Resource recommendation method and device, electronic equipment and storage medium

Abstract

The invention discloses a resource recommendation method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a first resource in a target resource set in response to a resource recommendation request of a target object; the first resource is a resource meeting a new resource condition in the target resource set; searching a target second resource corresponding to the first resource based on resource mapping information; the resource mapping information represents second resources respectively associated with a plurality of different preset first resources; the second resource is a resource of which the recommendation index exceeds a preset recommendation index threshold value in a preset historical time period; under the condition that the target second resource is found, obtaining resource feature representation of the target second resource; and based on the object feature representation of the target object and the resource feature representation of the target second resource, predicting a recommendation probability of recommending the first resource to the target object. The cold start success rate of new resources can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a resource recommendation method and device, electronic equipment and storage medium. BACKGROUND

[0002] At present, the resource recommendation system often uses the collaborative filtering method to recommend resources to users, for example, recommending the advertisement resources that the user may be interested in to the user.

[0003] However, the collaborative filtering method in the related art needs to rely on a large amount of interaction data between the user and the resource, so for new resources lacking interaction data, it is difficult to accurately match the target user, cannot make accurate recommendations, and needs to wait for the accumulation of interaction data of the new resource, thereby leading to poor recommendation accuracy of the new resource and poor cold start effect of the new resource. SUMMARY

[0004] In order to solve the problems of the prior art, the embodiments of the present application provide a resource recommendation method and device, electronic equipment and storage medium. The technical solution is as follows:

[0005] On the one hand, a resource recommendation method is provided, and the method comprises:

[0006] In response to a resource recommendation request of a target object, a first resource in a target resource set is determined; the first resource is a resource in the target resource set that meets a new resource condition;

[0007] Based on resource mapping information, a target second resource corresponding to the first resource is found; the resource mapping information represents a plurality of different preset first resources respectively associated with second resources; the second resource is a resource whose recommendation indicator in a preset historical time period exceeds a preset recommendation indicator threshold;

[0008] In the case where the target second resource is found, a resource feature representation of the target second resource is obtained;

[0009] Based on the object feature representation of the target object and the resource feature representation of the target second resource, a recommendation probability of recommending the first resource to the target object is predicted.

[0010] On the other hand, a resource recommendation device is provided, and the device comprises:

[0011] A first resource determination module is configured to determine a first resource in a target resource set in response to a resource recommendation request of a target object; the first resource is a resource in the target resource set that meets a new resource condition;

[0012] The associated resource search module is used to search for the target second resource corresponding to the first resource based on resource mapping information; the resource mapping information represents the second resources associated with multiple different preset first resources; the second resource is a resource whose recommendation index exceeds a preset recommendation index threshold within a preset historical time period;

[0013] The resource feature representation acquisition module is used to acquire the resource feature representation of the target second resource when the target second resource is found.

[0014] The first recommendation probability prediction module is used to predict the recommendation probability of recommending the first resource to the target object based on the object feature representation of the target object and the resource feature representation of the target second resource.

[0015] In one exemplary embodiment, the resource mapping information includes multiple preset correspondences between first resource identifiers and second resource identifiers; the associated resource lookup module includes:

[0016] The first resource identifier acquisition module is used to acquire the first resource identifier of the first resource.

[0017] The correspondence lookup module is used to search whether there is a target correspondence in the resource mapping information that matches the first resource identifier; the preset first resource identifier in the target correspondence is the first resource identifier.

[0018] The associated resource determination module is used to obtain the target second resource identifier in the target correspondence when the target correspondence is found, wherein the second resource indicated by the target second resource identifier is the target second resource.

[0019] In one exemplary embodiment, the apparatus further includes a resource mapping information generation module, the resource mapping information generation module comprising:

[0020] The first filtering module is used to select second resources that meet the filtering conditions from the reference resource set to obtain a candidate second resource set; the filtering conditions include the recommendation index exceeding the preset recommendation index threshold within the preset historical time period;

[0021] The second filtering module is used to select a first resource that meets the new resource conditions from the reference resource set to obtain a preset first resource set;

[0022] The similarity determination module is used to determine the similarity between each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set for each preset first resource.

[0023] The associated candidate resource determination module is used to determine the candidate second resource associated with the preset first resource based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set;

[0024] The resource mapping information generation submodule is used to generate the resource mapping information based on the candidate second resources associated with each of the preset first resources.

[0025] In one exemplary implementation, the associated candidate resource determination module includes:

[0026] The first determining module is used to determine a target candidate second resource whose similarity exceeds a preset similarity level based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set;

[0027] The display environment information acquisition module is used to acquire the first display environment information of the preset first resource and the second display environment information of the target candidate second resource;

[0028] The second determining module is used to determine the matching target candidate second resource as the candidate second resource associated with the preset first resource when the second display environment information matches the first display environment information.

[0029] In one exemplary embodiment, the second determining module includes:

[0030] A quantity determination module is used to determine the number of matching target candidate second resources;

[0031] The third determining module is used to determine the matching target candidate second resource with the highest similarity as the candidate second resource associated with the preset first resource when the number is multiple.

[0032] In one exemplary implementation, the similarity determination module includes:

[0033] The visual feature extraction module is used to extract visual features from each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set, to obtain the preset resource visual feature vector of each preset first resource and the candidate resource visual feature vector of each candidate second resource.

[0034] The similarity determination submodule is used to determine the similarity between the preset first resource and each candidate second resource for each preset first resource based on the preset resource visual feature vector of the preset first resource and the visual feature vector of each candidate resource.

[0035] In one exemplary implementation, the similarity determination submodule includes:

[0036] The graph embedding vector acquisition module is used to acquire the graph embedding vector of the resource node corresponding to each resource in the reference resource set in the graph structure; the graph structure is constructed based on the historical interaction data between each resource and the interaction object in the reference resource set.

[0037] The graph embedding vector determination module is used to determine the graph embedding vector of the resource node corresponding to the preset first resource, and the graph embedding vector of the resource node corresponding to each of the candidate second resources.

[0038] The first splicing module is used to splice the graph embedding vector of the preset first resource with the visual feature vector of the preset resource to obtain the preset comprehensive resource feature vector of the preset first resource.

[0039] The second splicing module is used to splice the graph embedding vector of each of the candidate second resources with the corresponding candidate resource visual feature vector to obtain the candidate comprehensive resource feature vector of each of the candidate second resources.

