Resource screening method and apparatus, computer device and storage medium

Abstract

The present application belongs to the technical field of computers. Disclosed are a resource screening method and apparatus, a computer device and a storage medium. The method comprises: acquiring historical behavior information of a target and model instruction information, and on the basis of the historical behavior information, generating prompt information, the prompt information being used for indicating resources on which the target has performed interaction behaviors; by means of a large language model, and on the basis of the model instruction information and the prompt information, generating interest description information; and, on the basis of the interest description information and information of a plurality of resources, screening out a first resource from amongst the plurality of resources, the similarity between the information of the first resource and the interest description information being greater than the similarity between the information of a second resource and the interest description information, and the second resource being a resource amongst the plurality of resources excepting the first resource.

Description

Resource screening method, device, computer equipment and storage medium

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on September 21, 2023, with application number 2023112203260 and entitled “Resource Screening Method, Device, Computer Equipment and Storage Medium,” the entire contents of which are incorporated herein by reference. Technical Field

[0002] The embodiments of the present application relate to the field of computer technology, and in particular to a resource screening method, apparatus, computer equipment, and storage medium.

[0003] background

[0004] With the development of Internet technology, the resources in the flower network are becoming more and more abundant and diverse. In order to enable users to perform interactive behaviors on the resources they are interested in, resources of interest to users are usually recommended.

[0005] Technical content

[0006] This embodiment of the present application provides a resource screening method, the method comprising:

[0007] Obtaining historical behavior information and model indication information of an object, wherein the historical behavior information indicates that the object has performed interactive behaviors on resources, and the model indication information is used to instruct a large language model to generate interest description information based on input information, wherein the interest description information is used to describe the resources that the object is interested in;

[0008] Based on the historical behavior information, generating prompt information, the prompt information is used to indicate the resources on which the object has performed interactive behaviors;

[0009] generating interest description information based on the large language model and the model indication information and the prompt information;

[0010] Based on the interest description information and information of multiple resources, a first resource is filtered out from the multiple resources, the similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information, and the second resource is a resource among the multiple resources other than the first resource.

[0011] An embodiment of the present application provides a resource screening device, the device comprising:

[0012] an acquisition module, configured to acquire historical behavior information and model indication information of an object, wherein the historical behavior information indicates that the object has performed interactive behaviors on resources, and the model indication information is used to instruct a large language model to generate interest description information based on input information, wherein the interest description information is used to describe the resources of interest to the object;

[0013] A generating module, configured to generate prompt information based on the historical behavior information, wherein the prompt information is used to indicate the resources on which the object has performed interactive behaviors;

[0014] The generating module is further configured to generate interest description information based on the large language model, the model indication information and the prompt information;

[0015] A screening module is used to screen out a first resource from the multiple resources based on the interest description information and the information of the multiple resources, wherein the similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information, and the second resource is a resource among the multiple resources other than the first resource.

[0016] An embodiment of the present application provides a computer device, which includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the resource screening method described in the above aspects.

[0017] An embodiment of the present application provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed by the resource screening method described in the above aspects.

[0018] An embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the operations performed by the resource screening method described in the above aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] FIG1 is a schematic diagram of a structure of an implementation environment provided by an embodiment of the present application;

[0021] FIG2 is a flow chart of a resource screening method provided in an embodiment of the present application;

[0022] FIG3 is a flow chart of another resource screening method provided in an embodiment of the present application;

[0023] FIG4 is a schematic structural diagram of a resource screening device provided in an embodiment of the present application;

[0024] FIG5 is a schematic structural diagram of another resource screening device provided in an embodiment of the present application;

[0025] FIG6 is a schematic structural diagram of a terminal provided in an embodiment of the present application;

[0026] FIG7 is a schematic diagram of the structure of a server provided in an embodiment of the present application.

[0027] Implementation Method

[0028] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0029] As used herein, the terms "first," "second," "third," "fourth," and the like may be used to describe various concepts herein, but unless otherwise specified, these concepts are not limited by these terms. These terms are used solely to distinguish one concept from another. For example, a first resource may be referred to as a second resource, and similarly, a second resource may be referred to as a first resource without departing from the scope of this application.

[0030] As used herein, the terms "at least one," "a plurality," "each," and "any" include one, two, or more, "a plurality" includes two or more, "each" refers to each of the corresponding plurality, and "any" refers to any one of the plurality. For example, a plurality of resources includes three resources, and "each" refers to each of the three resources. Any refers to any one of the three resources, which can be the first resource, the second resource, or the third resource.

[0031] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the historical behavior information and model indication information of the objects involved in this application are all obtained with full authorization.

[0032] In some embodiments, before recommending resources to a user, it is necessary to screen out resources of interest to the user. Resource screening methods in related technologies have poor accuracy. The present application provides a resource screening method that can improve the accuracy of the screened resources. The solution provided in the present application is based on artificial intelligence machine learning technology, which can train a large language model and then utilize the trained large language model to implement the resource screening method.

[0033] The resource screening method provided in the embodiments of the present application can be executed by a computer device. In some embodiments, the computer device is a terminal or a server. In some embodiments, the server is an independent physical server, or 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 communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. In some embodiments, the terminal is a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, intelligent voice interaction device, smart home appliance and car terminal, etc., but is not limited to this.

[0034] In some embodiments, the computer program involved in the embodiments of the present application can be deployed and executed on a computer device, or on multiple computer devices located at one location, or on multiple computer devices distributed at multiple locations and interconnected through a communication network. Multiple computer devices distributed at multiple locations and interconnected through a communication network can constitute a blockchain system.

[0035] In some embodiments, the computer device is provided as a server. FIG1 is a schematic diagram of an implementation environment provided by an embodiment of the present application. Referring to FIG1 , the implementation environment includes a terminal 101 and a server 102. The terminal 101 and the server 102 are connected via a wireless or wired network. The terminal 101 is used to display resources, and the server 102 is used to provide a resource recommendation service. The server 102 can recommend resources to the terminal 101 through a large language model so that the object to which the terminal 101 belongs can view the recommended resources.

