Object recommendation method, apparatus, device, and storage medium

Abstract

The application provides an object recommendation method and device, equipment and a storage medium, and relates to the technical field of Internet. The method comprises the following steps: determining a current target vector of each to-be-recommended object in a to-be-recommended object set and a previously ranked target vector, wherein the to-be-recommended object set comprises a plurality of to-be-recommended objects; determining an eigenvector of a change matrix of the current target vector and the previously ranked target vector; calculating a click rate of each to-be-recommended object according to target state information of each to-be-recommended object; calculating a ranking weight of each to-be-recommended object according to the eigenvector, the click rate and a maximum norm of the current target vector; ranking each to-be-recommended object in the to-be-recommended set according to the ranking weight of each to-be-recommended object, and performing object recommendation based on the ranking result. The scheme can ensure that the ranking result can better handle the recommendation of some non-high-heat, small-group content and sudden emergency content, and effectively ensure the fairness of ranking all to-be-recommended objects.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and more specifically, to an object recommendation method, apparatus, device, and storage medium. Background Technology

[0002] With the development of information technology and the internet, people have gradually moved from an era of information scarcity to an era of information overload, resulting in a surge in the number of videos. Therefore, it is necessary to effectively sort these massive amounts of videos, prioritizing those that users are likely to interact with, thereby improving the accuracy of recommended content.

[0003] Currently, for content such as internet TV programs—which are massive in volume, relatively traditional in origin, and possess strong traditional media attributes—simple goal ranking is mostly used to sort them. On the other hand, in the internet industry, the ranking of items and content often employs multi-objective optimization ranking techniques to achieve a comprehensive result of ranking and recommendation. Specifically, simple goal ranking involves sequentially ranking based on different objectives; multi-objective optimization ranking schemes typically use deep neural networks to build relevant models, optimizing multiple specified objectives / indicators within the content to be ranked, achieving targeted ranking of program content.

[0004] However, using a multi-objective optimization ranking method will limit the exposure of some less popular, niche, and new content, making it impossible to guarantee the fairness of ranking all video content in Internet TV programs. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing an object recommendation method, apparatus, device, and storage medium to ensure fairness in the ranking of all objects to be recommended.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:

[0007] In a first aspect, embodiments of this application provide an object recommendation method, including:

[0008] Determine the current target vector and the previously sorted target vectors for each object in the set of objects to be recommended; the set of objects to be recommended includes multiple objects to be recommended.

[0009] Determine the eigenvectors of the transformation matrix of the current target vector and the previously sorted target vectors;

[0010] The click-through rate of each object to be recommended is calculated based on the target status information of each object to be recommended.

[0011] Calculate the ranking weight of each of the objects to be recommended based on the feature vector, the click rate, and the maximum norm of the current target vector;

[0012] Based on the ranking weight of each of the objects to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results.

[0013] Optionally, the method further includes:

[0014] Based on the usage records of each object to be recommended in the set of objects to be recommended, determine whether the sorting result is reasonable;

[0015] If not, then object recommendations will be made based on a preset general recommendation strategy.

[0016] Optionally, after calculating the ranking weight of each of the objects to be recommended based on the feature vector, the click-through rate, and the maximum norm of the current target vector, the method further includes:

[0017] Based on the value distribution probability and value change rate corresponding to the current target vector, construct the reward function for the object to be recommended;

[0018] Based on the changing trend of the reward function, determine the target ranking weight of the object to be recommended;

[0019] The step of sorting the objects in the set to be recommended according to their ranking weights, and then recommending objects based on the ranking results, includes:

[0020] Based on the target ranking weight of each of the objects to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results.

[0021] Optionally, determining the target ranking weight of the object to be recommended based on the changing trend of the reward function includes:

[0022] The reward value corresponding to the ranking weight is obtained based on the ranking weight of the object to be recommended;

[0023] If the reward value is greater than or equal to a preset threshold, the ranking weight is used as the target ranking weight of the object to be recommended; otherwise, the current target vector of the object to be recommended is recalculated.