[0040] The vector similarity calculation module is used to determine the similarity between the preset comprehensive resource feature vector of the preset first resource and the candidate comprehensive resource feature vector of each of the candidate second resources, and use this as the similarity between the preset first resource and each candidate second resource.

[0041] In one exemplary implementation, the graph embedding vector acquisition module includes:

[0042] The graph structure construction module is used to construct a graph structure based on the historical interaction data between each resource and the interactive object in the reference resource set. The graph structure includes multiple graph nodes and connecting edges between adjacent graph nodes. The multiple graph nodes include resource nodes corresponding to resources and object nodes corresponding to interactive objects. The connecting edges represent the interaction relationship between the corresponding interactive object and the corresponding resource.

[0043] The graph embedding vector generation module is used to obtain the graph embedding vector of each resource node based on the pre-trained graph neural network model and the graph structure.

[0044] In one exemplary implementation, the new resource conditions include: the resource's online duration is less than a preset online duration, the duration of the most recently added material in the resource is less than a preset addition duration, and the resource does not meet the cold start success conditions.

[0045] In one exemplary embodiment, the apparatus further includes:

[0046] The update module is used to update the resource mapping information based on preset exit conditions;

[0047] The preset exit conditions include: the online duration of the preset first resource exceeds the preset online duration and the feedback index of the preset first resource exceeds the preset feedback index threshold.

[0048] In one exemplary implementation, the resource feature representation acquisition module includes:

[0049] The current status acquisition module is used to acquire the current status of the target second resource when the target second resource is found.

[0050] The second resource feature representation acquisition submodule is used to acquire the resource feature representation of the target second resource when the current state indicates that the target second resource is in a displayable state.

[0051] In one exemplary embodiment, the resource feature representation acquisition module further includes:

[0052] The first resource feature representation acquisition submodule is used to acquire the resource feature representation of the first resource when the current state indicates that the target second resource is in an undisplayable state.

[0053] The second recommendation probability prediction module is used to predict the recommendation probability of recommending the first resource to the target object based on the object feature representation of the target object and the resource feature representation of the first resource.

[0054] On the other hand, an electronic device is provided, including a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the resource recommendation method of any of the above aspects.

[0055] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the resource recommendation method as described above.

[0056] On the other hand, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the resource recommendation method of any of the above aspects.

[0057] This application embodiment, in response to a resource recommendation request from a target object, determines a first resource in a resource set that meets the conditions for a new resource. Then, based on resource mapping information, it searches for a target second resource corresponding to the first resource. If the target second resource is found, it obtains the resource feature representation of the target second resource. Based on the object feature representation of the target object and the resource feature representation of the target second resource, it predicts the recommendation probability of recommending the first resource to the target object. Since the resource mapping information represents multiple second resources associated with different preset first resources, and the associated second resources are resources whose recommendation indicators exceed a preset recommendation indicator threshold within a preset historical time period, it is possible to recommend the first resource belonging to the new resource type to the target second resource's high-conversion interactive object, thereby improving the recommendation accuracy of new resources and the cold start success rate of new resources. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a schematic diagram of an implementation environment provided in an embodiment of this application;

[0060] Figure 2 This is a flowchart illustrating a resource recommendation method provided in an embodiment of this application;

[0061] Figure 3 This is a flowchart illustrating another resource recommendation method provided in an embodiment of this application;

[0062] Figure 4 This is a specific example of a resource recommendation method provided in the embodiments of this application;

[0063] Figure 5 This is an example of the application effect of a resource recommendation method provided in the embodiments of this application;

[0064] Figure 6 This is a hardware structure block diagram of a resource recommendation device provided in an embodiment of this application;

[0065] Figure 7 This is a hardware structure block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0066] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0067] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0068] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0069] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0070] Please see Figure 1 The diagram shown is an implementation environment provided in this application embodiment. The implementation environment includes a terminal 110 and a server 120, wherein the terminal 110 and the server 120 can communicate through a wired or wireless network connection.

[0071] Terminal 110 includes, but is not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, vehicle terminals, and aircraft. Terminal 110 is equipped with client software such as an application (App) that has resource display functions. This application can be a standalone application or a subroutine within an application. In this embodiment, the resources displayed by the resource display function can be promotional resources such as advertising resources. Advertising resources are a type of multimedia resource and can include various types of materials, such as combinations of images, text, audio, and video. For example, an advertisement may be a video advertisement, meaning an advertisement is a video clip; an advertisement may be a graphic advertisement, meaning an advertisement includes both images and text; an advertisement may be a text advertisement (pure advertising copy), meaning an advertisement is a piece of text; or an advertisement may include a combination of various types of multimedia resources, such as an advertisement that is a resource obtained by adding text and a cover image to a video clip.

[0072] Server 120 can provide background services for applications in terminal 110, specifically resource recommendation services. Specifically, upon receiving a resource recommendation request from terminal 110, server 120 can respond to the request by performing resource recommendation processing to determine the resources to recommend to terminal 110. During resource recommendation processing, server 120 typically goes through one or more recommendation probability prediction stages, selecting resources to be recommended based on the predicted recommendation probability at each stage, and then determining the final target resource based on these resources. Taking advertising resources as an example, from ad entry to delivery, the process mainly involves multiple stages such as recall, coarse ranking, and fine ranking. Each subsequent stage typically performs further recommendation probability prediction and filtering based on the results of the previous stage, thereby improving the accuracy of the final ad delivery. The resource recommendation method of this embodiment can be applied to any recommendation probability prediction stage of resource recommendation processing. Taking the aforementioned advertising resources as an example, it can be applied to the recall stage of advertising resources. For new ads lacking interaction data, it can improve the recommendation accuracy of new ads, thereby increasing the cold start success rate of new ads.

[0073] It should be noted that the server involved in the embodiments of this application can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0074] In one exemplary embodiment, both terminal 110 and server 120 can be node devices in a blockchain system, capable of sharing acquired and generated information with other node devices in the blockchain system, thus enabling information sharing among multiple node devices. Multiple node devices in the blockchain system can be configured with the same blockchain, which consists of multiple blocks, and adjacent blocks are related, ensuring that any data tampering in any block can be detected by the next block. This prevents data tampering in the blockchain and guarantees the security and reliability of the data in the blockchain.

