodiments, the computer storage medium may be a memory such as FRAM, ROM, PROM, EEPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0153] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0154] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hypertext Markup Language (TML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files that store one or more modules, subroutines, or code sections).

[0155] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0156] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A parameter tuning method for multiple objectives, applied to a recommendation system, characterized in that, The method includes: Collect historical behavioral information of multiple users regarding recommended materials; the historical behavioral information includes: clicks, interactions, and viewing duration. Based on the historical behavior information, a revenue parameter vector is determined for each of the N sets of fusion parameter vectors in the fusion network. The fusion network fuses the predicted values ​​of x targets of the candidate material based on the fusion parameter vector. The revenue parameter vector is used to represent the user's interest in the recommended material. N and x are integers greater than or equal to 2. Determine the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector; Determine the incremental vector between the first revenue parameter vector and each group of second revenue parameter vectors in the N groups of revenue parameter vectors corresponding to the N groups of fusion parameter vectors. The first revenue parameter vector is the optimal revenue parameter vector in the N groups of revenue parameter vectors, and the second revenue parameter vector is the other groups of revenue parameter vectors besides the optimal revenue parameter vector. Based on each set of incremental vectors, update the mean vector of the fusion parameter vector corresponding to each set of second benefit parameter vectors; The N sets of fusion parameter vectors are updated based on the updated mean vector and the correlation coefficient; Wherein, determining the correlation coefficient between each set of fusion parameter vectors and the corresponding revenue parameter vector includes: Pearson correlation coefficients are calculated for x parameters in each set of fusion parameter vectors and x parameters in the corresponding benefit parameter vectors of each set of fusion parameter vectors to obtain x correlation arrays, wherein each of the x correlation arrays includes x correlation coefficients.

2. The method according to claim 1, characterized in that, Determining the incremental vector between the first revenue parameter vector and each second revenue parameter vector in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors includes: For any set of second revenue parameter vectors, the increment vector between the first revenue parameter vector and the second revenue parameter vector is obtained by calculating the increment of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector.

3. The method according to claim 2, characterized in that, For any set of second revenue parameter vectors, by calculating the increments of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector, the increment vector between the first and second revenue parameter vectors is obtained, satisfying the following expression: In the formula, Δr i Let r be the i-th increment vector, 1≤i≤N-1. ai ,r bi ,…,r xi Let x be the parameters in the second revenue parameter vector of the i-th group, and r be the parameters in the second revenue parameter vector. a0 ,r b0 ,…,r x0 Let x be the parameters in the first benefit parameter vector, [Δr ai ,Δr bi ,…,Δr xi ] represents the x parameters in the i-th group of increment vectors.

4. The method according to claim 1, characterized in that, The step of updating the mean vector of the fusion parameter vector corresponding to each set of second benefit parameter vectors based on each set of incremental vectors includes: For any set of incremental vectors, collect the values ​​of x parameters in the incremental vectors; The updated mean vector is obtained by subtracting the minimum value from the maximum value among the x parameters in the increment vector.

5. The method according to claim 1, characterized in that, The first benefit parameter vector corresponds to the first fusion parameter vector in the N sets of fusion parameter vectors, and each set of second benefit parameter vectors corresponds to each set of second fusion parameter vectors in the N sets of fusion parameter vectors other than the first fusion parameter vector. The step of updating the N sets of fusion parameter vectors based on the updated mean vector and the correlation coefficient includes: Based on the updated mean vector and the x correlation arrays, calculate the values ​​of x parameters in each updated second fusion parameter vector; Based on the updated second fusion parameter vectors and the first fusion parameter vectors, the updated N sets of fusion parameter vectors are obtained.