[0036] In one possible implementation, an application provided by server 102 is installed on terminal 101. Terminal 101 can use this application to perform functions such as viewing and sharing resources. In some embodiments, the application is an application in the operating system of terminal 101 or an application provided by a third party. For example, the application is a content sharing application that has a content sharing function. Of course, the content sharing application can also have other functions, such as review functions, shopping functions, navigation functions, game functions, etc.

[0037] Terminal 101 is used to log in to the application based on the object identifier and view the resources in the application through the application. Server 102 is used to recommend resources of interest to the object referred to by the object identifier to terminal 101 through a large language model, and terminal 101 displays the recommended resources for the object to view.

[0038] FIG2 is a flowchart of a resource screening method provided in an embodiment of the present application, which is executed by a computer device. As shown in FIG2 , the method includes the following steps.

[0039] 201. The computer device obtains historical behavior information and model indication information of the object. The historical behavior information is used to indicate that the object has performed interactive behaviors on resources. The model indication information is used to instruct the large language model to generate interest description information based on input information. The interest description information is information that describes the interests of the object. The interest description information is used to describe the resources that the object is interested in.

[0040] In an embodiment of the present application, since the historical behavior information of the object indicates the resources on which the object has performed interactive behaviors, and thus can reflect the interests of the object, through the generalization and reasoning capabilities of the large language model itself, it can combine the historical behavior information to generate a natural language description for describing the object's preferences, which is used as the object's interest representation. Combined with the information of the resources, the resources that the object is interested in can be screened out.

[0041] Large language models (LLMs) inherently contain rich knowledge and information, possessing generalization and reasoning capabilities. Examples include open-source large language models such as LLaMA (Large Language Model Meta AI, an open-source large model) and ChatGLM (Chat General Language Model). Large language models can be used to generate information related to input information. For example, if the input information for a large language model is "generate a description of a scenic recommendation," the large language model can generate text information about the scenic recommendation based on this input information. An object is any object, such as a user. Historical behavior information indicates which resource an object has interacted with. The interactions indicated by historical behavior information can be any type of behavior, such as positive behaviors such as clicks, likes, favorites, and shares. Resources are any resources, such as item links, multimedia resources, or other resources.

[0042] The model indication information is equivalent to the task indication information of the large language model, which instructs the large language model what task to perform, that is, instructs the large language model what processing to perform based on the input information and what information to generate. For example, the model indication information is "You are a personalized recommendation system, and I will tell you a list of items that the user has clicked on before. You mine the user's interest preferences from his previous historical behavior and write a paragraph of language to describe the user's preferences." Historical behavior information, model indication information, and interest description information can all be expressed in any form. For example, if historical behavior information, model indication information, and interest description information are all expressed in text form, then the model indication information instructs the large language model to generate text information that describes the interest of the object.

[0043] 202. The computer device generates prompt information based on the historical behavior information, where the prompt information is used to indicate the resources on which the object has performed interactive behaviors.

[0044] In an embodiment of the present application, since the historical behavior information indicates on which resource the object has performed interactive behavior, based on the historical behavior information of the object, it is possible to obtain which resources the object has performed interactive behavior on during the historical period, and then generate a prompt information to indicate the resources on which the object has performed interactive behavior.

[0045] For example, the prompt message is "The following are the products that the subject has recently clicked: badminton racket, sports pants, and outdoor sunscreen."

[0046] 203. The computer device generates interest description information based on the large language model, model indication information and prompt information.

[0047] In an embodiment of the present application, the large language model has generalization and reasoning capabilities. Through the large language model, according to the processing method indicated by the model indication information, combined with the resources on which the object has performed interactive behaviors in the prompt information, the resources that the object is interested in are inferred to obtain the interest description information.

[0048] The interest description information can be any form of information, for example, text information expressed in natural language. For example, the interest description information may be "Based on the items recently clicked by the subject, the subject's shopping interests may be related to sports, outdoor activities, personal care, and body protection products, such as fitness equipment, sportswear, outdoor gear, sunscreen, skin care products, protective gloves, knee pads, elbow pads, etc. In addition, the subject may also be interested in other sports and sports equipment, such as basketball, football, tennis, etc."

[0049] 204. The computer device selects a first resource from the multiple resources based on the interest description information and the information of the multiple resources, the similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information, and the second resource is a resource other than the first resource in the multiple resources.

[0050] In an embodiment of the present application, the interest description information is used to describe the resources that the object is interested in, and can express the object's interests and hobbies, while the resource information is used to characterize the resources. Therefore, by comparing the interest description information and the resource information, the similarity between the interest description information and the information of each resource can be determined. The similarity can reflect whether the object is interested in the resource, and then the first resource with the greater similarity is screened out from multiple resources, that is, the resources that the object is interested in are screened out to ensure the accuracy of the screened resources.

[0051] The resource information can be any form of information, for example, text information or other types of information, such as the resource information including the resource title, category, and brief introduction of the resource.

[0052] In an embodiment of the present application, since the historical behavior information of the object indicates the resources on which the object has performed interactive behaviors, it can reflect the object's interests, and the large language model itself has generalization and reasoning capabilities. Therefore, based on the model indication information, the large language model is instructed to combine the historical behavior information to generate a natural language description for describing the object's preferences, which is used as the object's interest representation. Combined with the information of the resources, the resources that the object is interested in are screened out, thereby realizing a resource screening method based on the large language model, ensuring that the screened resources match the object's interests, and ensuring the accuracy of the screened resources.

[0053] Based on the embodiment shown in Figure 2, the embodiment of the present application can also combine multiple historical behavior information to generate prompt information, and use similarity calculation to filter out the first resource and recommend the first resource to the object. The specific process is detailed in the following embodiment.

[0054] FIG3 is a flowchart of another resource screening method provided in an embodiment of the present application, which is executed by a computer device. As shown in FIG3 , the method includes the following steps.

[0055] 301. A computer device obtains multiple pieces of historical behavior information and model indication information of an object, where the historical behavior information indicates that the object has performed interactive behaviors on resources, and the model indication information indicates that a large language model generates interest description information based on input information, where the interest description information is used to describe resources that the object is interested in.