[0075] The embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0076] Artificial Intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or computers-controlled machines to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Pre-trained models, also known as large models or foundational models, can be widely applied to downstream tasks across various AI fields after fine-tuning. AI software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0077] Computer vision (CV) is the science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes in recognizing and measuring targets, and then performs image processing to create images more suitable for human observation or transmission to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Large model technology has brought significant changes to the development of computer vision. Pre-trained models in the vision field, such as Swin-transformer, ViT, V-MOE, and MAE, can be quickly and widely applied to specific downstream tasks after fine-tuning.

[0078] Please see Figure 2 The diagram shown is a flowchart illustrating a resource recommendation method provided in an embodiment of this application. This method can be applied to... Figure 1The server in the middle. It should be noted that this specification provides the operation steps of the method as described in the embodiments or flowcharts, but based on conventional or non-inventive labor, more or fewer operation steps may be included. The order of steps listed in the embodiments is only one of many possible execution orders and does not represent the only execution order. In actual system or product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Specifically, as shown in the figures... Figure 2 As shown, the method may include:

[0079] S201, in response to the resource recommendation request of the target object, determine the first resource in the target resource set, which is the resource in the target resource set that meets the new resource conditions.

[0080] Specifically, during terminal use, the terminal can receive resource recommendation operations triggered by the user, generate a corresponding resource recommendation request, and send the request to the backend server. This request can carry the user account information currently logged into the terminal, indicating the target object. For example, a resource recommendation operation could be a refresh of the currently viewed page, or a successful login to an application.

[0081] Taking a resource recommendation operation as an example of refreshing the currently viewed page, the terminal, in response to the refresh operation, can obtain user account information and display environment information, and then generate a resource recommendation request based on the user account information and display environment information and send it to the backend server. The display environment information can include any one or more of the following: application identifier, page identifier, and display location identifier.

[0082] When responding to a resource recommendation request for a target object, the backend server can first determine a target resource set. This target resource set is a collection of resources that match the resource recommendation request. For example, it could be a collection of multiple resources that match the display environment information in the resource recommendation request. Taking advertising resources as an example, the target resource set could be a collection of advertising resources whose placement environment information matches the display environment information in the resource recommendation request. The placement environment information could be pre-determined advertising placement sites by the advertising resource provider. Recommending resources based on the target resource set can improve the efficiency of resource recommendation.

[0083] In this embodiment, resources that meet the new resource conditions are referred to as new resources, and new resources in the target resource set are referred to as first resources. The new resource conditions can be used to filter out new resources. Specific new resource conditions include: the resource's online duration is less than a preset online duration; the addition time of each material in the resource is less than a preset addition time; and the resource does not meet the cold start success conditions.

[0084] Specifically, the conditions for a successful cold start can be represented by the amount of interaction data accumulated for that resource. When the amount of interaction data accumulated for a resource exceeds a preset amount, it indicates that the resource meets the conditions for a successful cold start; conversely, if the amount of interaction data accumulated for a resource does not exceed the preset amount, it indicates that the resource does not meet the conditions for a successful cold start. The preset amount of interaction data can be set based on practical experience.

[0085] The listing of a resource refers to it entering a displayable state, meaning it can be recommended to users. Resources that are listed are called online resources. Conversely, if a resource is not listed, it means it is not currently in a displayable state and cannot be recommended to users. Since unlisted resources cannot be recommended, the backend server can ignore these resources when determining the target resource set. Therefore, all resources in the target resource set can be considered as potentially recommendable to users, i.e., online resources. Taking advertising resources as an example, the listing of an advertisement means it has entered the delivery phase; advertisements in the delivery phase are called online advertisements. The listing duration is the duration starting from the time the resource enters a displayable state. The preset listing duration can be set based on actual needs, such as 3 days.

[0086] The resources in this application embodiment can consist of one or more materials. For example, an advertising resource can consist of one or more creative materials. As a new resource, the resource requires that the addition time of the most recently added material in the resource is less than a preset addition time. This preset addition time can be set based on actual needs, for example, it can be 24 hours. For advertising resources, one or more creative materials are generally added to the advertising resource when the advertisement goes live. For example, advertisement A has creative material a added when it goes live. The preset live duration is 3 days, and the preset addition time is 24 hours. Assuming that advertisement A does not meet the cold start success conditions, then on the first and second days after advertisement A goes live, based on the preset live duration and preset addition time, advertisement A can be determined to be a new advertisement. However, on the third day after going live, since the addition time of the most recently added material, i.e., creative material a, has exceeded 24 hours, although the live duration of advertisement A has not exceeded the preset live duration, the addition time of the most recently added material has exceeded the preset addition time. Therefore, advertisement A can be determined to be not a new advertisement at this time. If creative material b is added to ad A on the third day after it goes live, ad A can be considered a new ad because the added creative material b has not been added for more than 24 hours. However, on the fourth day after it goes live, although the added creative material b has not been added for more than 24 hours, ad A has been online for longer than the preset online time, so ad A is not a new ad.

[0087] S203, based on the resource mapping information, find the target second resource corresponding to the first resource mentioned above.

[0088] The resource mapping information represents multiple second resources associated with different preset first resources. The second resource is a resource whose recommendation index exceeds a preset recommendation index threshold within a preset historical time period. Specifically, the associated second resource is a second resource similar to the corresponding preset first resource.

[0089] Specifically, recommendation metrics can include indicators that characterize the recommendation effect, such as impressions and conversions. Recommendation metrics can also include indicators that characterize the processing effect of the resource processing stage, such as the number of recalls during the recall stage. Understandably, preset recommendation metric thresholds are threshold data that match the recommendation metrics. For example, if recommendation metrics include impressions, conversions, and recalls, then preset recommendation metric thresholds would include impression thresholds, conversion thresholds, and recall thresholds. The specific values ​​of each threshold can be set based on practical experience. The preset historical time period is used to meet real-time requirements and is a preset duration prior to the current time, such as the last 7 days.

[0090] In practice, resource mapping information can be generated offline in advance, and this information can be updated at preset time intervals, such as once a week. Understandably, this preset time interval can be adjusted based on the frequency of new resource deployments in actual applications. Generally, if the frequency of new resource deployments is high, the preset time interval can be set shorter to improve the accuracy of the resource mapping information, thereby improving the accuracy of resource recommendations. The generation of resource mapping information will be described in detail later in the embodiments of this application.

