Method for determining recommended resources, related device and computer program product

Abstract

The present disclosure provides a method, device, electronic device, computer-readable storage medium and computer program product for determining recommended resources, and relates to the fields of artificial intelligence technology such as intelligent recommendation, user understanding, and deep learning. A specific implementation of the method includes: for a target user, recalling a set of matching candidate resources, the candidate resource set includes: a first candidate resource and other second candidate resources except the first candidate resource; based on the first feature of the first candidate resource and the second feature of the second candidate resource, determining a set feature corresponding to the first candidate resource; based on the first feature, the set feature and the user feature of the target user, determining a rough ranking reference value of the first candidate resource; based on the rough ranking reference value, determining the resource to be recommended to the target user. Thus, by introducing a set feature that refers to the features of other candidate resources, more information can be provided for the current candidate resources, which can increase the accuracy of the rough ranking results, thereby improving the accuracy of determining the recommended resources based on the rough ranking results.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, specifically to the field of artificial intelligence technology such as intelligent recommendation, user understanding, and deep learning, and especially to methods, devices, electronic devices, computer-readable storage media, and computer program products for determining recommended resources. Background Art

[0002] With the development of society and computer technology, resource recommendation technology has emerged to facilitate users to obtain resources. In resource recommendation technology, it is possible to provide users with technology to filter and recommend resources based on user needs and preferences.

[0003] Resource recommendation technology is a system that can help users filter and recommend the most relevant content from a large number of resources. Therefore, it can be widely used in scenarios such as e-commerce platforms, online videos, news push, social media, etc. In resource recommendation technology, how to effectively process and recommend resources, especially to make fast and accurate preliminary screening in massive data, has become one of the key technologies.

[0004] Coarse ranking is an important stage in the resource recommendation process. It is responsible for selecting a relatively small candidate set from a large number of candidate resources to provide input for the subsequent fine ranking stage. The goal of coarse ranking is to preliminarily filter out resources that are irrelevant or do not meet the user's interests in a fast and low-computational way to improve the efficiency of the recommendation system. Therefore, how to balance the efficiency and quality of coarse ranking in the coarse ranking stage is worthy of attention and demand. Summary of the invention

[0005] The embodiments of the present disclosure provide a method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product for determining recommended resources.

[0006] In a first aspect, an embodiment of the present disclosure proposes a method for determining recommended resources, comprising: for a target user, recalling a set of matching candidate resources, the set of candidate resources comprising: a first candidate resource and other second candidate resources except the first candidate resource; determining a set feature corresponding to the first candidate resource based on a first feature of the first candidate resource and a second feature of the second candidate resource; determining a coarse ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user; and determining a resource to be recommended to the target user based on the coarse ranking reference value.

[0007] In a second aspect, an embodiment of the present disclosure proposes a device for determining recommended resources, comprising: a candidate resource recall unit, configured to recall a set of matching candidate resources for a target user, the candidate resource set comprising: a first candidate resource and other second candidate resources except the first candidate resource; a set feature determination unit, configured to determine a set feature corresponding to the first candidate resource based on a first feature of the first candidate resource and a second feature of the second candidate resource in the candidate resource set; a coarse ranking reference value determination unit, configured to determine a coarse ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user; and a recommended resource determination unit, configured to determine the resources to be recommended to the target user based on the coarse ranking reference values ​​corresponding to the candidate resources in the candidate resource set.

[0008] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that when the at least one processor executes, it is able to implement the method for determining recommended resources as described in any implementation method in the first aspect.

[0009] In a fourth aspect, an embodiment of the present disclosure provides a non-transitory computer-readable storage medium storing computer instructions, which are used to enable a computer to implement the method for determining recommended resources as described in any implementation manner in the first aspect when executed.

[0010] In a fifth aspect, an embodiment of the present disclosure provides a computer program product comprising a computer program, which, when executed by a processor, can implement the method for determining recommended resources as described in any implementation manner in the first aspect.

[0011] The method, device, electronic device, computer-readable storage medium and computer program product for determining recommended resources provided by the embodiments of the present disclosure recall a set of matching candidate resources for a target user, the set of candidate resources including: a first candidate resource and other second candidate resources except the first candidate resource; determine a set feature corresponding to the first candidate resource based on a first feature of the first candidate resource and a second feature of the second candidate resource; determine a rough ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user; and determine a resource to be recommended to the target user based on the rough ranking reference value.

[0012] The present disclosure can provide more information for the current candidate resources by introducing a collection of features that reference features of other candidate resources, which can increase the accuracy of the rough ranking results, thereby improving the accuracy of determining recommended resources based on the rough ranking results.

[0013] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Other features, objects and advantages of the present disclosure will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0015] Figure 1 is an exemplary system architecture in which the present disclosure may be applied;

[0016] Figure 2 A flowchart of a process for determining recommended resources provided by an embodiment of the present disclosure;

[0017] Figure 3 A flowchart of another process for determining recommended resources provided by an embodiment of the present disclosure;

[0018] Figure 4 A schematic diagram of an architecture for implementing a process of determining recommended resources in an application scenario provided by an embodiment of the present disclosure;

[0019] Figure 5 A structural block diagram of a device for determining recommended resources provided in an embodiment of the present disclosure;

[0020] Figure 6 A schematic diagram of the structure of an electronic device suitable for executing a method for determining recommended resources provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0021] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description. It should be noted that the embodiments in the present disclosure and the features in the embodiments may be combined with each other without conflict.