[0056] In the embodiment of the present application, different historical behavior information is generated at different times, and the generation time is equivalent to the time when the object performs the interactive behavior on the resource.

[0057] In one implementation, the historical behavior information includes an object identifier, a resource identifier, and a generation time, and the object identifier indicates the object that performs the interactive behavior.

[0058] Among them, the resource identifier indicates the resource on which the object has performed interactive behavior. The resource identifier is a unique identifier used to represent a resource, which is unique and stable. The resource identifier is represented by a number or a string. Based on the resource identifier, the indicated resource can be determined, and the resource indicated by the resource identifier can also be queried in the resource library. In some embodiments, the historical behavior information also includes an operation identifier, which indicates the interactive behavior performed by the object on the resource.

[0059] In one implementation, the model indication information further indicates the type of input information.

[0060] In an embodiment of the present application, the type of input information can reflect what kind of information the input information is, and the type of input information is indicated in the model indication information so that the large language model can process the input information in a processing method that matches the type indicated by the model indication information, and then generate information for describing the object's interest to ensure the accuracy of subsequently generated information.

[0061] In one implementation, the obtained model indication information matches the recommended domain.

[0062] The recommended fields refer to the categories of resources, and different recommended fields have different resources. For example, recommended fields include movies, music, books, news, etc.

[0063] In an embodiment of the present application, the large language model can be applied to multiple recommendation fields. Each recommendation field corresponds to a model indication information. In any recommendation field, the model indication information matching the recommendation field is obtained to ensure that the subsequent large language model can output information related to the recommendation field. In this way, it can be ensured that the large language model can be applicable to multiple recommendation scenarios. There is no need to deploy a large language model for each recommendation field, which ensures the applicability of the large language model and saves the cost of deploying the large language model.

[0064] In some embodiments, the method of obtaining model indication information includes: in response to a resource recommendation instruction, obtaining model indication information matching the current recommendation field.

[0065] The resource recommendation instruction indicates recommending resources to the object, and the resource recommendation instruction is actively triggered by the object or triggered in response to a resource acquisition instruction of the object.

[0066] 302. The computer device determines first behavior information from the multiple pieces of historical behavior information based on the generation time of the multiple pieces of historical behavior information of the object, where the generation time of the first behavior information is later than the generation time of the second behavior information, and the second behavior information is the historical behavior information in the multiple pieces of historical behavior information other than the first behavior information.

[0067] In an embodiment of the present application, historical behavior information has a generation time, which is equivalent to the time when the object performs interactive behavior on the resource. The later the generation time, the more the corresponding historical behavior information can reflect the interests and preferences of the object. Therefore, from the multiple historical behavior information of the object, the historical behavior information with a later generation time is selected so that the user's interest description can be determined subsequently based on the historical behavior information with a later generation time.

[0068] In one possible implementation, step 302 includes: determining a first number of first behavior information from multiple historical behavior information based on the generation time of multiple historical behavior information of the object; or, determining historical behavior information whose interval between the generation time and the current time is less than the target duration as the first behavior information.

[0069] The first number is any number, and the target duration is any duration.

[0070] 303. The computer device extracts the resources on which the object has performed the interactive behavior from the first behavior information.

[0071] In the embodiment of the present application, each piece of historical behavior information records the resources with which the subject has performed interactive behaviors. Therefore, when a first piece of behavior information is determined from multiple pieces of historical behavior information, the resources with which the subject has performed interactive behaviors can be extracted from each piece of first behavior information. Therefore, the extracted resources are the resources with which the subject has performed interactive behaviors extracted from the first piece of behavior information.

[0072] In one possible implementation, the historical behavior information also includes an operation identifier, then step 303 includes: when multiple first behavior information are determined, based on the operation identifier contained in the first behavior information, filtering the first behavior information containing the target operation identifier from the multiple first behavior information, and extracting the resources on which the object has performed interactive behaviors from the first behavior information containing the target operation identifier.

[0073] Among them, the target operation identifier indicates positive interactive behavior, which refers to interactive behavior that has a positive effect on resources. For example, positive interactive behavior includes click, like, favorite, share and other interactive behaviors.

[0074] 304. The computer device generates prompt information based on the extracted resources, where the prompt information indicates the resources on which the object has performed interactive behaviors.

[0075] In an embodiment of the present application, when resources on which the object has performed interactive behaviors during the historical period are extracted, these resources are summarized into prompt information to indicate the resources on which the object has performed interactive behaviors during the historical period. In an embodiment of the present application, since the historical behavior information of the object is relatively sparse, prompt information is constructed through the historical behavior information of the object so that the subsequent large language model can generate interest description information based on the prompt information. In an embodiment of the present application, the later the generation time of the historical behavior information, the more the historical behavior information can reflect the interest of the object. Therefore, according to the generation time of the historical behavior information, the prompt information is generated using the historical behavior information with a later generation time to ensure that the prompt information can reflect the resources on which the object has recently performed interactive behaviors, so as to ensure the accuracy of the prompt information, so that when the subsequent large language model generates interest description information based on the prompt information, it can take into account the recent changes in the object's interests and ensure the accuracy of the generated interest description information.

[0076] In one possible implementation, step 304 includes: obtaining a prompt information template, the prompt information template includes a resource filling position and relationship description information, the relationship description information is used to describe the relationship between the resource and the object in the resource filling position, and specifically may include the interactive behavior between the object and the resource; adding the extracted resource to the resource filling position to obtain prompt information.

[0077] In an embodiment of the present application, a prompt information template is used to generate prompt information. In order to facilitate the generation of prompt information, a prompt information template is preset. Prompt information can be generated by simply filling in resources in the resource filling position contained in the prompt information template, so as to ensure the convenience of generating prompt information, and to ensure that the generated prompt information can be easily understood by the large language model, thereby ensuring the accuracy of the prompt information.

[0078] For example, the relationship description information is "The following are products that the user has recently clicked on" or "The following are products that the user has recently purchased."