[0024] Optionally, determining the current target vector of each object in the set of objects to be recommended includes:

[0025] Based on the pre-constructed target vector model corresponding to the object to be recommended, the current target vector of the object to be recommended is determined.

[0026] Optionally, determining the eigenvectors of the transformation matrices of the current target vector and the previously sorted target vectors includes:

[0027] Obtain the sorted feature vectors;

[0028] The eigenvectors of the transformation matrix are calculated based on the current target vector and the prior eigenvectors.

[0029] Optionally, calculating the click-through rate of each object to be recommended based on the target state information of each object includes:

[0030] The click-through rate of the object to be recommended is calculated based on the pre-selected target click-through rate model and the historical target status information of the object to be recommended.

[0031] Secondly, embodiments of this application also provide an object recommendation device, the device comprising:

[0032] The determination module is used to determine the current target vector and the pre-sorted target vectors of each object in the set of objects to be recommended, wherein the set of objects to be recommended includes multiple objects to be recommended; and to determine the feature vectors of the transformation matrices of the current target vector and the pre-sorted target vectors.

[0033] The calculation module is used to calculate the click-through rate of each object to be recommended based on the target state information of each object to be recommended; and to calculate the ranking weight of each object to be recommended based on the feature vector, the click-through rate, and the maximum norm of the current target vector.

[0034] The recommendation module is used to sort the objects to be recommended in the set of objects to be recommended according to their ranking weights, and to recommend objects based on the ranking results.

[0035] Optionally, the device further includes:

[0036] The verification module is used to determine whether the ranking result is reasonable based on the usage records of each object in the set of objects to be recommended; if not, object recommendation is performed based on a preset general recommendation strategy.

[0037] Optionally, the device further includes:

[0038] The construction module is used to construct the reward function of the object to be recommended based on the value distribution probability and value change rate corresponding to the current target vector;

[0039] The determining module is further configured to determine the target ranking weight of the object to be recommended based on the changing trend of the reward function;

[0040] The recommendation module is also used for:

[0041] Based on the target ranking weight of each of the objects to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results.

[0042] Optionally, the determining module is further configured to:

[0043] The reward value corresponding to the ranking weight is obtained based on the ranking weight of the object to be recommended;

[0044] If the reward value is greater than or equal to a preset threshold, the ranking weight is used as the target ranking weight of the object to be recommended; otherwise, the current target vector of the object to be recommended is recalculated.

[0045] Optionally, the determining module is further configured to:

[0046] Based on the pre-constructed target vector model corresponding to the object to be recommended, the current target vector of the object to be recommended is determined.

[0047] Optionally, the determining module is further configured to:

[0048] Obtain the sorted feature vectors;

[0049] The eigenvectors of the transformation matrix are calculated based on the current target vector and the prior eigenvectors.

[0050] Optionally, the computing module is further configured to:

[0051] The click-through rate of the object to be recommended is calculated based on the pre-selected target click-through rate model and the historical target status information of the object to be recommended.

[0052] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the method provided in the first aspect above.

[0053] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method provided in the first aspect above.

[0054] The beneficial effects of this application are:

[0055] This application provides an object recommendation method, apparatus, device, and storage medium. The method includes: determining the current target vector of each object to be recommended in a set of objects to be recommended, wherein the set of objects to be recommended includes multiple objects to be recommended; determining the feature vectors of the change matrices of the current target vectors and the previously sorted target vectors; calculating the click-through rate of each object to be recommended based on the target state information of each object to be recommended; calculating the ranking weight of each object to be recommended based on the feature vectors, click-through rates, and the maximum norm of the current target vector; sorting the objects to be recommended in the set of objects to be recommended based on the ranking weights of each object to be recommended, and recommending objects based on the ranking results. In this scheme, to maximize the target effect of the ranking results of each object to be recommended, a ranking weight is calculated based on the maximum norm of the current target vector of each object in the set of objects to be recommended, the eigenvectors of the transformation matrix of the current target vector and the previously ranked target vectors, and the click-through rate of each object. Then, the objects to be recommended in the set of objects to be recommended are ranked according to their ranking weights, and object recommendations are made based on the ranking results. This ensures that the ranking results can better handle the recommendation of some low-popularity, niche, or sudden and urgent content, effectively guaranteeing the fairness of the ranking of all objects to be recommended. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0058] Figure 2 A flowchart illustrating an object recommendation method provided in an embodiment of this application;