[0022] In addition, in the technical solutions involved in this disclosure, the acquisition, storage, use, processing, transportation, provision and disclosure of user personal information involved are in compliance with the relevant laws and regulations and do not violate public order and good morals.

[0023] Furthermore, the acquisition and application of contents such as user characteristics in the technical solutions involved in the present disclosure are already known to the users and have been permitted by them.

[0024] Figure 1 An exemplary system architecture 100 is shown to which embodiments of the method, apparatus, electronic device, and computer-readable storage medium for determining recommended resources of the present disclosure can be applied.

[0025] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0026] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various applications for realizing information communication between the terminal devices 101, 102, 103 and server 105 can be installed, such as resource recommendation applications, online shopping applications, instant messaging applications, etc.

[0027] Terminal devices 101, 102, 103 and server 105 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc.; when terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above, which can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server; when the server is software, it can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here.

[0028] The server 105 can provide various services through various built-in applications. Taking a resource recommendation application that can provide a resource recommendation function as an example, the server 105 can achieve the following effects when running the resource recommendation application: first, the server 105 recalls a matching candidate resource set for the target user, and the candidate resource set includes: a first candidate resource and other second candidate resources except the first candidate resource (for example, the target user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to request resource recommendation. Accordingly, the server 105 can respond to the request to determine the target user and recall the matching candidate resource set accordingly); then, the server 105 determines the set feature corresponding to the first candidate resource based on the first feature of the first candidate resource and the second feature of the second candidate resource; next, the server 105 determines the rough ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user; finally, the server 105 determines the resource to be recommended to the target user based on the rough ranking reference value corresponding to the candidate resource in the candidate resource set.

[0029] It should be noted that if the user characteristics of the target user have been pre-stored in the server 105 in various ways, when the server 105 detects that the data have been stored locally (for example, when starting to process the previously reserved resource rough sorting task), the execution subject may choose to directly obtain the data locally. In this case, the exemplary system architecture 100 may also not include the terminal devices 101, 102, 103 and the network 104. For example, in some scenarios, the process of determining the recommended resources may be an "offline task" independently completed by the server 105 locally.

[0030] Since storage resources, characteristics of storage resources, and determination of rough ranking reference values ​​may require more computing resources and stronger computing power, the methods for determining recommended resources provided in the subsequent embodiments of the present disclosure are generally performed by the server 105 with stronger computing power and more computing resources, and accordingly, the device for determining recommended resources is generally also set in the server 105. However, it should also be pointed out that when the terminal devices 101, 102, and 103 also have computing power and computing resources that meet the requirements, the terminal devices 101, 102, and 103 can also complete the above-mentioned various operations that are handed over to the server 105 through the resource recommendation application installed thereon, and then output the same result as the server 105. Especially in the case where there are multiple terminal devices with different computing powers at the same time, but when the resource recommendation application determines that the terminal device where it is located has stronger computing power and more remaining computing resources, it can allow the terminal device to perform the above-mentioned operations, thereby appropriately reducing the computing pressure of the server 105, and accordingly, the device for determining recommended resources can also be set in the terminal devices 101, 102, and 103. In this case, the exemplary system architecture 100 may also not include the server 105 and the network 104 .

[0031] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0032] Please refer to Figure 2 , Figure 2 A flowchart of a process for determining recommended resources provided by an embodiment of the present disclosure includes process 200.

[0033] The process 200 specifically includes the following steps:

[0034] Step 201: Recall a set of matching candidate resources for a target user;

[0035] In the embodiment of the present disclosure, this step is intended to be performed by an execution subject (eg, Figure 1 The server 105 shown in the figure recalls a matching candidate resource set for the target user. The candidate resource set includes a first candidate resource and other second candidate resources except the first candidate resource.

[0036] For example, the execution subject may actively send a request (e.g., a resource recommendation request) to request the execution subject to recall a set of candidate resources and provide recommended resources accordingly. Alternatively, the execution subject may also actively provide resource recommendation services to (target) users based on a pre-configuration, and recall a set of candidate resources to determine recommended resources.

[0037] Generally, the execution subject can recall the corresponding and matching candidate resource set according to the needs and preferences of the target user. The candidate resource set includes multiple "candidate resources" that can be used to recommend to the user. The candidate resource set can be a candidate resource set for "rough sorting" based on the initial recall from the candidate resource library.

[0038] In practice, the candidate resource set may be a set storing candidate resources of the same type, in which case the candidate resource set may be correspondingly configured with a "classification label". Subsequently, the execution subject may determine the corresponding candidate resource set based on the classification (label) pointed to by the demand, for example, after determining the needs of the target user.

[0039] For example, the classification label of the candidate resource set may correspond to products with function A, products with function B, etc. Accordingly, if the target user's "demand" points to products with function A, the execution subject may correspondingly recall the candidate resource set with the classification label of products with function A, and select those "candidate resources" that can be recommended to the user as "recommended resources" from the candidate resource set.

[0040] In practice, candidate resources may include, for example, “physical objects” such as commodities, audio, text, video, etc. that can be obtained by users (including links for obtaining “physical objects”), services, channels for obtaining online information, etc.

[0041] Step 202: Determine a set feature corresponding to the first candidate resource based on a first feature of a first candidate resource and a second feature of a second candidate resource in the candidate resource set;

[0042] In an embodiment of the present disclosure, based on step 201, the execution entity can choose to obtain a first feature corresponding to a first candidate resource in the candidate resource set, and obtain a second feature corresponding to a second candidate resource other than the first candidate resource and different from the first candidate resource from the candidate resource set.