[0091] In some exemplary embodiments, the resource mapping information may include multiple preset correspondences between first resource identifiers and second resource identifiers, wherein the resource identifier is used to uniquely identify a resource. Based on this, step S203 above may include:

[0092] Obtain the first resource identifier of the aforementioned first resource;

[0093] Search the resource mapping information to see if there is a target correspondence that matches the first resource identifier, where the preset first resource identifier in the target correspondence is the first resource identifier;

[0094] If the target correspondence is found, the target second resource identifier in the target correspondence is obtained, and the second resource indicated by the target second resource identifier is the target second resource.

[0095] The above implementation method, by finding the target second resource corresponding to the first resource based on resource identifier matching, is beneficial to improving the resource recommendation efficiency of the embodiments of this application.

[0096] S205, if the target second resource is found, obtain the resource feature representation of the target second resource.

[0097] S207, based on the object feature representation of the target object and the resource feature representation of the second resource of the target object, predict the recommendation probability of recommending the first resource to the target object.

[0098] Specifically, when a target second resource is found, its resource feature representation can be obtained in two ways: One way is to utilize an existing set of resource feature representations. This set stores the resource feature representations of each resource, and these representations are updated based on the actual interaction data of the corresponding resource. The other way is to extract features from the target second resource in real time to obtain its corresponding resource feature representation. Understandably, the former method is more efficient for resource recommendation.

[0099] The object feature representation of the target object can be obtained by feature extraction of the target object.

[0100] Taking the ad recall phase as an example, a dual-tower model is typically used in this phase, consisting of a user-side feature extraction network and an item-side feature extraction network. The user-side feature extraction network extracts features from the input user features (which can be understood as the object features of the target object) to obtain the user's feature representation (referred to as User Embedding). The item-side feature extraction network extracts features from the input candidate ad features to obtain the candidate ad item's feature representation (referred to as Item Embedding). Then, a matching layer calculates the similarity between the user embedding and the item embedding to obtain the probability of recommending a candidate ad item to that user. Therefore, the item embeddings extracted by the item-side feature extraction network during the recall phase can be stored. Subsequently, when dealing with the same item, the feature representation of that item can be directly retrieved. Understandably, to improve the accuracy of the stored item embeddings, the feature representations of each item can be updated at preset time intervals.

[0101] In specific implementation, in step S207 above, when predicting the recommendation probability of recommending the first resource to the target object, the similarity between the object feature representation of the target object and the resource feature representation of the target second resource can be calculated, and this similarity can be used as the recommendation probability of recommending the first resource to the target object. This similarity may include, but is not limited to, cosine similarity.

[0102] For example, if the target second resource corresponding to the first resource is not found based on the resource mapping information, the resource feature representation of the first resource can be obtained, and then the recommendation probability of recommending the first resource to the target object can be predicted based on the resource feature representation of the first resource.

[0103] In the above implementation, based on the collaborative filtering idea that similar items may be of interest to the same object, for a new resource (i.e., the first resource) in the target resource set, the old resources associated with the new resource (i.e., the target second resource) are determined by combining resource mapping information. Then, when predicting the recommendation probability of recommending the new resource to the target object, the resource feature representation of the associated old resource is used to replace the resource feature representation of the new resource. In this way, when the new resource lacks interaction data, it helps the new resource find high-potential conversion interaction objects, improves the recommendation accuracy of the new resource, and increases the cold start success rate of the new resource.

[0104] As can be seen from the above technical solutions of the embodiments of this application, the embodiments of this application may further include generating the aforementioned resource mapping information, which will be discussed below in conjunction with... Figure 3 The process of generating resource mapping information in the embodiments of this application will be described in detail, such as... Figure 3 As shown, the method in this application embodiment may further include:

[0105] S301, Select a second resource that meets the screening criteria from the reference resource set to obtain a candidate second resource set.

[0106] The reference resource set can be a real-time collection of all resources. The filtering criteria include recommending metrics exceeding a preset recommendation metric threshold within a preset historical time period. Specifically, recommending metrics can include indicators characterizing the recommendation effect, such as exposure and conversion rates. They can also include indicators characterizing the processing effect of the resource processing stage, such as the number of recalls during the recall stage. Understandably, the preset recommendation metric threshold is a threshold data matching the recommending metrics. For example, if the recommending metrics include exposure, conversion, and recall, the preset recommending metric threshold would include an exposure threshold, a conversion threshold, and a recall threshold. The specific values ​​of each threshold can be set based on practical experience. The preset historical time period is used to meet real-time requirements and is a preset duration prior to the current time, such as the last 7 days.

[0107] S303, Select the first resource that meets the new resource conditions from the reference resource set to obtain the preset first resource set.

[0108] S305, for each preset first resource in the preset first resource set, determine the similarity between the preset first resource and each candidate second resource in the candidate second resource set.

[0109] In some exemplary implementations, to improve the efficiency and accuracy of resource mapping information generation, step S305 may include the following:

[0110] Visual features are extracted from each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set to obtain the preset resource visual feature vector of each preset first resource and the candidate resource visual feature vector of each candidate second resource.

[0111] For each of the preset first resources, the similarity between the preset first resource and each of the candidate second resources is determined based on the preset resource visual feature vector of the preset first resource and the visual feature vector of each candidate resource.

[0112] Specifically, a pre-trained ViT (Vision Transformer) model can be used for visual feature extraction. In practice, each element in a preset first resource can be acquired first. The ViT model is then used to extract visual features from each element in the preset first resource, resulting in a visual feature vector for each element. These visual feature vectors are then concatenated to obtain the preset resource visual feature vector for the preset first resource. Similarly, a similar visual feature extraction method can be used for each candidate second resource to obtain its candidate resource visual feature vector.

[0113] Therefore, when determining the similarity between the preset first resource and each candidate second resource based on the preset resource visual feature vector and each candidate resource visual feature vector, the similarity between the preset resource visual feature vector of the preset first resource and each candidate resource visual feature vector (e.g., cosine similarity) can be calculated to obtain the similarity between the preset first resource and each candidate second resource.

[0114] The above implementation combines the visual feature representation of resources to characterize the degree of similarity. It can introduce the visual attributes of resources as a basis for judging similarity, which can not only reduce the amount of computation when calculating the degree of similarity, but also ensure the accuracy of the calculation of the degree of similarity to a certain extent, thereby improving the efficiency and accuracy of resource mapping information generation.