[0079] In one implementation, the prompt information template includes a first information template or a second information template, the first information template includes a resource filling position and a first relationship description information, the first relationship description information is used to describe the relationship between the resource in the resource filling position and the object; the second information template includes a first filling position, a second filling position, the first relationship description information and the second relationship description information, the first filling position is used to fill in the resources on which the object has performed interactive behaviors, and the second filling position is used to fill in the resources on which the object has performed interactive behaviors most recently, the first relationship description information is used to describe the relationship between the resource in the first filling position and the object, and the second relationship description information is used to describe the relationship between the resource in the second filling position and the object.

[0080] The resource on which the object has most recently performed an interactive behavior is, that is, a resource in the historical behavior information of the object whose generation time is closest to the current time.

[0081] In some embodiments, the process of generating prompt information includes: when the prompt information template is a first information template, adding the extracted resources to the resource filling position to obtain prompt information; when the prompt information template is a second information template, adding the extracted resources to the first filling position, and adding the resource of the object in the extracted resources that most recently performed an interactive behavior to the second filling position to obtain prompt information.

[0082] In this embodiment of the present application, the prompt information generated based on the first information template can only reflect the resources with which the subject has interacted historically. However, the prompt information generated based on the second information template can not only reflect the resources with which the subject has interacted historically, but also highlight the resources with which the subject has interacted recently. This allows the subsequent large language model to take into account the subject's recent interest changes when generating interest description information based on the prompt information, ensuring the accuracy of the generated interest description information.

[0083] For example, when the prompt information template is the first information template, the first relationship description information is "The following are the products that the user has recently clicked on", then the generated prompt information is "The following are the products that the user has recently clicked on: badminton rackets, sports pants, outdoor sunscreen".

[0084] For example, when the prompt information template is the second information template, the first relationship description information is "The following are the products that the user has clicked recently", and the second relationship description information is "The product that the user clicked on most recently is", then the generated prompt information is "The following are the products that the user has clicked on recently: badminton racket, sports pants, outdoor sunscreen. The product that the user clicked on most recently is outdoor sunscreen."

[0085] It should be noted that the embodiment of the present application generates prompt information based on the first behavior information screened out from multiple historical behavior information. In another embodiment, there is no need to execute the above steps 302-304, but other methods are adopted to generate prompt information based on historical behavior information.

[0086] 305. The computer device generates interest description information based on the large language model, model indication information and prompt information, and the interest description information is used to describe the resources that the object is interested in.

[0087] In one implementation, the interest description information is used to describe the resource type that the object is interested in and the resources that the object is interested in under the resource type; or, the interest description information is used to describe the resources and the types of resources that the object may perform interactive behaviors on next time.

[0088] The resource type can be any resource type, for example, the resource type can be a sports type, an outdoor type, a sports equipment type, etc.

[0089] In this embodiment of the present application, since a resource type contains multiple resources, the interest description information not only describes the resource type the object is interested in, but also describes which resources within the resource type the object is interested in, thereby ensuring the accuracy of the interest description information. When the interest description information describes the resource and resource type with which the object is likely to perform an interactive behavior next, it reflects the object's historical behavior information and predicts which resource the object is likely to perform an operation on in the future, thereby reflecting the object's future interest in the resource, thereby ensuring the accuracy of the interest description information.

[0090] In one implementation, the model indication information further indicates the format of the interest description information; step 305 includes: generating interest description information in the format based on the model indication information and the prompt information through the large language model.

[0091] In an embodiment of the present application, the model indication information indicates the format of the generated interest description information, so that the interest description information belonging to this format is generated through the large language model to ensure that the generated interest description information can accurately describe the interest of the object, thereby ensuring the accuracy of the interest description information.

[0092] For example, the model indication information indicates that the interest description information is text information, and the interest description information contains a certain number of words; then the interest description information generated by the large language model is text information containing the certain number of words.

[0093] In one implementation, the prompt information is generated based on the first information template or the second information template. When the prompt information is generated based on the first information template or the second information template, the generated interest description information is different. In an embodiment of the present application, when the prompt information is generated based on the first information template, the interest description information obtained by the large language model inference does not specifically distinguish the interests of the object, and the interest description information focuses on the stable long-term interests of the object. When the prompt information is generated based on the second information template, the interest description information obtained by the large language model inference pays more attention to the short-term interest preferences of the object, while also taking into account the stable long-term interests of the object.

[0094] For example, when the prompt information is generated based on the first information template, the interest description information inferred by the large language model is "Based on the products that the user recently clicked, it can be predicted that the user's shopping interest preferences at the next moment may be products related to sports, outdoor activities, personal care and body protection, such as fitness equipment, sportswear, outdoor equipment, sunscreen, skin care products, protective gloves, knee pads, elbow pads, etc. In addition, the user may also be interested in other sports and sports equipment, such as basketball, football, tennis, etc."

[0095] For example, when the prompt information is generated based on the second information template, the interest description information inferred by the large language model is "Based on the user's previous behavior, it can be seen that he is more interested in sports and outdoor activities, and also pays attention to personal care and body protection. Since the product clicked by the user at the most recent moment is outdoor sunscreen, it can be inferred that the user's shopping interest at the next moment may continue to the field of personal care and body protection. The user may be interested in other products related to personal care and body protection, such as skin care products, sunscreen spray, sun hats, sun protection clothing, etc. In addition, since he previously clicked on badminton rackets and sweatpants, he may also be interested in other sports and sports equipment, such as other sports equipment, sports shoes, sports accessories, etc."

[0096] Based on the interest description information in the above two examples, the interest description information contains vocabulary such as product names of "fitness equipment, sportswear, outdoor equipment, sunscreen, skin care products, protective gloves, knee pads, elbow pads, basketball, football, tennis" and vocabulary of product categories such as "sports, outdoor activities, personal care". It can be seen that the interest description information generated based on the large language module can represent the object's preferences. The interest description information already contains rich object interest characteristics and potential resource-related information. Therefore, the interest description information can represent the object's interests.

[0097] In one implementation, step 305 includes: extracting features of the model indication information and the prompt information through a large language model to obtain features of the model indication information and features of the prompt information, and generating interest description information based on the features of the model indication information and features of the prompt information.