[0059] Figure 3 A flowchart illustrating another object recommendation method provided in an embodiment of this application;

[0060] Figure 4 A flowchart illustrating yet another object recommendation method provided in an embodiment of this application;

[0061] Figure 5 A flowchart illustrating another object recommendation method provided in an embodiment of this application;

[0062] Figure 6 A flowchart illustrating yet another object recommendation method provided in an embodiment of this application;

[0063] Figure 7 This is a schematic diagram of the structure of an object recommendation device provided in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0065] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0066] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0067] First, before providing a detailed explanation of the technical solutions provided in this application, a brief explanation of the relevant background will be given.

[0068] Currently, for content such as internet TV programs, which are characterized by their massive volume, relatively traditional sources, and strong traditional media attributes, multi-objective optimization ranking methods are used to recommend video content. Specifically, multi-objective optimization ranking methods generally use deep neural networks to build relevant models to optimize multiple specified objectives / indicators in the content to be ranked, thereby achieving targeted ranking of program content.

[0069] However, using a multi-objective optimization ranking method will limit the exposure of some less popular content, niche content, and breaking news and emergencies, making it impossible to guarantee the fairness of ranking all video content.

[0070] To address the technical problems existing in the prior art, this application proposes an object recommendation method that achieves fair ranking of recommended objects, so as to better handle the recommendation of certain low-popularity, niche, or sudden and urgent content during the ranking process.

[0071] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; the electronic device may be a processing device such as a computer or server, used to implement the object recommendation method provided in this application. Figure 1 As shown, the electronic device includes: a processor 101 and a memory 102.

[0072] The processor 101 and the memory 102 are electrically connected directly or indirectly to enable data transmission or interaction. For example, the electrical connection can be achieved through one or more communication buses or signal lines.

[0073] The processor 101 can be an integrated circuit chip with signal processing capabilities. The processor 101 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0074] The memory 102 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0075] Understandable. Figure 1 The structure described is for illustrative purposes only; the electronic device 100 may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown. Figure 1 The components shown can be implemented using hardware, software, or a combination thereof.

[0076] The memory 102 is used to store programs, and the processor 101 calls the programs stored in the memory 102 to execute the object recommendation method provided in the following embodiments.

[0077] The following will illustrate, through multiple embodiments, a method for recommending objects provided in this application and the corresponding beneficial effects.

[0078] Figure 2 This is a flowchart illustrating an object recommendation method provided in an embodiment of this application. Optionally, the execution entity of this method can be an electronic device such as a server or computer, possessing data processing capabilities. It should be understood that in other embodiments, the order of some steps in the object recommendation method can be interchanged according to actual needs, or some steps can be omitted or deleted. Figure 2 As shown, the method includes:

[0079] S201. Determine the current target vector of each object to be recommended in the set of objects to be recommended.

[0080] The set of objects to be recommended includes multiple objects. In this embodiment, for ease of explanation, "Internet TV program content" is used as an example of an "object to be recommended." However, the "object to be recommended" can also be any other target that can be used for sorting.

[0081] Among them, the objects to be recommended take Internet TV programs as an example. The objects to be recommended include multiple optimization goals, such as the paid ratio, viewing time, number of views, and completion rate.

[0082] The "target vector of the object to be recommended" is used to characterize the state changes of multiple objectives to be optimized for each object to be recommended. In this embodiment, the target vector of the object to be recommended can be the product of "a certain objective to be optimized" and "the weight of the objective to be optimized".