[0115] In some exemplary embodiments, to further improve the accuracy of resource mapping information, step S305, when determining the similarity between the preset first resource and each candidate second resource based on the preset resource visual feature vector of the preset first resource and the visual feature vectors of each candidate resource, can adopt the following method:

[0116] Obtain the graph embedding vector of the resource node corresponding to each resource in the reference resource set in the graph structure, wherein the graph structure is constructed based on the historical interaction data between each resource in the reference resource set and the interaction object;

[0117] Determine the graph embedding vector of the resource node corresponding to the preset first resource, and the graph embedding vector of the resource node corresponding to each candidate second resource;

[0118] By concatenating the graph embedding vector of the preset first resource with the visual feature vector of the preset resource, a preset comprehensive resource feature vector of the preset first resource is obtained.

[0119] The graph embedding vector of each candidate second resource is concatenated with the corresponding visual feature vector of the candidate resource to obtain the candidate comprehensive resource feature vector of each candidate second resource.

[0120] The similarity between the preset comprehensive resource feature vector of the preset first resource and the candidate comprehensive resource feature vector of each candidate second resource is determined as the similarity between the preset first resource and each candidate second resource.

[0121] For example, obtaining the graph embedding vector of the resource node corresponding to each resource in the reference resource set in the graph structure may include:

[0122] Based on the historical interaction data between each resource and interactive object in the reference resource set, a graph structure is constructed. The graph structure includes multiple graph nodes and connecting edges between adjacent graph nodes. The multiple graph nodes include resource nodes corresponding to resources and object nodes corresponding to interactive objects. The connecting edges represent the interaction relationship between the corresponding interactive object and the corresponding resource.

[0123] Based on the pre-trained graph neural network model and the graph structure, the graph embedding vector of each resource node is obtained.

[0124] Graph Neural Networks (GNNs) are neural networks that operate directly on graph structures, primarily processing data with non-Euclidean spatial structures (graph structures). They ignore the input order of nodes; during computation, the representation of a node is influenced by its neighboring nodes, while the graph's connections remain unchanged; the graph structure allows for graph-based reasoning. Typically, a GNN consists of two modules: a propagation module and an output module. The propagation module transmits information and updates the state between nodes in the graph, while the output module provides vector representations of the nodes and edges, defining the objective function according to different tasks. Types of GNNs include Graph Convolutional Networks (GCNs), Gated Graph Neural Networks (GGNNs), and Graph Attention Networks (GATs), which utilize attention mechanisms.

[0125] In the above embodiments, a graph structure is constructed by utilizing historical interaction data between resources and objects, and the embedding vectors of each graph node in the graph structure are extracted using a graph neural network model. This allows the graph embedding vector of the resource node corresponding to each resource to be obtained. Then, the graph embedding vector of the preset first resource can be combined with the preset visual feature vector of the resource to obtain the preset comprehensive resource feature vector. Similarly, the graph embedding vector of each candidate second resource can be combined with the corresponding visual feature vector of the candidate resource to obtain the candidate comprehensive resource feature vector. By utilizing the similarity between the preset comprehensive resource feature vector and each candidate comprehensive resource feature vector (such as calculating their cosine similarity), the accuracy of the similarity calculation is improved, which is beneficial to improving the accuracy of resource mapping information.

[0126] S307, Based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set, determine the candidate second resource associated with the preset first resource.

[0127] Among them, the candidate second resource associated with the preset first resource is the candidate second resource with the highest similarity to the preset first resource.

[0128] For example, determining the candidate second resource associated with the preset first resource may include:

[0129] Based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set, a target candidate second resource with a similarity exceeding a preset similarity is determined.

[0130] Obtain the first display environment information of the preset first resource and the second display environment information of the target candidate second resource;

[0131] If the second display environment information matches the first display environment information, then the matching target candidate second resource is determined as the candidate second resource associated with the preset first resource.

[0132] Specifically, the preset similarity level can be set based on actual experience, such as 0.8 or 0.9, etc.

[0133] The first display environment information can characterize the display environment of the preset first resource, and the second display environment information can characterize the display environment of the target candidate second resource. Taking an advertisement as an example, the first display environment information can characterize the placement site of the preset advertisement, and the second display environment information can characterize the placement site of the candidate second advertisement. When the second display environment information matches the first display environment information, the matching target candidate second resource is determined as the candidate second resource associated with the preset first resource. This ensures that the candidate second resource associated with the preset first resource has the same display environment. For advertisements, this means that new (preset first resource) and old (associated candidate second resource) advertisements have the same placement site, thereby making the traffic characteristics of new and old advertisements consistent and improving the accuracy of recommendations based on resource mapping information.

[0134] In some exemplary embodiments, when there are multiple target candidate second resources with a similarity exceeding a preset similarity level, determining the matching target candidate second resource as a candidate second resource associated with a preset first resource may include: determining the number of matching target candidate second resources; and when there are multiple matching target candidate second resources, determining the matching target candidate second resource with the highest similarity as the candidate second resource associated with the preset first resource. For advertisements, the old advertisement with the highest similarity to the new advertisement can be selected from multiple old advertisements on the same site as the new advertisement as the associated old advertisement for the new advertisement, further improving the accuracy of resource mapping information.

[0135] S309, Based on the candidate second resources associated with each of the preset first resources, the resource mapping information is generated.

[0136] In specific implementation, after determining the candidate second resources associated with each preset first resource, the resource identifier of each preset first resource and the resource identifier of each associated candidate second resource can be obtained, and then the correspondence between the resource identifier of the preset first resource and the resource identifier of the associated candidate second resource can be established, thereby obtaining resource mapping information.

[0137] In some exemplary embodiments, in order to improve the efficiency of resource recommendation based on resource mapping information, the method may further include: updating the resource mapping information based on preset exit conditions; wherein the preset exit conditions include: the online duration of the preset first resource exceeds the preset online duration and the feedback index of the preset first resource exceeds a preset feedback index threshold.

[0138] Specifically, feedback metrics can include indicators that characterize the recommendation effect, such as exposure and conversion rates. Preset feedback metric thresholds correspond to the feedback metrics, and may include, for example, exposure thresholds and conversion rate thresholds. When a preset first resource that meets the aforementioned preset exit conditions exists in the resource mapping information, the preset first resource and its associated candidate second resource are deleted. For example, the correspondence between the preset first resource and the resource identifier in the resource mapping information can be deleted.

[0139] In some exemplary embodiments, to ensure the real-time accuracy of resource feature representation and thus improve the accuracy of resource recommendations, the aforementioned step S205 may include:

[0140] If the target second resource is found, obtain the current state of the target second resource;

[0141] If the current state of the target second resource indicates that the target second resource is in a displayable state, then the resource feature representation of the target second resource is obtained, and then the aforementioned step S207 is executed.