6. The method according to claim 5, characterized in that, The step of calculating the values ​​of x parameters in each updated second fusion parameter vector based on the updated mean vector and the x correlation arrays satisfies the following expression: In the formula, a i '、b i '、…、x i ' represents the values ​​of x parameters in the updated i-th group of the second fusion parameter vector, 1≤i≤N-1, a i b i ... x i Let x be the values ​​of the second fusion parameter vector in the i-th group, and μ be the values ​​of the x parameters. ai ',μ bi ',…,μ xi ' represents the values ​​of x parameters in the updated i-th group of mean vectors, ρ (a,r) For each set of fusion parameter vectors, ρ represents the correlation coefficient set between the first parameter and x parameters in the corresponding revenue parameter vector. (b,r) For each set of fusion parameter vectors, ρ represents the correlation coefficient set between the second parameter and x parameters in the corresponding return parameter vector. (x,r) Let r be the correlation coefficient set between the x-th parameter in each set of fusion parameter vectors and the x-th parameters in the corresponding revenue parameter vector of each set of fusion parameter vectors, where r min r is the minimum value among the x parameters of the revenue parameter vector. max It is the maximum value among the x parameters of the revenue parameter vector.

7. A parameter tuning device for multiple objectives, installed in a recommendation system, characterized in that, The device includes: The data acquisition module is used to collect historical behavior information of multiple users on recommended materials; the historical behavior information includes: clicks, interactions, and viewing duration; The data processing module is used to determine, based on the historical behavior information, the revenue parameter vector corresponding to each of the N sets of fusion parameter vectors in the fusion network, wherein the fusion network fuses the predicted values ​​of x targets of the candidate material based on the fusion parameter vectors, and the revenue parameter vector is used to represent the user's interest in the recommended material, where N and x are integers greater than or equal to 2; determine the correlation coefficient between each set of fusion parameter vectors and the revenue parameter vector corresponding to each set of fusion parameter vectors; and determine the incremental vector between the first revenue parameter vector and each second revenue parameter vector in the N sets of revenue parameter vectors corresponding to the N sets of fusion parameter vectors, wherein the first revenue parameter vector is the optimal revenue parameter vector in the N sets of revenue parameter vectors, and the second revenue parameter vector is the other sets of revenue parameter vectors besides the optimal revenue parameter vector. The update module is used to update the mean vector of the fusion parameter vector corresponding to each group of second benefit parameter vectors based on each group of incremental vectors; and to update the N groups of fusion parameter vectors according to the updated mean vector and the correlation coefficient. Specifically, the data processing module is used to calculate the Pearson correlation coefficient for x parameters in each set of fusion parameter vectors and x parameters in the corresponding revenue parameter vectors of each set of fusion parameter vectors, to obtain x correlation arrays, wherein each of the x correlation arrays includes x correlation coefficients.

8. The apparatus according to claim 7, characterized in that, The data processing module is specifically used to calculate the increment vector between the first and second revenue parameter vectors by calculating the increment of x parameters in the second revenue parameter vector relative to x parameters in the first revenue parameter vector for any set of second revenue parameter vectors.

9. The apparatus according to claim 7, characterized in that, The data acquisition module is specifically used to acquire the values ​​of x parameters in any given set of incremental vectors. The data processing module is specifically used to subtract the minimum value from the maximum value among the x parameters in the incremental vector to obtain the updated mean vector.

10. The apparatus according to claim 7, characterized in that, The first benefit parameter vector corresponds to the first fusion parameter vector in the N sets of fusion parameter vectors, and each set of second benefit parameter vectors corresponds to each set of second fusion parameter vectors in the N sets of fusion parameter vectors other than the first fusion parameter vector. The data processing module is specifically used to calculate the values ​​of x parameters in each updated set of second fusion parameter vectors based on the updated mean vector and the x correlation arrays; The update module is specifically used to obtain updated N sets of fusion parameter vectors based on the updated sets of second fusion parameter vectors and the first fusion parameter vector.

11. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; A processor, when executing executable instructions or computer programs stored in the memory, implements the method according to any one of claims 1 to 6.

12. A computer storage medium storing executable instructions or a computer program, characterized in that, When the executable instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images