[0083] For example, the "current target vector of the object to be recommended" can be denoted as... , where i is one of the targets to be optimized among the objects to be recommended.

[0084] S202. Determine the eigenvectors of the transformation matrix of the current target vector and the previously sorted target vectors.

[0085] In this embodiment, the eigenvectors of the current target vector and the previously sorted target vector transformation matrix are denoted as... It can be represented as follows:

[0086] (1)

[0087] Where m represents the maximum number of cycles in the calculation, i.e., the target number m, the calculation , represents matrix generation, This represents the current target vector of the object to be recommended. This represents the target vector for the prior sorting of the objects to be recommended, specifically the target vector from the previous sorting.

[0088] S203. Calculate the click-through rate of each object to be recommended based on the target status information of each object to be recommended.

[0089] The target status information of the object to be recommended can refer to information such as viewership ratings, viewing time, click-through rate, and playback duration.

[0090] Optionally, the click-through rate (CTR) of the recommended object can be calculated based on its target state information, and the CTR can be recorded as... .

[0091] in, For a specific object in the set of objects to be recommended, ( (where A is the set of objects to be recommended).

[0092] S204. Calculate the ranking weight of each object to be recommended based on the feature vector, click rate, and the maximum norm of the current target vector.

[0093] Here, the maximum norm of the current target vector refers to the target effect after sorting, and the maximum norm of the current target vector is denoted as... .

[0094] It's understandable that, given the click-through rate (CTR), the order in which the recommended objects are arranged is the order that maximizes the target effect. Therefore, the final arrangement is a strategy weighted average of CTR, where the strategy is the degree of influence of CTR on each target. Hence:

[0095] (2)

[0096] Where p is the number of strategies, i.e. .

[0097] In this embodiment, for the above formula (2), the current value can be obtained by maximum likelihood estimation. , The strategy obtained is used to determine the final ranking order of each object to be recommended. For each object to be recommended, the calculated ranking weight is... .

[0098] Therefore, in this embodiment, the ranking weight of each object to be recommended is mainly calculated based on the feature vector of the change matrix of the current target vector of the object to be recommended, the click rate, the strategy, and the maximum norm of the current target vector.

[0099] S205. Sort each object in the recommendation set according to its ranking weight, and make object recommendations based on the ranking results.

[0100] Based on the above embodiment, the sorting order of each object to be recommended in the set can be obtained according to the sorting weight of each object to be recommended obtained above, and the objects to be recommended in the set can be recommended and displayed based on the obtained sorting result.

[0101] Furthermore, the ranking problem can be viewed as a sequence decision problem, and the selection of user groups can be seen as a verification of the ranking results. Therefore, the goal of ranking is to improve the benefits of multiple objectives that need to be optimized.

[0102] In summary, this application provides an object recommendation method, which includes: determining the current target vector of each object in a set of objects to be recommended, wherein the set of objects to be recommended includes multiple objects to be recommended; determining the feature vectors of the change matrices of the current target vector and the previously sorted target vectors; calculating the click-through rate of each object to be recommended based on the target state information of each object to be recommended; calculating the ranking weight of each object to be recommended based on the feature vector, the click-through rate, and the maximum norm of the current target vector; sorting each object to be recommended in the set of objects to be recommended based on the ranking weight, and recommending objects based on the ranking result. In this scheme, to maximize the target effect of the ranking results of each object to be recommended, a ranking weight is calculated based on the maximum norm of the current target vector of each object in the set of objects to be recommended, the eigenvectors of the transformation matrix of the current target vector and the previously ranked target vectors, and the click-through rate of each object. Then, the objects to be recommended in the set of objects to be recommended are ranked according to their ranking weights, and object recommendations are made based on the ranking results. This ensures that the ranking results can better handle the recommendation of some low-popularity, niche, or sudden and urgent content, effectively guaranteeing the fairness of the ranking of all objects to be recommended.