[0142] Specifically, when the target second resource corresponding to the first resource is found based on the resource mapping information, it is further determined whether the target second resource is in a displayable state based on its current state. Only when it is in a displayable state is the resource feature representation of the target second resource obtained, and the resource feature representation of the target second resource is used to replace the resource feature representation of the first resource to predict the recommendation probability of recommending the first resource to the target object. For advertising, this can ensure that new (first resource) and old (target second resource) ads are online at the same time, thereby ensuring that the target object is a high-conversion target of the first resource.

[0143] For example, if the current state of the target second resource indicates that the target second resource is in a non-displayable state, then the resource feature representation of the first resource is obtained;

[0144] Based on the object feature representation of the target object and the resource feature representation of the first resource, the probability of recommending the first resource to the target object is predicted.

[0145] Specifically, when a target second resource corresponding to a first resource is found based on resource mapping information, if it is determined that the target second resource is in an undisplayable state based on its current state, the resource feature representation of the first resource is directly used to predict the recommendation probability of recommending the first resource to the target object.

[0146] To facilitate understanding of the technical solutions in the embodiments of this application, the following is combined with... Figure 4 and Figure 5 Let's take the recall phase of advertising recommendations as an example to illustrate this.

[0147] like Figure 4 As shown, the ad recommendation method in the recall stage of this application embodiment is divided into two main stages: offline similarity calculation of ads to generate resource mapping information and accessing the resource mapping information into the recall module for ad recall.

[0148] Phase 1: Offline similarity calculation of advertisements based on resource mapping information.

[0149] 1) Filter old advertisements that meet the criteria

[0150] Specifically, the conditions that old ads need to meet are: the old ads have good performance, such as the number of times the ad is exposed, converted, and recalled is greater than the threshold, and in order to meet the real-time requirements, the ad must be an ad within 7 days.

[0151] 2) Filter new advertisements that meet the criteria

[0152] Specifically, a new ad needs to meet the following conditions: the ad is newly launched, the creative (material) is newly generated, and the ad has not yet met the requirements for a successful cold start.

[0153] 3) Feature extraction to form feature vectors (Embedding)

[0154] Specifically, one approach is to use the ViT model to extract visual feature vectors from both new and old ads, then calculate the cosine similarity between them as their similarity score. The old ad with the highest similarity score (greater than a threshold) is then selected to form a set of new and old ad correspondences in the resource mapping information. Another approach is to introduce a Graph Neural Network (GNN) model. This model uses historical user-ad interaction data to construct a graph structure. The GNN model extracts the graph embedding vectors of each node in the graph structure. The graph embedding vectors corresponding to the new and old ads are then found and combined (concatenated) with their respective visual feature vectors to form their corresponding comprehensive feature vectors. The cosine similarity between the new and old ads is then calculated based on these comprehensive feature vectors, and the old ad with the highest similarity score (greater than a threshold) is selected to form a set of new and old ad correspondences in the resource mapping information.

[0155] To ensure that the traffic characteristics of new and old ads match, the placement sites for new and old ads in a set of new and old ad correspondences need to be the same.

[0156] 4) Exit update of resource mapping information

[0157] Specifically, the corresponding relationship of a new ad that meets the following exit conditions will be removed from the resource mapping information: the new ad has been online for more than 3 days; the new ad's impressions, conversions, and other performance data exceed the threshold.

[0158] Phase Two: Integrate resource mapping information into the recall module for ad recall.

[0159] Specifically, the recall module accesses offline resource mapping information. When recalling a new ad, it finds similar older ads based on the resource mapping information, replaces the recall feature vector of the new ad with the recall feature vector of the older ads, and predicts the recommendation probability of the new ad. This allows for... Figure 5 As shown, new ads can be recommended to the high-conversion target group of old ads, optimizing the recall and ranking effect of new ads, thereby optimizing the initial launch of new ads and improving the cold start success rate of new ads.

[0160] Corresponding to the resource recommendation methods provided in the above embodiments, this application also provides a resource recommendation device. Since the resource recommendation device provided in this application corresponds to the resource recommendation methods provided in the above embodiments, the implementation methods of the aforementioned resource recommendation methods are also applicable to the resource recommendation device provided in this embodiment, and will not be described in detail in this embodiment.

[0161] Please see Figure 6The diagram shown is a structural schematic of a resource recommendation device provided in an embodiment of this application. This device has the function of implementing the resource recommendation method in the above-described method embodiments. This function can be implemented by hardware or by hardware executing corresponding software. Figure 6 As shown, the resource recommendation device 600 may include:

[0162] The first resource determination module 610 is used to determine a first resource in the target resource set in response to a resource recommendation request from the target object; the first resource is a resource in the target resource set that meets the new resource conditions;

[0163] The associated resource search module 620 is used to search for the target second resource corresponding to the first resource based on the resource mapping information; the resource mapping information represents the second resources associated with multiple different preset first resources; the second resource is a resource whose recommendation index exceeds a preset recommendation index threshold within a preset historical time period.

[0164] The resource feature representation acquisition module 630 is used to acquire the resource feature representation of the target second resource when the target second resource is found;

[0165] The first recommendation probability prediction module 640 is used to predict the recommendation probability of recommending the first resource to the target object based on the object feature representation of the target object and the resource feature representation of the target second resource.

[0166] In one exemplary embodiment, the resource mapping information includes multiple preset correspondences between first resource identifiers and second resource identifiers; the associated resource lookup module 620 includes:

[0167] The first resource identifier acquisition module is used to acquire the first resource identifier of the first resource.

[0168] The correspondence lookup module is used to search whether there is a target correspondence in the resource mapping information that matches the first resource identifier; the preset first resource identifier in the target correspondence is the first resource identifier;

[0169] The associated resource determination module is used to obtain the target second resource identifier in the target correspondence when the target correspondence is found, and the second resource indicated by the target second resource identifier is the target second resource.

[0170] In one exemplary embodiment, the apparatus further includes a resource mapping information generation module, which includes:

[0171] The first filtering module is used to select second resources that meet the filtering criteria from the reference resource set to obtain a candidate second resource set; the filtering criteria include that the recommended index in the preset historical time period exceeds the preset recommended index threshold.

[0172] The second filtering module is used to select the first resource that meets the new resource conditions from the reference resource set, so as to obtain the preset first resource set;

[0173] The similarity determination module is used to determine the similarity between each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set for each preset first resource.