[0098] In an embodiment of the present application, the large language model adopts a feature extraction method to extract the features of the model indication information and the prompt information respectively, so as to generate accurate interest description information by utilizing the features of the model indication information and the features of the prompt information.

[0099] In some embodiments, the process of generating interest description information includes: splicing features of the model indication information and features of the prompt information, and generating interest description information based on the spliced ​​features.

[0100] In an embodiment of the present application, the features of the model indication information and the features of the prompt information are spliced ​​together to ensure that the features of the model indication information and the features of the prompt information can be fully integrated, so that the large language model can know how to process the prompt information based on the spliced ​​features, and then generate interest description information to ensure the accuracy of the interest description information.

[0101] 306. The computer device determines the similarity between the information of each resource and the interest description information.

[0102] In an embodiment of the present application, the similarity between the resource information and the interest description information indicates that the resource has a higher correlation with the object, and can also reflect the degree of interest of the object in the resource. The greater the similarity, the more interested the object is in the resource, and the smaller the similarity, the less interested the object is in the resource.

[0103] In one possible implementation, step 306 includes: using a large language model, extracting features from the information and interest description information of each resource to obtain the features and interest features of each resource; and determining the similarity between the information and interest description information of each resource based on the features and interest features of each resource.

[0104] In the embodiment of the present application, since the large language model itself contains rich knowledge information and has generalization and reasoning capabilities, the characteristics of resource information and interest description information are extracted through the large language model, which can ensure the quality of the extracted features, thereby ensuring the accuracy of the extracted features and further ensuring the accuracy of the determined similarity.

[0105] The interest feature is a feature corresponding to the interest description information. The similarity between the resource information and the interest description information is determined based on the feature, and can be determined by using cosine similarity or other methods.

[0106] For example, through a large language model, the information of each resource is encoded to obtain the characteristics of each resource. The characteristics are used to represent the semantic representation vector. The characteristics of each resource are represented as a vector e i , vector e i ∈R 1×d , where R is used to represent the feature space, i is used to represent the sequence number of the resource, and d is used to represent the vector e i The dimension of each resource is , that is, the feature of each resource is a 1×d vector.

[0107] In some embodiments, the similarity is determined based on the characteristics and interest characteristics of each resource, satisfying the following relationship:

[0108] in, Represents the semantic representation vector e of the i-th resourcei The transpose of p f Used to represent interest features, it can be a semantic representation vector, r i Used to represent the similarity between the information of the i-th resource and the interest description information.

[0109] It should be noted that the embodiment of the present application obtains the characteristics of the resource in the process of determining the similarity between the resource information and the interest description information. In another embodiment, before step 301, the information of each resource can be feature extracted through a large language model to obtain the characteristics of each resource.

[0110] In one implementation, after obtaining the characteristics of each resource, the characteristics of multiple resources are stored in a feature library. In an embodiment of the present application, by pre-generating the characteristics of multiple resources and storing the characteristics of multiple resources in the feature library, when subsequently recommending resources to an object, the characteristics of the resources in the feature library can be used, combined with the interest description information, to filter out the resources of interest from the multiple resources, and then recommend the resources of interest to the object.

[0111] The feature library is a database of any form. For example, the feature library is Faiss (an open source, high-performance similarity search library). Faiss has the functions of large-scale vector retrieval and clustering, and is used to solve the similarity search problem of high-dimensional vectors. It can be applied in various fields, such as images, audio, text and other fields.

[0112] In some embodiments, the resource characteristics and resource identifiers are stored in a feature library in a corresponding manner. In the embodiments of the present application, the resource characteristics and resource identifiers are stored in a feature library in a corresponding manner, that is, an index between the resource characteristics and resource identifiers is established. Subsequently, based on the index in the feature library, the resource represented by the characteristic of each resource can be determined.

[0113] In some embodiments, when features of multiple resources are stored in a feature library, features of multiple resources are extracted from a special library, combined with interest features, to determine the similarity between the information of each resource and the interest information, so that resources can be subsequently filtered based on the similarity.

[0114] 307. The computer device selects a first resource from multiple resources based on the similarity between the information of each resource and the interest description information, the similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information, and the second resource is a resource other than the first resource among the multiple resources.

[0115] In an embodiment of the present application, based on the similarity between the information of each resource and the interest description information, a resource with sufficiently large similarity can be selected from multiple resources as a first resource, and the first resource is equivalent to the resource of interest to the object.

[0116] In one implementation, step 307 includes: sorting the multiple resources based on the similarity between the information of each resource and the interest description information, and selecting the first n resources from the sorted multiple resources as the first resources.

[0117] Where n is a positive integer. Multiple resources are sorted in descending order of similarity. In the embodiment of the present application, the first resource is sorted in descending order of similarity to ensure that the similarity corresponding to the first resource selected is greater than the similarity corresponding to the unselected resources, thereby ensuring that the first resource selected is the resource of interest to the object and ensuring the accuracy of the selected resources.

[0118] 308. When the similarity between the information of the third resource and the interest description information is greater than a similarity threshold, the computer device determines the third resource as the first resource, where the third resource is a resource among the multiple resources whose number of recommendations is less than a number threshold.

[0119] In an embodiment of the present application, for a resource whose number of recommendations among multiple resources is less than a threshold number, the resource is a new resource, that is, the third resource is a new resource among the multiple resources. When the similarity between the information of the new resource and the interest description information is large enough, it means that the new resource meets the screening conditions, and also means that the user is interested in the new resource. Among them, the similarity threshold is an arbitrary value, and the similarity threshold is set by the developer in the recommendation system. In an embodiment of the present application, when the similarity between the information of the new resource and the interest description information is large enough, the new resource is regarded as a resource of interest to achieve a cold start of the new resource to ensure that the new resource can be accurately recommended later, thereby ensuring the accuracy of the recommendation.

[0120] In an embodiment of the present application, the above method is only used to screen resources whose recommendation times are less than the number threshold. Once the recommendation times of a resource reaches the number threshold, the resource is screened according to the above step 307 and no longer screened according to the above step 308.