[0043] Then, the execution subject determines a set feature corresponding to the first candidate resource based on the first feature of the first candidate resource and the second feature of the second candidate resource in the candidate resource set.

[0044] For example, for candidate resource A that is the first candidate resource in the candidate resource set, the execution entity can obtain the first feature of "A" and the second feature of another candidate resource that is different from A in the candidate resource set and is the second candidate resource (for example, candidate resource "B"), and then process the first feature of "A" and the second feature of "B" based on the target attention mechanism to determine the set feature corresponding to "A".

[0045] As for the first feature and the second feature, both of them can be pre-processed by combining different resource types with corresponding neural networks. For example, when the resource is in the form of a picture, the first feature and the second feature can be processed by using a convolutional neural network or the like to process the image.

[0046] For another example, when the resource is a commodity, the first feature and the second feature may be obtained by, for example, processing text description information of the commodity through LSTM or the like.

[0047] It should be understood that the first feature and the second feature can actually be the same in feature form and feature type. The distinction between the first and the second is intended to distinguish the correspondence between the feature and the candidate resource and other candidate resources (i.e., for example, the first candidate resource corresponds to the first feature, and the second candidate resource corresponds to the second feature). For example, for a specific candidate resource, when it is expected that its corresponding set feature and rough reference value are determined through the following steps, the feature of the "specific candidate resource" (or the current candidate resource, the target candidate resource) can be referred to as the first feature. Correspondingly, the corresponding features of other candidate resources different from the "specific candidate resource" can be referred to as the second feature.

[0048] Then, after determining the first feature and the second feature, the execution subject can perform a weighted summation based on the similarity between the first feature and each second feature for the first feature, thereby generating a set feature corresponding to the candidate resource. In some scenarios, such a set feature can also be called a set-aware feature.

[0049] For example, the execution entity may calculate, for a first feature, the cosine similarity, dot product similarity, Euclidean distance, or Jaccard similarity between the first feature and (each) other second feature.

[0050] Then, after calculating the similarity between the first feature and each second feature, the calculation results are weighted averaged to obtain a set-aware representation. As a result, other candidate resources that are similar to the candidate resource corresponding to the first feature can be given a higher weight, so that these other similar candidate resources play a greater role in the final representation.

[0051] Step 203: determining a rough ranking reference value of the first candidate resource based on the first feature, the set feature, and the user feature of the target user;

[0052] In an embodiment of the present disclosure, after the execution entity determines the set feature corresponding to the candidate feature based on the above step 202, the execution entity may determine a rough ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user.

[0053] For example, the execution entity can determine a "combined feature" that can correspond to the first candidate resource and includes reference to the second feature based on the combination of the first feature and the set feature, and then the execution entity determines a rough reference value for the first candidate resource based on the similarity between the combined feature and the user feature.

[0054] The size of the rough ranking reference value of the first candidate resource is positively correlated with the similarity, that is, for a candidate resource with a higher similarity, it can be understood that it has a higher recommendation value for the user (or, the probability that the candidate resource can be clicked and obtained by the user is higher), and accordingly, its corresponding rough ranking reference value will be higher. In some scenarios, the rough ranking reference value can also be called a rough ranking reference score, a rough ranking score, etc.

[0055] It should be understood that for each candidate resource in the candidate resource set, the executing entity can, based on the process and method discussed above, take each candidate resource as the "first candidate resource" respectively, and determine the rough reference value corresponding to each candidate resource based on the first feature of the first candidate resource and the second feature of the second candidate resource different from the first candidate resource.

[0056] In other words, even if a candidate resource is used as a second candidate resource by other candidate resources, the candidate resource can still be used as a "first candidate resource" to determine a corresponding rough reference value by the execution entity.

[0057] Step 204: Determine the resources to be recommended to the target user based on the rough ranking reference values ​​corresponding to the candidate resources in the candidate resource set.

[0058] It should be understood that in scenarios such as rough sorting, the determined recommended resources can be directly recommended to users as recommended resources based on different needs, or can be used as materials for the next task stage (for example, the fine sorting stage).

[0059] In an embodiment of the present disclosure, the execution entity may determine the rough ranking reference value corresponding to each first candidate resource (or each candidate resource) in the candidate resource set based on the above method. Then, the execution entity determines the to-be-recommended resources that can be recommended to the target user based on the rough ranking reference value. For example, the execution entity may select those candidate resources that can be used as the to-be-recommended resources (recommended to the target user) based on whether the rough ranking reference value meets the requirements. For example, the execution entity may determine the first candidate resource whose rough ranking reference value is greater than or equal to the reference value threshold as the to-be-recommended resource.

[0060] The method for determining recommended resources provided by the disclosed embodiment is to recall a set of matching candidate resources for a target user, the set of candidate resources including: a first candidate resource and other second candidate resources except the first candidate resource; determine a set feature corresponding to the first candidate resource based on a first feature of the first candidate resource and a second feature of the second candidate resource; determine a rough ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user; and determine the resource to be recommended to the target user based on the rough ranking reference value. Thus, by introducing a set feature that refers to the features of other candidate resources, more information can be provided for the current candidate resources, which can increase the accuracy of the rough ranking results, thereby improving the accuracy of determining the recommended resources based on the rough ranking results.