[0174] The associated candidate resource determination module is used to determine the associated candidate second resources based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set;

[0175] The resource mapping information generation submodule is used to generate the resource mapping information based on the candidate second resources associated with each preset first resource.

[0176] In one exemplary implementation, the associated candidate resource determination module includes:

[0177] The first determining module is used to determine a target candidate second resource whose similarity exceeds a preset similarity level based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set.

[0178] The display environment information acquisition module is used to acquire the first display environment information of the preset first resource and the second display environment information of the target candidate second resource;

[0179] The second determining module is used to determine the matching target candidate second resource as the candidate second resource associated with the preset first resource when the second display environment information matches the first display environment information.

[0180] In one exemplary implementation, the second determining module includes:

[0181] The quantity determination module is used to determine the number of matching target candidate second resources;

[0182] The third determining module is used to determine the matching target candidate second resource with the highest similarity as the candidate second resource associated with the preset first resource when there are multiple such resources.

[0183] In one exemplary implementation, the similarity determination module includes:

[0184] The visual feature extraction module is used to extract visual features from each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set, so as to obtain the preset resource visual feature vector of each preset first resource and the candidate resource visual feature vector of each candidate second resource.

[0185] The similarity determination submodule is used to determine the similarity between the preset first resource and each candidate second resource for each preset first resource based on the preset resource visual feature vector of the preset first resource and the visual feature vector of each candidate resource.

[0186] In one exemplary implementation, the similarity determination submodule includes:

[0187] The graph embedding vector acquisition module is used to obtain the graph embedding vector of the resource node corresponding to each resource in the reference resource set in the graph structure; the graph structure is constructed based on the historical interaction data between each resource and the interaction object in the reference resource set;

[0188] The graph embedding vector determination module is used to determine the graph embedding vector of the resource node corresponding to the preset first resource, and the graph embedding vector of the resource node corresponding to each candidate second resource.

[0189] The first splicing module is used to splice the graph embedding vector of the preset first resource with the visual feature vector of the preset resource to obtain the preset comprehensive resource feature vector of the preset first resource.

[0190] The second splicing module is used to splice the graph embedding vector of each candidate second resource with the corresponding candidate resource visual feature vector to obtain the candidate comprehensive resource feature vector of each candidate second resource.

[0191] The vector similarity calculation module is used to determine the similarity between the preset comprehensive resource feature vector of the preset first resource and the candidate comprehensive resource feature vector of each candidate second resource, and use this as the similarity between the preset first resource and each candidate second resource.

[0192] In one exemplary implementation, the graph embedding vector acquisition module includes:

[0193] The graph structure construction module is used to construct a graph structure based on the historical interaction data between each resource and interactive object in the reference resource set. The graph structure includes multiple graph nodes and connecting edges between adjacent graph nodes. The multiple graph nodes include resource nodes corresponding to resources and object nodes corresponding to interactive objects. The connecting edges represent the interaction relationship between the corresponding interactive object and the corresponding resource.

[0194] The graph embedding vector generation module is used to obtain the graph embedding vector of each resource node based on the pre-trained graph neural network model and the graph structure.

[0195] In one exemplary implementation, the new resource conditions include: the resource's online duration is less than a preset online duration, the duration of the most recently added material in the resource is less than a preset addition duration, and the resource does not meet the cold start success conditions.

[0196] In one exemplary embodiment, the device further includes:

[0197] The update module is used to update the resource mapping information based on preset exit conditions;

[0198] The preset exit conditions include: the online duration of the preset first resource exceeds the preset online duration and the feedback indicators of the preset first resource exceed the preset feedback indicator threshold.

[0199] In one exemplary implementation, the resource feature representation acquisition module 630 includes:

[0200] The current status acquisition module is used to acquire the current status of the target second resource when the target second resource is found;

[0201] The second resource feature representation acquisition submodule is used to acquire the resource feature representation of the target second resource when the current state indicates that the target second resource is in a displayable state.

[0202] In one exemplary embodiment, the resource feature representation acquisition module 630 further includes:

[0203] The first resource feature representation acquisition submodule is used to acquire the resource feature representation of the first resource when the current state indicates that the target second resource is in an undisplayable state.

[0204] The second recommendation probability prediction module is used to predict the recommendation probability of recommending the first resource to the target object based on the object feature representation of the target object and the resource feature representation of the first resource.

[0205] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0206] This application provides an electronic device including a processor and a memory. The memory stores at least one instruction or at least one program segment, which is loaded and executed by the processor to implement any of the resource recommendation methods provided in the above method embodiments.

[0207] Memory can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for the functions, etc.; the data storage area can store data created based on the use of the device, etc. Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, memory can also include a memory controller to provide the processor with access to the memory.

[0208] The methods and embodiments provided in this application can be executed in a computer terminal, server, or similar computing device; that is, the aforementioned electronic device may include a computer terminal, server, or similar computing device. Taking running on a server as an example... Figure 7 This is a hardware structure block diagram of an electronic device running a resource recommendation method provided in an embodiment of this application, such as... Figure 7 As shown, the server 700 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 710 (CPUs 710 may include, but are not limited to, microprocessors such as MCUs or programmable logic devices such as FPGAs), a memory 730 for storing data, and one or more storage media 720 (e.g., one or more mass storage devices) for storing application programs 723 or data 722. The memory 730 and storage media 720 may be temporary or persistent storage. The program stored in the storage media 720 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the CPU 710 may be configured to communicate with the storage media 720 and execute the series of instruction operations stored in the storage media 720 on the server 700. Server 700 may also include one or more power supplies 760, one or more wired or wireless network interfaces 750, one or more input / output interfaces 740, and / or one or more operating systems 721, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0209] The input / output interface 740 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 700. In one example, the input / output interface 740 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the input / output interface 740 may be a radio frequency (RF) module for wireless communication with the Internet.

[0210] Those skilled in the art will understand that Figure 7 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 700 may also include... Figure 7 The more or fewer components shown, or having the same Figure 7 The different configurations shown.

[0211] Embodiments of this application also provide a computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one program related to implementing a resource recommendation method. The at least one instruction or the at least one program is loaded and executed by the processor to implement any of the resource recommendation methods provided in the embodiments of this application.

[0212] Embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the resource recommendation methods provided in the embodiments of this application.