[0121] In one implementation, only the second number of resources is filtered out from multiple resources each time. Therefore, the screening process includes: when the similarity between the information of the third resource and the interest description information is greater than the similarity threshold, the third resource is determined as the first resource, and the third number is determined, the third number is the number of the currently determined first resource, and the difference between the second number and the third number is determined. According to the above step 307, based on the similarity between the information of each resource and the interest description information, the first resource with the difference is filtered out from the remaining resources.

[0122] The second number is an arbitrary number, and the third number is an arbitrary number.

[0123] In an embodiment of the present application, when only a certain number (for example, the second number) of resources are screened each time, as long as any new resource meets the screening conditions, the new resource that meets the screening conditions will be used as the screened first resource (for example, the third number), and according to the similarity, the resource with the greatest similarity is screened out from the remaining resources as the first resource, ensuring that the first number (the difference between the second number and the third number) of first resources are screened out. In this way, not only is it ensured that the number of screened first resources is equal to the first number, and the screened first resources are ensured to be the resources of interest to the object, but it can also be ensured that the screened first resources contain new resources. When the first resource is recommended to the object subsequently, a cold start of the new resource can be achieved.

[0124] 309. The computer device recommends a first resource to the object.

[0125] In the embodiment of the present application, the screened first resources are all resources that the object is interested in. Therefore, the first resources are recommended to the object to ensure the accuracy of the resource recommendation.

[0126] In an embodiment of the present application, the similarity between the information of the first resource recommended to the object and the interest description information is large enough, which not only ensures that the resources recommended to the object are in line with the object's interests, but also, for new resources, through the method provided in an embodiment of the present application, when the similarity between the information of the new resource and the interest description information is large enough, the new resource can be used as a resource of interest to the object, so as to accurately recommend the new resource, thereby avoiding the problem of being unable to make accurate recommendations due to lack of sufficient historical data and user feedback, realizing the cold start of the new resource, and ensuring the accuracy of the recommendation.

[0127] In an embodiment of the present application, since the historical behavior information of the object indicates the resources on which the object has performed interactive behaviors, it can reflect the object's interests, and the large language model itself has generalization and reasoning capabilities. Therefore, based on the model indication information, the large language model is instructed to combine the historical behavior information to generate a natural language description for describing the object's preferences, which is used as the object's interest representation. Combined with the information of the resources, the resources that the object is interested in are screened out, thereby realizing a resource screening method based on the large language model. This screening method can be applied to a variety of scenarios, ensuring that the screened resources match the object's interests, and ensuring the accuracy of the screened resources.

[0128] Furthermore, the similarity between the information of the first resource recommended to the subject and the interest description information is sufficiently large to ensure that the resource recommended to the subject is consistent with the subject's interests, thereby improving the accuracy of the recommendation. Furthermore, for new resources, the method provided in the embodiments of this application allows for accurate recommendation of new resources, avoiding the problem of inability to make accurate recommendations due to a lack of sufficient historical data and user feedback, achieving a cold start for new resources, and ensuring the accuracy of recommendations.

[0129] The solution provided in the embodiments of the present application can be applied to a variety of small and medium-sized scenarios. Due to the small amount of historical behavior data of objects in small and medium-sized scenarios, it is impossible to use the historical behavior data of the objects to train a recommendation model for small and medium-sized scenarios. However, the large language model itself contains rich knowledge information and has generalization and reasoning capabilities. Therefore, a universal recommendation system is built based on the large language model. The generalization ability and rich semantic knowledge information of the large language model on small data sets are utilized to reduce the dependence on the interactive information of the behavior sequence, so that it can achieve resource cold start in small and medium-sized scenarios with few resources and sparse historical behavior data of objects, effectively improving the accuracy of resource recommendation in cold start scenarios.

[0130] Based on the embodiment shown in FIG. 3 , the large language model is used to pre-acquire the features of each resource. The features of multiple resources and their corresponding resource identifiers are pre-stored in a feature library, thereby pre-establishing an index between the resource features and resource identifiers. If a new resource is added to the recommendation system, feature extraction is performed on the new resource information based on the large language model in accordance with step 306 . The features of the new resource and its resource identifier are then stored in the feature library in a corresponding manner to update the index. When recommending resources for an object, first generate interest description information according to the above steps 301 to 305; then, according to the above step 306, extract features of the interest description information based on the large language model to obtain interest features; then, according to the above step 307, determine the similarity between the features of the resources in the feature library and the interest features, filter out the features of the top n resources with the greatest similarity, determine the resource identifiers corresponding to the filtered features based on the index in the feature library, and determine the resource indicated by the filtered resource identifier as the first resource; for new resources, according to the above step 308, filter out the first resource from the new resources, and then, according to the above step 309, recommend the first resource to the object to achieve a cold start of the new resource. In addition, once the number of recommendations for a new resource reaches the number threshold, the new resource will be filtered according to the above step 307 and will no longer be filtered according to the above step 308.

[0131] It should be noted that the embodiment shown in FIG3 utilizes a similarity threshold to select the first resource from the second resource. However, in another embodiment, steps 306-308 are not required. Instead, other methods are employed to select the first resource from the multiple resources based on the interest description information and information about the multiple resources. In one implementation, the process of selecting the first resource includes steps 1-3 below.

[0132] Step 1: Determine the similarity between the information of each resource and the interest description information.

[0133] Step 2: Based on the amplification coefficient, amplify the similarity corresponding to the third resource, where the third resource is a resource whose number of recommendations is less than a number threshold among the multiple resources.

[0134] The amplification factor is used to amplify the similarity of the third resource, and the amplification factor is an arbitrary factor, for example, the amplification factor is a value greater than 1. The similarity corresponding to the third resource is the similarity between the information of the third resource and the interest description information.

[0135] In one implementation, the process of amplifying the similarity corresponding to the third resource includes: multiplying the similarity corresponding to the third resource by the amplification coefficient as the amplified similarity of the third resource.

[0136] Step 3. Based on the amplified similarity of the third resource and the corresponding similarity of the fourth resource, the first resource is selected from multiple resources. The fourth resource is a resource other than the third resource among the multiple resources. The similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information. The second resource is a resource other than the first resource among the multiple resources.