[0061] In some embodiments, the execution subject may, as an alternative or alternative, select a target-attention mechanism in the process of executing the above step 202, and determine the set feature corresponding to the first candidate resource by processing the first feature of the first candidate resource and the second feature corresponding to the second candidate resource in the candidate resource set. For example, each second feature may be regarded as a "resource sequence", and the first feature may be used as a query vector to utilize the target-attention mechanism to fuse and process to obtain the set feature corresponding to the first candidate resource.

[0062] Specifically, the target attention mechanism can guide the model to focus on the target object (Target) when processing input data, thereby improving the model's understanding and reasoning ability of the target.

[0063] Accordingly, the execution subject can use the first feature as the query vector (Q) in the target attention mechanism and the second feature as the key vector (K). Then, based on the following formula (1), the similarity of the first candidate resource corresponding to the first feature to the candidate resource set is determined:

[0064] Similarity = softmax(Q*K T )(1)

[0065] Then, the following formula (2) is finally used as the set feature corresponding to the first candidate resource:

[0066] Set feature = softmax(Q*K T )*V(2)

[0067] Among them, V is the value vector under the target attention mechanism, which corresponds to the output weighted information.

[0068] Therefore, through the target attention mechanism, the target attention mechanism can be used to determine the set features corresponding to the first candidate resource faster and with lower computational cost.

[0069] In some optional implementations of this embodiment, because a large number of candidate resources may be included in the candidate resource set, the process of determining the set characteristics may require a large amount of time consumption and computing resources. Because the embodiments of the present disclosure are applied to coarse sorting scenarios, it is possible to choose to use only some of the candidate resources in the candidate resource set as the second candidate resources for reference (rather than using all the candidate resources in the candidate resource set except the first candidate resource as the second candidate resource) to reduce the time consumption and computing resource consumption requirements.

[0070] Therefore, it is expected that the computing time and computing consumption will not be significantly increased while at least part of the set characteristics that refer to other candidate resources (i.e., the second candidate resources) can be introduced as a reference and provide auxiliary information, so as to balance the relationship between the computing time and computing consumption.

[0071] For easier understanding, you can also refer to Figure 3 , Figure 3 A flowchart of another process of determining recommended resources provided by an embodiment of the present disclosure includes process 300 .

[0072] The process 300 specifically includes the following steps:

[0073] Step 301: Recall a set of matching candidate resources for a target user;

[0074] The process implemented by this step, as well as the subsequent steps 304 and 305, is actually similar to steps 201, 203 and 204 in process 200. Therefore, reference may be made to the above discussion of steps 201, 203 and 204, and no repetition will be given here.

[0075] Step 302: In response to receiving a quantity setting instruction for a second candidate resource, determining a target quantity of second candidate resources from a candidate resource set based on a target quantity indicated by the quantity setting instruction;

[0076] Specifically, if the execution subject receives a quantity setting instruction for the second candidate resource, the execution subject may respond to this and determine the target number of candidate resources from the candidate resource set as the second candidate resource based on the target number indicated by the quantity setting instruction (for example, the target number may be determined based on the standard of being able to achieve the above-mentioned purpose of reference and providing auxiliary information without significantly increasing the computing time and computing consumption). In other words, the execution subject may determine a subset of candidate resources from the candidate resource set. For example, the execution subject may determine the subset of candidate resources by random sampling, multiple sampling and aggregation, etc.

[0077] Accordingly, the number of candidate resources included in the candidate resource subset is the target number, and then the execution subject can use the candidate resources in the candidate resource subset as second candidate resources that can be used.

[0078] In some optional implementations of this embodiment, the quantity setting instruction may also indicate a “ratio value.” Accordingly, the execution subject may also determine the target quantity based on the product of the ratio value and the total amount of candidate resources included in the candidate resource set.

[0079] Step 303: Determine a set feature corresponding to the first candidate resource based on the first feature of the first candidate resource and the second feature of the second candidate resource;

[0080] Specifically, after determining the target number of "second candidate resources", or "candidate resource subsets" based on the above step 302, when selecting "second candidate resources", the execution subject can choose to extract "other candidate resources" or "second candidate resources" that are different from the first candidate resources corresponding to the first feature only from the determined "second candidate resources", or in other words, from the candidate resource subsets. That is, in the process of determining the corresponding set feature for each candidate resource, the execution subject can select those "second candidate resources" from the candidate resource subset (for example, based on all the "second candidate resources" selected from the candidate resource subset that are different from the first candidate resource, determine the set feature corresponding to the first candidate resource).

[0081] Then, similar to the above step 202, based on the second features of the "second candidate resources" in the candidate resource subset, the set features corresponding to the first candidate resource are determined.

[0082] Step 304: determining a rough ranking reference value of the first candidate resource based on the first feature, the set feature, and the user feature of the target user;

[0083] Step 305: Based on the rough ranking reference values, determine the resources to be recommended to the target user.

[0084] In some embodiments, when the executing entity determines a coarse ranking reference value and determines the recommended resources to be recommended to the target user based on the coarse ranking reference value (for example, executing steps 204 and 305), the executing entity, as an alternative or substitute, chooses to sort each candidate resource in the candidate resource set in descending order based on the size of the coarse ranking reference value corresponding to each candidate resource in the candidate resource set (in fact, the candidate resources here are usually all the candidate resources in the candidate resource set, that is, all the candidate resources including the first candidate resource and the second candidate resource).

[0085] For ease of understanding, the description is also made with “first candidate resources”. For example, the execution subject may sort the first candidate resources in the candidate resource set from large to small, from high to low according to the corresponding rough sorting reference values.