[0103] like Figure 3 As shown, after recommending objects based on the ranking results, the method also includes:

[0104] S301. Based on the usage records of each object to be recommended in the set of objects to be recommended, determine whether the sorting result is reasonable.

[0105] The usage records of each object to be recommended refer to the user's actions such as clicking, browsing, or watching on each object to be recommended.

[0106] Therefore, the user group's usage records of each recommended object can be seen as a test of the ranking results. If the target user group's usage records of each recommended object maximize the effect, then the ranking result can be determined to be reasonable, and the current ranking scheme can continue to be used.

[0107] S302. If not, then object recommendation is performed based on the preset general recommendation strategy.

[0108] In another possible approach, if the target user group's usage records for each recommended object do not achieve the maximum effect, then the ranking result is deemed unreasonable and the "current ranking method" needs to be adjusted.

[0109] Optionally, in this embodiment, the "currently used ranking scheme" can be adjusted using the following reinforcement learning method.

[0110] In this embodiment, before sorting each object to be recommended, it is necessary to check whether there is a clear sorting scheme in the current environment (i.e., the target user group).

[0111] The ranking scheme used by the target user group may include: a preset general recommendation strategy, and the object recommendation method mentioned in the above embodiments.

[0112] The aforementioned "preset general recommendation strategy" is as follows:

[0113] 1. Use a preset or selected sorting target as the sorting rule;

[0114] 2. The order in which the sorting targets are set is used as the sorting priority standard;

[0115] 3. When no sorting rules are manually set, the default sorting rules will be used:

[0116] a) Prioritize the program's release year, using reverse sorting;

[0117] b) Next, use program sets to create a reverse sort by month;

[0118] c) Then sort by the number of plays from the previous day in reverse order;

[0119] d) Program update time: Cataloging data - last update time, sorted in descending order;

[0120] e) When the content to be sorted does not have any of the above four attributes, only the number of clicks is used for sorting.

[0121] The following section will continue to explain how to determine whether the sorting result is reasonable, and how to adjust the sorting scheme when the sorting result is unreasonable.

[0122] First, if the sorting scheme used in the current environment is the object recommendation method mentioned in the above embodiments, then a random tag is generated;

[0123] a) If the random label is less than 0.05%, the sorting will use the preset general recommendation strategy, and the sorting effect will be checked after sorting; if the target effect is improved by more than 1.5%, the current sorting scheme will be modified to the general recommendation strategy.

[0124] b) If the random label is greater than or equal to 0.05%, the object recommendation method mentioned in the above embodiments will still be used.

[0125] The second approach is to generate a random label if the current environment uses a general recommendation strategy for ranking.

[0126] a) If the random label is less than 0.05%, then use the "object recommendation method mentioned in the above embodiments" to perform a one-time sorting, and then examine the change in the target after such sorting;

[0127] If randomized object sorting improves the target effect by 0.1% over the general sorting target effect in 5 consecutive random uses, then the default sorting of this sorting environment shall be changed to the object recommendation method mentioned in the above embodiment.

[0128] b) If the random labeling rate is greater than or equal to 0.05%, the general recommendation strategy is still used.

[0129] In this embodiment, in order to ensure the rationality of the ranking result based on the ranking weight of each object to be recommended, it is proposed to check whether the ranking result is reasonable based on the usage record of each object to be recommended. If it is not reasonable, then object recommendation is performed based on a preset general recommendation strategy to adjust the "currently used ranking scheme" so as to generate a better ranking result.

[0130] In this application, to verify the ranking results obtained based on the ranking weights of each object to be recommended, this application also proposes using a "reward function for the objects to be recommended" to determine the effectiveness of the final ranking results. Specifically, as follows... Figure 4 As shown, after S205 above, the method further includes:

[0131] S401. Construct a reward function for the object to be recommended based on the value distribution probability and value change rate corresponding to the current target vector.

[0132] To optimize the total value that the "object to be recommended" can obtain across all objectives, a reward function for the object to be recommended can be defined based on the value distribution probability and the rate of change of value corresponding to the current objective vector, as follows:

[0133] (3)

[0134] Where R is the rate of change of the value of the current target vector, and Q is the probability distribution of the value corresponding to the current target vector.