[0137] In an embodiment of the present application, when determining the similarity between the information of each resource and the interest description information, the similarity corresponding to the new resource is amplified, and then based on the amplified similarity of the new resource, that is, the similarity corresponding to other resources, the resources of interest to the object are screened out from multiple resources according to the size of the similarity, thereby realizing a cold start method for resources. On the basis of recommending resources according to the degree of user interest, the similarity of the new resource is preferentially amplified to realize the cold start of the resource as much as possible, while also ensuring that the screened resources are the resources of interest to the object, so as to ensure the accuracy of subsequent recommendations.

[0138] FIG4 is a schematic diagram of the structure of a resource screening device provided in an embodiment of the present application. As shown in FIG4 , the device includes:

[0139] Acquisition module 401 is used to acquire historical behavior information and model indication information of an object, where the historical behavior information indicates that the object has performed interactive behaviors on resources, and the model indication information is used to instruct the large language model to generate interest description information based on input information, where the interest description information is used to describe the resources that the object is interested in;

[0140] A generating module 402 is used to generate prompt information based on the historical behavior information, where the prompt information is used to indicate the resources on which the object has performed interactive behaviors;

[0141] The generating module 402 is further configured to generate the interest description information based on the model indication information and the prompt information through the large language model;

[0142] The screening module 403 is used to screen out a first resource from multiple resources based on the interest description information and the information of multiple resources. The similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information. The second resource is a resource other than the first resource in the multiple resources.

[0143] In one possible implementation, the screening module 403 is used to determine the similarity between the information of each resource and the interest description information; based on the amplification coefficient, the similarity corresponding to the third resource is amplified, and the third resource is a resource among multiple resources whose number of recommendations is less than the number threshold; based on the amplified similarity of the third resource and the similarity corresponding to the fourth resource, the first resource is screened out from the multiple resources, and the fourth resource is the resource among the multiple resources other than the third resource.

[0144] In another possible implementation, the screening module 403 is used to extract features from the information and interest description information of each resource using a large language model to obtain features and interest features of each resource; and based on the features and interest features of each resource, determine the similarity between the information and interest description information of each resource.

[0145] In another possible implementation, as shown in FIG5 , the apparatus further includes:

[0146] The determination module 404 is configured to determine the third resource as the first resource when the similarity between the information of the third resource and the interest description information is greater than a similarity threshold, wherein the third resource is a resource with a recommendation count less than a count threshold among the multiple resources.

[0147] In another possible implementation, the model indication information further indicates the format of the interest description information; the generating module 402 is configured to generate the interest description information in the format based on the model indication information and the prompt information through the large language model.

[0148] In another possible implementation, generation module 402 is used to determine first behavior information from multiple pieces of historical behavior information based on the generation time of multiple pieces of historical behavior information of the object, where the generation time of the first behavior information is later than the generation time of the second behavior information, and the second behavior information is the historical behavior information in the multiple pieces of historical behavior information other than the first behavior information; extract resources on which the object has performed interactive behaviors from the first behavior information; and generate prompt information based on the extracted resources.

[0149] In another possible implementation, the generation module 402 is used to obtain a prompt information template, which includes a resource filling position and relationship description information, and the relationship description information is used to describe the relationship between the resource and the object in the resource filling position; the extracted resource is added to the resource filling position to obtain prompt information.

[0150] In another possible implementation, the interest description information is used to describe the resource type that the object is interested in and the resources that the object is interested in under the resource type; or, the interest description information is used to describe the resources and the types of resources that the object may perform interactive behaviors on next time.

[0151] It should be noted that the resource screening device provided in the above embodiment is merely an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the resource screening device provided in the above embodiment and the resource screening method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0152] An embodiment of the present application further provides a computer device, which includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the resource screening method of the above embodiment.

[0153] In some embodiments, the computer device is provided as a terminal. FIG6 shows a block diagram of a terminal 600 provided in an exemplary embodiment of the present application. The terminal 600 includes: a processor 601 and a memory 602.

[0154] The processor 601 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 601 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 601 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 601 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 601 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0155] The memory 602 may include one or more computer-readable storage media, which may be non-transitory. The memory 602 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 602 is used to store at least one computer program, which is executed by the processor 601 to implement the resource screening method provided in the method embodiment of the present application.

[0156] In some embodiments, terminal 600 may optionally include a peripheral device interface 603 and at least one peripheral device. Processor 601, memory 602, and peripheral device interface 603 may be connected via a bus or signal lines. Each peripheral device may be connected to peripheral device interface 603 via a bus, signal lines, or circuit boards. Specifically, the peripheral device may include at least one of a radio frequency circuit 604, a display screen 605, a camera assembly 606, an audio circuit 607, and a power supply 608.

[0157] Those skilled in the art will appreciate that the structure shown in FIG6 does not limit the terminal 600 and may include more or fewer components than shown, or combine certain components, or adopt a different component arrangement.

[0158] In some embodiments, the computer device is provided as a server. Figure 7 is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server 700 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 701 and one or more memories 702, wherein at least one computer program is stored in the memory 702, and at least one computer program is loaded and executed by the processor 701 to implement the methods provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The server may also include other components for implementing device functions, which will not be described in detail here.

[0159] An embodiment of the present application further provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the operations performed by the resource screening method of the above embodiment.

[0160] An embodiment of the present application further provides a computer program product, including a computer program, which implements the operations performed by the resource screening method of the above embodiment when the computer program is executed by a processor.

[0161] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0162] The above description is only an optional embodiment of the embodiment of the present application and is not intended to limit the embodiment of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the embodiment of the present application should be included in the scope of protection of the present application.

Claims

1. A resource screening method, executed by a computer device, comprising: Acquire historical behavior information and model indication information of the object, wherein the historical behavior information indicates that the object has performed interactive behaviors on resources, and the model indication information is used to instruct the large language model to generate interest description information based on input information, and the interest description information is used to describe the resources that the object is interested in; Based on the historical behavior information, generating prompt information, the prompt information is used to indicate the resources on which the object has performed interactive behaviors; Generate the interest description information through the large language model based on the model indication information and the prompt information; Based on the interest description information and information of multiple resources, a first resource is filtered out from the multiple resources, the similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information, and the second resource is a resource among the multiple resources other than the first resource.