[0086] Then, based on the sorting results in descending order, the first candidate resources ranked in front of a preset number (usually, the preset number is determined based on the number of recommended resources that can be recommended) in the sorting results are determined as the resources to be recommended to the target user.

[0087] Therefore, the execution entity can select candidate resources that have higher recommendation value and are more likely to be clicked by users for recommendation based on the rough ranking reference value.

[0088] In some optional implementations of this embodiment, in the process of selecting and determining recommended resources based on the sorting results, the execution entity may also choose to "take into account the reference value threshold" to further constrain the candidate resources to avoid recommending candidate resources that may not meet the recommendation requirements.

[0089] Accordingly, in the process of executing the sorting result based on descending sorting and determining the first preset number of candidate resources as recommended resources recommended to the target user, the execution subject may, as an alternative or substitute, select the sorting result based on descending sorting and determine the first candidate resource whose ranking is within the first preset number of the sorting results and whose corresponding rough ranking reference value is greater than or equal to the reference value threshold as the resource to be recommended to the target user.

[0090] In some embodiments, during the process of the execution entity determining the coarse-ranked reference values ​​of the candidate resources, the execution entity may choose to introduce a cross-feature approach in the hope of more accurately determining the coarse-ranked reference values.

[0091] For example, when the execution subject is executing step 203 and / or step 304, it may, as an alternative or substitution, first select to combine the first feature and the set feature to obtain a combination result.

[0092] Then, the feature cross processing combination result is combined with the user features of the target user to obtain the cross features.

[0093] For example, the execution subject can perform feature crossover to obtain crossover features through a "crossover operation" manually created by an operation and maintenance personnel, or by using tools such as Polynomial Features or Feature Crosses, or by using a neural network.

[0094] Accordingly, the execution entity can generate new features (such as "combination result-user feature" features) by cross-crossing the combination results and user features, so that in the process of determining the coarse-ranking reference value, it can learn the complex interactions between these features and determine the coarse-ranking reference value more accurately.

[0095] Accordingly, in the process of determining the rough ranking reference value of the candidate resources, the execution subject may further select to determine the rough ranking reference value of the first candidate resource based on the first feature, the set feature, the cross feature and the user feature of the target user.

[0096] For example, the execution entity may determine the rough ranking reference values ​​of the candidate resources based on the determination of the rough ranking reference values ​​discussed above, and further use the "cross feature" as a newly added input.

[0097] In some embodiments, the execution subject may also aggregate requirements and choose to add resource multi-objective Q-Learning, etc., in the process of determining the rough ranking reference values ​​of the candidate resources, to further improve the determination process.

[0098] Resource-based multi-objective Q-Learning is a method that combines resource allocation and multi-objective optimization, and can be used for decision-making, optimization, and learning in an environment with multiple objectives. It can be understood as an extension of the Q-Learning algorithm, where "resources" refer to the limited resources in the system (such as time, computing power, storage space, etc.), and "multi-objective" means that in the decision-making process, not only a single objective must be considered, but multiple objectives must be balanced.

[0099] Therefore, by using resource multi-objective Q-learning, we can better balance the usage relationship between the "first feature" and the "collective feature", so that the executor can pay more attention to the two features in a more balanced and sufficient manner when jointly using the first feature and the collective feature to determine the rough reference value, while not completely biased towards one of them, and can make more full use of the first feature and the collective feature.

[0100] In some embodiments, in the process of the execution subject determining the rough ranking reference value of the first candidate resource, the execution subject may choose to use a multi-layer perceptron to determine the rough ranking reference value of the first candidate resource. For example, in the process of executing step 203 and / or step 304, the execution subject may, as an alternative or alternative, choose to use a multi-layer perceptron to process the first feature, the set feature, and the user feature of the target user to determine the rough ranking reference value of the first candidate resource.

[0101] Multilayer Perceptron (MLP) is a feedforward artificial neural network model that can map multiple input data sets to a single output data set. For example, a deep neural network (DNN) based on MLP can be selected to process the first feature, the set feature, and the user feature of the target user to determine the rough ranking reference value of the first candidate resource. Alternatively, the first feature, the set feature, the user feature of the target user, and the above cross-features can be processed to determine the rough ranking reference value of the first candidate resource.

[0102] Therefore, based on the simpler structure of MLP compared to network architectures such as CNN, it is possible to process the first feature, set features and user features of the target user at a lower cost and with lower resource consumption to determine the rough ranking reference value of the candidate resources.

[0103] In some optional implementations of this embodiment, in the process of using a multilayer perceptron to process the first feature, the set feature and the user feature of the target user to determine the coarse-ranking reference value of the candidate resources, the execution entity may also choose to use at least two or more multilayer perceptrons to improve the processing capability so as to determine a more accurate coarse-ranking reference value.

[0104] For example, in the above DNN, more than two multilayer perceptrons may actually be configured.

[0105] For such a situation, if there are at least two multilayer perceptrons, the execution subject may respond to this and determine, for each multilayer perceptron, an initial rough ranking reference value for the first candidate resource corresponding to the multilayer perceptron based on the first feature, the set feature, and the user feature of the target user. That is, the execution subject may determine the initial rough ranking reference value for the first candidate resource corresponding to each multilayer perceptron based on the first feature, the set feature, and the user feature.

[0106] The execution subject may choose to use each MLP to independently and correspondingly implement the above steps to determine the initial coarse ranking reference value for the first candidate resource corresponding to each multi-layer perceptron.