[0135] S402. Determine the target ranking weight of the object to be recommended based on the changing trend of the reward function.

[0136] The reward function's changing trend can be either positive or negative. If the reward function fails to achieve a positive value increase, the strategy selection needs to be iterated again until the reward function achieves a positive value increase. The final ranking weight corresponding to this result is the target ranking weight of the object to be recommended.

[0137] The following examples will explain in detail how to determine the target ranking weight of the object to be recommended based on the changing trend of the reward function.

[0138] Optionally, such as Figure 5 As shown, based on the changing trend of the reward function, the target ranking weight of the object to be recommended is determined, including:

[0139] S402a. Obtain the reward value corresponding to the ranking weight of the object to be recommended.

[0140] Optionally, the sorting weights obtained above are... Substituting into the above formula (3), the reward value corresponding to the ranking weight is calculated.

[0141] S402b. If the reward value is greater than or equal to the preset threshold, the ranking weight is used as the target ranking weight of the object to be recommended. Otherwise, the current target vector of the object to be recommended is recalculated. That is, the steps of the aforementioned embodiment are executed repeatedly to obtain a new reward value until the reward value meets the requirements of this step. Then the loop stops and the ranking weight at this time is used as the target ranking weight.

[0142] Specifically, in one feasible approach, for example, if the reward value calculated above is greater than or equal to 0, then the above ranking weight is used as the target ranking weight of the object to be recommended.

[0143] In another possible implementation, for example, if the reward value calculated above is less than 0, then the next target vector for the object to be recommended is recalculated. Accordingly, the latest ranking order is the click-through rate-weighted strategy, as follows:

[0144] (4)

[0145] The target vector of the object to be recommended in formula (4) , as well as Given the information, it can be recalculated. That is, the new ranking weight of the object to be recommended is recalculated. If the reward value corresponding to the new ranking weight is greater than or equal to 0, then the "new ranking weight" of the object to be recommended obtained in this calculation can be used as the target ranking weight of the object to be recommended.

[0146] Step S205 above: Sort each object to be recommended in the recommendation set according to its ranking weight, and make object recommendations based on the ranking results, including:

[0147] Based on the target ranking weight of each object to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results.

[0148] In this embodiment, the current target vector of the object to be recommended can be calculated in the following manner. Specifically:

[0149] Optionally, determining the current target vector for each object in the set of objects to be recommended includes:

[0150] Based on the pre-built target vector model corresponding to the object to be recommended, the current target vector of the object to be recommended is determined.

[0151] In this embodiment, the pre-constructed target vector model corresponding to the object to be recommended can be as follows:

[0152] (5)

[0153] in, The weights of the target to be optimized The target to be optimized is... For a certain target that needs optimization, A set of optimization objectives.

[0154] The weights of the target to be optimized can be determined using the following calculation formulas (6)-(7). The details are as follows:

[0155] (6)

[0156] (7)

[0157] in Indicate target numerical values ​​and targets The correlation coefficient between the values, Indicate target The standard deviation.

[0158] If you need to adjust the weight of a certain target Then, based on the determined weight values, the change ratio is adjusted, that is:

[0159] (8)

[0160] in, For the new weights, To adjust the proportions, and .

[0161] Therefore, under multi-objective conditions, what is sought is a larger norm for the current objective vector. (i.e., the target effect). For any distinguishable user group, the required optimization target can be established. When combined with recommendation algorithms, the user group is optimized and ranked for each different user.

[0162] The following examples will illustrate how to determine the eigenvectors of the transformation matrix of the current target vector and the previously sorted target vectors. Specifically, as follows... Figure 6 As shown, the eigenvectors of the transformation matrices of the current target vector and the previously sorted target vectors are determined, including:

[0163] S202a. Obtain the sorted feature vectors.

[0164] S202b: Calculate the eigenvectors of the transformation matrix based on the current target vector and the prior eigenvectors.