2. The method according to claim 1, wherein: The step of selecting a first resource from the multiple resources based on the interest description information and the information of the multiple resources includes: Determining the similarity between the information of each resource and the interest description information; Based on the amplification coefficient, amplify the similarity corresponding to the third resource, wherein the third resource is a resource among the multiple resources whose recommendation times are less than the times threshold; Based on the amplified similarity of the third resource and the similarity corresponding to the fourth resource, the first resource is screened out from the multiple resources, and the fourth resource is a resource among the multiple resources except the third resource.

3. The method according to claim 1 or 2, wherein: The determining the similarity between the information of each resource and the interest description information includes: By using the large language model, feature extraction is performed on the information of each resource and the interest description information to obtain features and interest features of each resource; Based on the characteristics of each resource and the interest characteristics, the similarity between the information of each resource and the interest description information is determined.

4. The method according to any one of claims 1 to 3, wherein: After selecting the first resource from the multiple resources based on the interest description information and the information of the multiple resources, the method further includes: When the similarity between the information of the third resource and the interest description information is greater than a similarity threshold, the third resource is determined as the first resource, and the third resource is a resource among the multiple resources whose recommendation times are less than a times threshold.

5. The method according to any one of claims 1 to 4, wherein: The model indication information further indicates a format of the interest description information; and generating the interest description information by using the large language model based on the model indication information and the prompt information includes: The interest description information in the format is generated through the large language model based on the model indication information and the prompt information.

6. The method according to any one of claims 1 to 5, wherein: The generating prompt information based on the historical behavior information includes: Based on the generation time of the multiple pieces of historical behavior information of the object, determine first behavior information from the multiple pieces of historical behavior information, where the generation time of the first behavior information is later than the generation time of the second behavior information, and the second behavior information is the historical behavior information other than the first behavior information in the multiple pieces of historical behavior information; Extracting, from the first behavior information, resources on which the object has performed interactive behaviors; The prompt information is generated based on the extracted resources.

7. The method according to any one of claims 1 to 6, wherein: The generating the prompt information based on the extracted resources includes: Acquire a prompt information template, the prompt information template including a resource filling position and relationship description information, the relationship description information being used to describe the relationship between the resource in the resource filling position and the object; The extracted resources are added to the resource filling position to obtain the prompt information.

8. The method according to any one of claims 1 to 7, wherein: The interest description information is used to describe the resource type that the object is interested in and the resources that the object is interested in under the resource type; or, the interest description information is used to describe the resources that the object may perform interactive behavior on next time and the type of the resources.

9. A resource screening device, comprising: The acquisition module is used to acquire the historical behavior information and model indication information of the object, wherein the historical behavior information indicates the The resource has performed an interactive behavior, and the model indication information is used to instruct the large language model to generate interest description information based on the input information, and the interest description information is used to describe the resource that the object is interested in; A generating module, used to generate prompt information based on the historical behavior information, wherein the prompt information is used to indicate the resources on which the object has performed interactive behaviors; The generating module is further used to generate interest description information based on the model indication information and the prompt information through the large language model; A screening module is used to screen out a first resource from the multiple resources based on the interest description information and the information of the multiple resources, wherein the similarity between the information of the first resource and the interest description information is greater than the similarity between the information of the second resource and the interest description information, and the second resource is a resource among the multiple resources other than the first resource.

10. The device according to claim 9, wherein: The screening module is used to determine the similarity between the information of each resource and the interest description information; based on the amplification coefficient, amplify the similarity corresponding to the third resource, the third resource being a resource with a recommendation number less than a number threshold among the multiple resources; Based on the amplified similarity of the third resource and the similarity corresponding to the fourth resource, the first resource is screened out from the multiple resources, and the fourth resource is a resource among the multiple resources except the third resource.

11. The device according to claim 9 or 10, wherein: The screening module is specifically used to extract features of the information of each resource and the interest description information respectively through the large language model to obtain features and interest features of each resource; Based on the characteristics of each resource and the interest characteristics, the similarity between the information of each resource and the interest description information is determined.

12. The device according to any one of claims 9 to 11, wherein: The device also includes a determination module, which is used to determine the third resource as the first resource when the similarity between the information of the third resource and the interest description information is greater than a similarity threshold, and the third resource is a resource among the multiple resources whose number of recommendations is less than a number threshold.

13. The device according to any one of claims 9 to 12, wherein: The model indication information also indicates the format of the interest description information; the generating module is used to generate the interest description information in the format based on the large language model, the model indication information and the prompt information.

14. The device according to any one of claims 9 to 13, wherein: The generation module is used to determine first behavior information from multiple pieces of historical behavior information of the object based on the generation time of the multiple pieces of historical behavior information, the generation time of the first behavior information is later than the generation time of the second behavior information, and the second behavior information is the historical behavior information in the multiple pieces of historical behavior information other than the first behavior information; extract the resources on which the object has performed interactive behaviors from the first behavior information; and generate the prompt information based on the extracted resources.

15. The device according to any one of claims 9 to 14, wherein: The generation module is used to obtain a prompt information template, which includes a resource filling position and relationship description information, and the relationship description information is used to describe the relationship between the resource in the resource filling position and the object; the extracted resource is added to the resource filling position to obtain the prompt information.

16. The device according to any one of claims 9 to 15, wherein: The interest description information is used to describe the resource type that the object is interested in and the resources that the object is interested in under the resource type; or, the interest description information is used to describe the resources that the object may perform interactive behavior on next time and the type of the resources.

17. A computer device, comprising a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the operations performed by the resource screening method according to any one of claims 1 to 8.

18. A computer-readable storage medium, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the operations performed by the resource screening method according to any one of claims 1 to 8.

19. A computer program product, comprising computer instructions, wherein the computer instructions are stored in a computer-readable storage medium, and a processor obtains the computer instructions from the computer-readable storage medium, so that the processor loads and executes the computer instructions to implement the operations performed by the resource screening method as described in any one of claims 1 to 8.

Images