[0107] Then, the execution subject determines the coarse-ranked reference value of the first candidate resource based on the power fusion results of the initial coarse-ranked reference values ​​for the first candidate resource corresponding to each multi-layer perceptron.

[0108] For example, for different MLPs, after obtaining their corresponding initial rough sorting reference values, a "power operation" can be performed on them based on the weight coefficient (or power weight) corresponding to the MLP.

[0109] Then, the execution entity adds up the results of the "power operation" corresponding to each MLP to obtain the above-mentioned power fusion result, and uses the power fusion result as a rough ranking reference value for the candidate resources.

[0110] Therefore, through the power fusion method, the initial rough reference values ​​corresponding to each MLP can be "fused" through power nonlinear transformation. The result of such fusion is no longer a simple linear weighted average, which can improve the performance of at least two MLPs when used together.

[0111] Based on any of the above embodiments, for the first feature and the second feature, one or both of them can be determined based on a code table of a variational autoencoder with reparameterized quantization.

[0112] Reparameterized Quantization Variational Autoencoder (RQ-VAE) is a model proposed for quantization and reconstruction tasks under the variational autoencoder framework. It can usually be applied to image processing, data compression, generative models and other fields.

[0113] The codebook of RQ-VAE refers to a discretized set of codes. RQ-VAE can convert the continuous representation of the latent space into a discrete representation through quantization, for example, by mapping to predefined discrete codes (codebook entries) in a codebook.

[0114] Accordingly, by using such discrete representations as the first and second features, quantization and reparameterization techniques can be combined, so that, for example, a DNN model can handle the discretization of the latent space to convert the latent representation from a continuous space to a finite set of discrete vectors. This approach can not only avoid situations such as continuous latent variables, but also enable efficient compression and reconstruction of data through discrete codebooks, so that the final features can express information more completely and definitively.

[0115] To deepen understanding, the present disclosure also provides a specific implementation solution in combination with a specific application scenario, see Figure 4 The architecture 400 is shown as an example.

[0116] In the architecture 400, the candidate resource set 410 may be recalled for a target user (not shown in the figure). The candidate resource set may include candidate resources 411, 412, 413, 414, ... 41N (where N is a positive integer).

[0117] To facilitate a clearer understanding of the present disclosure, an example will be given using the above-discussed case where there is a “candidate resource subset”.

[0118] In this case, the exemplary execution subject (e.g., the server 105 described above) may respond upon receiving a quantity setting instruction for the second candidate resource, and determine the target quantity of second candidate resources from the candidate resource set based on the target quantity indicated by the quantity setting instruction. For ease of understanding, the description is made herein in the form of a "candidate resource subset".

[0119] For example, the target number is exemplarily 3, and the execution subject can determine 3 candidate resources as the second candidate resources from the candidate UAN combination 410. For example, based on the candidate resource set 410, the candidate resources can be correspondingly extracted to determine the candidate resource subset 410' (for example, the candidate resource subset 410' can include 3 candidate resources as the second candidate resources, for example, the candidate resources 412, 413 and 414 are included as the second candidate resources).

[0120] Next, exemplarily in the case where it is desired to determine a coarse-ranked reference value corresponding to candidate resource 411 (for example, coarse-ranked reference value 451 of candidate resource 411), the execution entity may be based on the target attention mechanism and fuse feature 421 of candidate resource 411, as well as feature 422 of candidate resource 412, feature 423 of candidate resource 413, and feature 424 of candidate resource 414 as the "second candidate resource" (for example, using feature 422 of candidate resource 412, feature 423 of candidate resource 413, and feature 424 of candidate resource 414 as the compared resource sequence corresponding to the above-mentioned key vector K).

[0121] According to the above-mentioned fusion processing results based on the target attention mechanism, the set features corresponding to the candidate resource 411 (ie, the set features 431) can be further obtained.

[0122] Next, the execution entity may input at least the features 421 , the collection features 431 , and the user features 440 of the candidate resource 411 into the deep neural network 450 for processing (for example, the deep neural network 450 may include multiple, multi-layer MLPs) to obtain a rough ranking reference value 451 for the candidate resource 411 .

[0123] Similarly, when the execution subject uses the deep neural network 450 for processing, it can also choose to add "resource multi-objective Q-learning", "cross features", etc. based on different needs, which will not be repeated here.

[0124] Subsequently, the execution subject may determine whether to use the candidate resource 411 as a resource to be recommended, for example, based solely on the comparison result with the reference value threshold, based solely on the descending sort result, or based on the descending sort result and the comparison result with the reference value threshold, as discussed above. (For example, in the case of only the comparison result of the reference value threshold, if the rough ranking reference value 451 of the candidate resource 411 is greater than or equal to the reference value threshold, it can be determined as the above-mentioned resource to be recommended).

[0125] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a device for determining recommended resources. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0126] like Figure 5 As shown, the device 500 for determining recommended resources in this embodiment may include: a candidate resource recall unit 501, a set feature determination unit 502, a rough ranking reference value determination unit 503 and a recommended resource determination unit 504. Among them, the candidate resource recall unit 501 is configured to recall a matching candidate resource set for a target user, and the candidate resource set includes: a first candidate resource and other second candidate resources except the first candidate resource; the set feature determination unit 502 is configured to determine a set feature corresponding to the first candidate resource based on a first feature of the first candidate resource and a second feature of the second candidate resource in the candidate resource set; the rough ranking reference value determination unit 503 is configured to determine a rough ranking reference value of the first candidate resource based on the first feature, the set feature and the user feature of the target user; the recommended resource determination unit 504 is configured to determine the resource to be recommended to the target user based on the rough ranking reference value corresponding to the candidate resource in the candidate resource set.