[0165] In this embodiment, the "pre-sorted first feature vector" can be denoted as... The current target vector is denoted as Then, based on the current target vector and the prior eigenvectors, the eigenvectors of the transformation matrix are calculated. Specifically, as shown in formula (1) above, that is:

[0166] (1)

[0167] The following examples will explain in detail how to calculate the click-through rate of each object to be recommended based on the target status information of each object to be recommended.

[0168] Optionally, the click-through rate of the object to be recommended can be calculated based on the pre-selected target click-through rate model and the historical target status information of the object to be recommended.

[0169] In this embodiment, for the object to be recommended The corresponding "pre-selected target click-through rate model" is "Historical target status information of the object to be recommended" is denoted as .

[0170] Therefore, the click-through rate of the object to be recommended can be calculated using the following formula (9). .

[0171] (9)

[0172] The following describes the object recommendation apparatus and storage medium used to execute the application provided in this application. The specific implementation process and technical effects are described above and will not be repeated here.

[0173] like Figure 7 As shown in the illustration, this application also provides an object recommendation device, which includes:

[0174] The determination module 701 is used to determine the current target vector and the pre-sorted target vector of each object in the set of objects to be recommended. The set of objects to be recommended includes multiple objects to be recommended. The module also determines the feature vectors of the transformation matrices of the current target vector and the pre-sorted target vectors.

[0175] The calculation module 702 is used to calculate the click-through rate of each object to be recommended based on the target state information of each object to be recommended; and to calculate the ranking weight of each object to be recommended based on the feature vector, the click-through rate, and the maximum norm of the current target vector.

[0176] The recommendation module 703 is used to sort each object in the recommendation set according to the ranking weight of each object to be recommended, and to make object recommendations based on the ranking results.

[0177] Optionally, the device further includes:

[0178] The verification module is used to determine whether the ranking result is reasonable based on the usage records of each object in the set of objects to be recommended; if not, it recommends objects based on a preset general recommendation strategy.

[0179] Optionally, the device further includes:

[0180] The construction module is used to construct the reward function for the object to be recommended based on the value distribution probability and value change rate corresponding to the current target vector;

[0181] The determination module 701 is also used to determine the target ranking weight of the object to be recommended based on the changing trend of the reward function;

[0182] Recommended module 703 is also used for:

[0183] Based on the target ranking weight of each object to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results.

[0184] Optionally, the determining module 701 is also used for:

[0185] The reward value corresponding to the ranking weight is obtained based on the ranking weight of the object to be recommended;

[0186] If the reward value is greater than or equal to the preset threshold, the ranking weight is used as the target ranking weight of the object to be recommended; otherwise, the current target vector of the object to be recommended is recalculated.

[0187] Optionally, the determining module 701 is also used for:

[0188] Based on the pre-built target vector model corresponding to the object to be recommended, the current target vector of the object to be recommended is determined.

[0189] Optionally, the determining module 701 is also used for:

[0190] Obtain the sorted feature vectors;

[0191] Based on the current target vector and the prior eigenvectors, the eigenvectors of the transformation matrix are calculated.

[0192] Optionally, the computing module 702 is also used for:

[0193] The click-through rate of the target object is calculated based on the pre-selected target click-through rate model and the historical target status information of the object to be recommended.

[0194] The above-described device is used to execute the method provided in the foregoing embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

[0195] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SOC).

[0196] Optionally, the present invention also provides a program product, such as a computer-readable storage medium, including a program that, when executed by a processor, is used to perform the above-described method embodiments.