[0213] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0214] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0215] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0216] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0217] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A resource recommendation method, characterized in that, The method includes: In response to a resource recommendation request from a target object, a first resource in the target resource set is determined; the first resource is a resource in the target resource set that meets the conditions for a new resource. Based on resource mapping information, the target second resource corresponding to the first resource is found; the resource mapping information represents the second resources associated with multiple different preset first resources; the second resource is a resource whose recommendation index exceeds a preset recommendation index threshold within a preset historical time period. If the target second resource is found, obtain the resource feature representation of the target second resource; Based on the object feature representation of the target object and the resource feature representation of the target second resource, the probability of recommending the first resource to the target object is predicted.

2. The method according to claim 1, characterized in that, The resource mapping information includes multiple preset correspondences between first resource identifiers and second resource identifiers; the step of finding the target second resource corresponding to the first resource based on the resource mapping information includes: Obtain the first resource identifier of the first resource; Search the resource mapping information to see if there is a target correspondence that matches the first resource identifier; the preset first resource identifier in the target correspondence is the first resource identifier. If the target correspondence is found, the target second resource identifier in the target correspondence is obtained, and the second resource indicated by the target second resource identifier is the target second resource.

3. The method according to claim 1, characterized in that, The method further includes generating the resource mapping information, wherein generating the resource mapping information includes: A second resource that meets the screening criteria is selected from the reference resource set to obtain a candidate second resource set; the screening criteria include the recommendation index exceeding the preset recommendation index threshold within the preset historical time period. Select a first resource that meets the new resource conditions from the reference resource set to obtain a preset first resource set; For each preset first resource in the preset first resource set, determine the degree of similarity between the preset first resource and each candidate second resource in the candidate second resource set; Based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set, the candidate second resource associated with the preset first resource is determined; Based on the candidate second resources associated with each of the preset first resources, the resource mapping information is generated.

4. The method according to claim 3, characterized in that, The step of determining the candidate second resource associated with the preset first resource based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set includes: Based on the similarity between the preset first resource and each candidate second resource in the candidate second resource set, a target candidate second resource with a similarity exceeding a preset similarity is determined. Obtain the first display environment information of the preset first resource and the second display environment information of the target candidate second resource; If the second display environment information matches the first display environment information, then the matching target candidate second resource is determined as the candidate second resource associated with the preset first resource.

5. The method according to claim 4, characterized in that, The step of determining the matching target candidate second resource as the candidate second resource associated with the preset first resource includes: Determine the number of matching target candidate second resources; When there are multiple such resources, the matching target candidate second resource with the highest similarity is determined as the candidate second resource associated with the preset first resource.

6. The method according to claim 3, characterized in that, Determining the similarity between each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set includes: Visual features are extracted from each preset first resource in the preset first resource set and each candidate second resource in the candidate second resource set to obtain the preset resource visual feature vector of each preset first resource and the candidate resource visual feature vector of each candidate second resource. For each of the preset first resources, the similarity between the preset first resource and each of the candidate second resources is determined based on the preset resource visual feature vector of the preset first resource and the visual feature vector of each candidate resource.

7. The method according to claim 6, characterized in that, The step of determining the similarity between the preset first resource and each candidate second resource based on the preset resource visual feature vector of the preset first resource and the visual feature vectors of each candidate resource includes: Obtain the graph embedding vector of the resource node corresponding to each resource in the reference resource set in the graph structure; the graph structure is constructed based on the historical interaction data between each resource in the reference resource set and the interaction object; Determine the graph embedding vector of the resource node corresponding to the preset first resource, and the graph embedding vector of the resource node corresponding to each candidate second resource; By concatenating the graph embedding vector of the preset first resource with the visual feature vector of the preset resource, a preset comprehensive resource feature vector of the preset first resource is obtained. The graph embedding vector of each candidate second resource is concatenated with the corresponding visual feature vector of the candidate resource to obtain the candidate comprehensive resource feature vector of each candidate second resource. The similarity between the preset comprehensive resource feature vector of the preset first resource and the candidate comprehensive resource feature vector of each candidate second resource is determined as the similarity between the preset first resource and each candidate second resource.

8. The method according to claim 7, characterized in that, The step of obtaining the graph embedding vector of the resource node corresponding to each resource in the reference resource set in the graph structure includes: Based on the historical interaction data between each resource and interactive object in the reference resource set, a graph structure is constructed. The graph structure includes multiple graph nodes and connecting edges between adjacent graph nodes. The multiple graph nodes include resource nodes corresponding to resources and object nodes corresponding to interactive objects. The connecting edges represent the interaction relationship between the corresponding interactive object and the corresponding resource. Based on the pre-trained graph neural network model and the graph structure, the graph embedding vector of each resource node is obtained.

9. The method according to claim 1, characterized in that, The conditions for the new resource include: the online duration of the resource is less than the preset online duration, the addition duration of the most recently added material in the resource is less than the preset addition duration, and the resource does not meet the conditions for a successful cold start.

10. The method according to claim 9, characterized in that, The method further includes: The resource mapping information is updated based on preset exit conditions; The preset exit conditions include: the online duration of the preset first resource exceeds the preset online duration and the feedback index of the preset first resource exceeds the preset feedback index threshold.

11. The method according to claim 1, characterized in that, The step of obtaining the resource feature representation of the target second resource when the target second resource is found includes: If the target second resource is found, obtain the current state of the target second resource; If the current state indicates that the target second resource is in a displayable state, then the resource feature representation of the target second resource is obtained.

12. The method according to claim 1, characterized in that, The method further includes: If the current state indicates that the target second resource is in an undisplayable state, then obtain the resource feature representation of the first resource; Based on the object feature representation of the target object and the resource feature representation of the first resource, the probability of recommending the first resource to the target object is predicted.

13. A resource recommendation device, characterized in that, The device includes: The first resource determination module is used to determine a first resource in the target resource set in response to a resource recommendation request from the target object; the first resource is a resource in the target resource set that meets the new resource conditions; The associated resource search module is used to search for the target second resource corresponding to the first resource based on resource mapping information; the resource mapping information represents the second resources associated with multiple different preset first resources; the second resource is a resource whose recommendation index exceeds a preset recommendation index threshold within a preset historical time period; The resource feature representation acquisition module is used to acquire the resource feature representation of the target second resource when the target second resource is found. The first recommendation probability prediction module is used to predict the recommendation probability of recommending the first resource to the target object based on the object feature representation of the target object and the resource feature representation of the target second resource.

14. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the resource recommendation method as described in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor to implement the resource recommendation method as described in any one of claims 1 to 12.

Images