[0127] In the device 500 for determining recommended resources in this embodiment, the specific processing of the candidate resource recall unit 501, the set feature determination unit 502, the rough sorting reference value determination unit 503 and the recommended resource determination unit 504 and the technical effects thereof can be referred to in detail. Figure 2 The relevant descriptions of steps 201 - 204 in the corresponding embodiment are not repeated here.

[0128] In some optional implementations of this embodiment, the set feature determination unit 502 is further configured to determine the set feature based on the target attention mechanism and by fusing the first feature and the second feature.

[0129] In some optional implementations of this embodiment, the device 500 also includes: a second candidate resource range determination unit, configured to determine a target number of second candidate resources from the candidate resource set based on the target number indicated by the quantity setting instruction in response to receiving a quantity setting instruction for the second candidate resources.

[0130] In some optional implementations of the present embodiment, the recommended resource determination unit 504 includes: a candidate resource sorting subunit, configured to sort the first candidate resources in descending order based on a coarse ranking reference value corresponding to the first candidate resources to obtain a sorting result; and a recommended resource determination subunit, configured to determine, based on the sorting result, the first candidate resources that are ranked in front of a preset number in the sorting result as resources to be recommended.

[0131] In some optional implementations of this embodiment, the recommended resource determination subunit is further configured to, based on the sorting result, determine the first candidate resource whose sorting is within the previous preset number and whose corresponding rough ranking reference value is greater than or equal to the reference value threshold as the resource to be recommended.

[0132] In some optional implementations of this embodiment, the coarse-ranking reference value determining unit 503 is further configured to use a multi-layer perceptron to process the first feature, the set feature, and the user feature of the target user to determine the coarse-ranking reference value of the first candidate resource.

[0133] In some optional implementations of this embodiment, the coarse-rank reference value determination unit 503 is further configured to, in response to determining that there are at least two multi-layer perceptrons, process the first feature, the set feature, and the user feature, and determine the initial coarse-rank reference value of the first candidate resource corresponding to each multi-layer perceptron; and determine the coarse-rank reference value of the first candidate resource based on the power fusion result of each initial coarse-rank reference value.

[0134] In some optional implementations of the present embodiment, the coarse ranking reference value determination unit 503 includes: a feature combination subunit, configured to combine the first feature and the set feature to obtain a combination result; a feature cross subunit, configured to perform feature cross-processing on the combination result and the user feature to obtain a cross feature; and a coarse ranking reference value determination subunit, configured to determine the coarse ranking reference value of the first candidate resource based on the first feature, the set feature, the cross feature and the user feature of the target user.

[0135] In some optional implementations of this embodiment, at least one of the first feature and the second feature is determined based on a code table of a variational autoencoder with reparameterized quantization.

[0136] This embodiment exists as an apparatus embodiment corresponding to the above method embodiment. The apparatus for determining recommended resources provided in this embodiment recalls a set of matching candidate resources for a target user, and the set of candidate resources includes: a first candidate resource and other second candidate resources except the first candidate resource; based on the first feature of the first candidate resource and the second feature of the second candidate resource, determines the set feature corresponding to the first candidate resource; based on the first feature, the set feature and the user feature of the target user, determines the rough ranking reference value of the first candidate resource; based on the rough ranking reference value, determines the resource to be recommended to the target user. Thus, by introducing the set feature that refers to the features of other candidate resources, more information can be provided for the current candidate resources, which can make the rough ranking results more accurate, thereby improving the accuracy of determining the recommended resources based on the rough ranking results.

[0137] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0138] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0139] like Figure 6As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0140] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0141] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above, such as a method for determining recommended resources. For example, in some embodiments, the method for determining recommended resources may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the method for determining recommended resources described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform a method for determining recommended resources in any other appropriate manner (e.g., by means of firmware).

[0142] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0143] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0144] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0145] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0146] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0147] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and virtual private servers (VPS) services. The server can also be divided into a server of a distributed system, or a server combined with a blockchain.

[0148] According to the technical solution of the disclosed embodiment, for the target user, a set of matching candidate resources is recalled, and the candidate resource set includes: a first candidate resource and other second candidate resources except the first candidate resource; based on the first feature of the first candidate resource and the second feature of the second candidate resource, a set feature corresponding to the first candidate resource is determined; based on the first feature, the set feature and the user feature of the target user, a rough ranking reference value of the first candidate resource is determined; based on the rough ranking reference value, a resource to be recommended to the target user is determined. Thus, by introducing a set feature that refers to the features of other candidate resources, more information can be provided for the current candidate resources, which can make the rough ranking results more accurate, thereby improving the accuracy of determining the recommended resources based on the rough ranking results.

[0149] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions provided by this disclosure can be achieved, and this document does not limit this.

[0150] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for determining a recommended resource, comprising: For a target user, a matching candidate resource set is retrieved, wherein the candidate resource set includes: a first candidate resource and other second candidate resources except the first candidate resource; Determining a set feature corresponding to the first candidate resource based on a first feature of the first candidate resource and a second feature of the second candidate resource; Determine a rough ranking reference value of the first candidate resource based on the first feature, the set feature, and the user feature of the target user; Based on the rough ranking reference value, the resource to be recommended to the target user is determined.

2. The method according to claim 1, wherein: The determining, based on the first feature of the first candidate resource and the second feature of the second candidate resource, a set feature corresponding to the first candidate resource includes: Based on a target attention mechanism, the set feature is determined by fusing the first feature and the second feature.