[0197] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0198] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0199] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0200] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. An object recommendation method, characterized in that, include: Determine the current target vector of each object in the set of objects to be recommended. The set of objects to be recommended includes multiple objects to be recommended. The current target vector of each object to be recommended is used to characterize the current state change of multiple objectives to be optimized for each object to be recommended. Determine the eigenvectors of the transformation matrix of the current target vector and the previously sorted target vectors; The click-through rate of each object to be recommended is calculated based on the target status information of each object to be recommended. Calculate the ranking weight of each of the objects to be recommended based on the feature vector, the click rate, and the maximum norm of the current target vector; Based on the ranking weight of each of the objects to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results. Specifically, after calculating the ranking weight of each object to be recommended based on the feature vector, the click-through rate, and the maximum norm of the current target vector, a reward function for the object to be recommended is constructed based on the value distribution probability and value change rate corresponding to the current target vector; wherein, the reward function for the object to be recommended is defined as follows: ； in, The rate of change of the current target vector's value. This represents the probability distribution of the value corresponding to the current target vector. Based on the changing trend of the reward function, determine the target ranking weight of the object to be recommended; The step of sorting the objects in the set of objects to be recommended according to their ranking weights, and then recommending objects based on the ranking results, includes: Based on the target ranking weight of each of the objects to be recommended, the objects in the set to be recommended are ranked, and object recommendations are made based on the ranking results. The step of determining the target ranking weight of the object to be recommended based on the changing trend of the reward function includes: The reward value corresponding to the ranking weight is obtained based on the ranking weight of the object to be recommended; If the reward value is greater than or equal to a preset threshold, the ranking weight is used as the target ranking weight of the object to be recommended; otherwise, the current target vector of the object to be recommended is recalculated.

2. The method according to claim 1, characterized in that, The method further includes: Based on the usage records of each object to be recommended in the set of objects to be recommended, determine whether the sorting result is reasonable; If not, then object recommendations will be made based on a preset general recommendation strategy.

3. The method according to claim 1, characterized in that, Determining the current target vector of each object in the set of objects to be recommended includes: Based on the pre-constructed target vector model corresponding to the object to be recommended, the current target vector of the object to be recommended is determined.

4. The method according to claim 1, characterized in that, The step of determining the eigenvectors of the transformation matrix of the current target vector and the previously sorted target vectors includes: Obtain the sorted feature vectors; The eigenvectors of the transformation matrix are calculated based on the current target vector and the prior eigenvectors.

5. The method according to claim 1, characterized in that, The step of calculating the click-through rate of each object to be recommended based on the target status information of each object includes: The click-through rate of the object to be recommended is calculated based on the pre-selected target click-through rate model and the historical target status information of the object to be recommended.

6. An object recommendation device, characterized in that, The device includes: The determination module is used to determine the current target vector and the pre-sorted target vectors of each object in the set of objects to be recommended, wherein the set of objects to be recommended includes multiple objects to be recommended; and to determine the feature vectors of the change matrices of the current target vectors and the pre-sorted target vectors; wherein the current target vector of the object to be recommended is used to characterize the current state changes of multiple targets to be optimized for each object to be recommended. The calculation module is used to calculate the click-through rate of each object to be recommended based on the target state information of each object; and to calculate the ranking weight of each object to be recommended based on the feature vector, the click-through rate, and the maximum norm of the current target vector. The recommendation module is used to sort the objects in the set of objects to be recommended according to the ranking weight of each object to be recommended, and to recommend objects based on the ranking result. The construction module is used to construct the reward function of the object to be recommended based on the value distribution probability and value change rate corresponding to the current target vector; The reward function for the object to be recommended is defined as follows: ； in, The rate of change of the current target vector's value. This represents the probability distribution of the value corresponding to the current target vector. The determining module is further configured to determine the target ranking weight of the object to be recommended based on the changing trend of the reward function; The recommendation module is further configured to sort each object to be recommended in the set of objects to be recommended according to the target ranking weight of each object to be recommended, and to recommend objects based on the ranking result; The determining module is further configured to obtain a reward value corresponding to the ranking weight of the object to be recommended; if the reward value is greater than or equal to a preset threshold, the ranking weight is used as the target ranking weight of the object to be recommended; otherwise, the current target vector of the object to be recommended is recalculated.

7. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, performs the steps of the method as described in any one of claims 1-5.

Images