3. The method according to claim 1, further comprising: In response to receiving a quantity setting instruction for the second candidate resources, the target quantity of second candidate resources is determined from the candidate resource set based on the target quantity indicated by the quantity setting instruction.

4. The method according to claim 1, wherein: The determining the resource to be recommended to the target user based on the rough ranking reference value includes: Based on the coarse ranking reference value corresponding to the first candidate resource, the first candidate resources are sorted in descending order to obtain a sorting result; Based on the ranking result, the first candidate resources ranked at the front by a preset number in the ranking result are determined as the resources to be recommended.

5. The method according to claim 4, wherein: The step of determining, based on the ranking result, a first candidate resource ranked in front of a preset number as the resource to be recommended includes: Based on the ranking result, a first candidate resource whose ranking is within a preset number and whose corresponding rough ranking reference value is greater than or equal to a reference value threshold is determined as the resource to be recommended.

6. The method according to claim 1, wherein: The determining, based on the first feature, the set feature, and the user feature of the target user, a rough ranking reference value of the first candidate resource includes: The first feature, the set feature and the user feature are processed by a multi-layer perceptron to determine a rough ranking reference value of the first candidate resource.

7. The method according to claim 6, wherein: The using a multi-layer perceptron to process the first feature, the set feature, and the user feature to determine a rough ranking reference value of the first candidate resource includes: In response to determining that there are at least two of the multilayer perceptrons, determining an initial coarse ranking reference value of the first candidate resource corresponding to each of the multilayer perceptrons based on the first feature, the set feature, and the user feature; Based on a power fusion result of each of the initial coarse-ranked reference values, a coarse-ranked reference value of the first candidate resource is determined.

8. The method according to claim 1, wherein: The determining, based on the first feature, the set feature, and the user feature of the target user, a rough ranking reference value of the first candidate resource includes: combining the first feature and the set feature to obtain a combination result; Cross-processing the combination result and the user feature to obtain a cross-feature; A rough ranking reference value of the first candidate resource is determined based on the first feature, the set feature, the cross feature and the user feature.

9. The method according to any one of claims 1 to 8, wherein: At least one of the first feature and the second feature is determined based on a code table of a variational autoencoder with reparameterization and quantization.

10. A device for determining recommended resources, comprising: A candidate resource recall unit is configured to recall a matching candidate resource set for a target user, wherein the candidate resource set includes: a first candidate resource and other second candidate resources except the first candidate resource; a set feature determination unit, configured to determine a set feature corresponding to the first candidate resource based on a first feature of a first candidate resource and a second feature of a second candidate resource in the candidate resource set; a coarse ranking reference value determining unit, configured to determine a coarse ranking reference value of the first candidate resource based on the first feature, the set feature, and a user feature of the target user; The recommended resource determination unit is configured to determine the to-be-recommended resource to be recommended to the target user based on the coarse-ranked reference values ​​corresponding to the candidate resources in the candidate resource set.

11. The device according to claim 10, wherein: The set feature determination unit is further configured to determine the set feature based on a target attention mechanism and by fusing the first feature and the second feature.

12. The device according to claim 10, further comprising: The second candidate resource range determining unit is configured to, in response to receiving a quantity setting instruction for the second candidate resources, determine the target quantity of second candidate resources from the candidate resource set based on the target quantity indicated by the quantity setting instruction.

13. The device according to claim 10, wherein: The recommended resource determination unit includes: a candidate resource sorting subunit, configured to sort the first candidate resources in descending order based on the coarse ranking reference values ​​corresponding to the first candidate resources to obtain a sorting result; The recommended resource determination subunit is configured to determine, based on the ranking result, the first candidate resources that are ranked at the front by a preset number in the ranking result as the to-be-recommended resources.

14. The device according to claim 13, wherein: The recommended resource determination subunit is further configured to, based on the ranking result, determine as the resource to be recommended a first candidate resource whose ranking is within a pre-set number and whose corresponding rough ranking reference value is greater than or equal to a reference value threshold.

15. The device according to claim 10, wherein: The coarse-ranking reference value determination unit is further configured to use a multi-layer perceptron to process the first feature, the set feature, and the user feature of the target user to determine the coarse-ranking reference value of the first candidate resource.

16. The device according to claim 15, wherein: The coarse-rank reference value determination unit is further configured to, in response to determining that there are at least two of the multilayer perceptrons, process the first feature, the set feature and the user feature, and determine an initial coarse-rank reference value of the first candidate resource corresponding to each of the multilayer perceptrons; and determine the coarse-rank reference value of the first candidate resource based on a power fusion result of each of the initial coarse-rank reference values.

17. The device according to claim 10, wherein: The coarse arrangement reference value determination unit comprises: a feature combination subunit, configured to combine the first feature and the set feature to obtain a combination result; A feature cross subunit is configured to perform feature cross processing on the combination result and the user feature to obtain a cross feature; The coarse ranking reference value determination subunit is configured to determine a coarse ranking reference value of the first candidate resource based on the first feature, the set feature, the cross feature and the user feature of the target user.

18. The device according to any one of claims 10 to 17, wherein: At least one of the first feature and the second feature is determined based on a code table of a variational autoencoder with reparameterization and quantization.

19. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for determining recommended resources according to any one of claims 1 to 9.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method for determining recommended resources according to any one of claims 1 to 9.

21. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method for determining recommended resources according to any one of claims 1 to 9.

Images