Resource processing method and device, electronic equipment and storage medium

Abstract

The invention discloses a resource processing method and device, electronic equipment and a storage medium. The embodiment of the invention relates to the technical field of artificial intelligence and the like. The method comprises the steps of obtaining a plurality of candidate resources and obtaining object features of a target object under each interest description granularity in N interest description granularities; for each candidate resource, according to a sequence from small interest description granularity to large interest description granularity, performing focusing fusion on the resource features of the candidate resources and the object features of the target object under the N interest description granularities in sequence to obtain object fusion features of the target object relative to the candidate resources; according to the object fusion feature of the target object relative to each candidate resource, determining the interestingness of the target object to each candidate resource; and determining a target resource pushed to the target object from the plurality of candidate resources according to the interestingness of the target object to each candidate resource. According to the method provided by the invention, the accuracy of the selected target resource is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a resource processing method, apparatus, electronic device, and storage medium. Background Technology

[0002] Currently, existing resource selection systems can determine an object's interest level in each candidate resource based on its object information, and then select the target resource recommended to the object from among the candidate resources according to the interest level. However, the accuracy of the interest level determined by this method is low, resulting in the selected target resource failing to meet the object's needs. Summary of the Invention

[0003] In view of this, embodiments of this application propose a resource processing method, apparatus, electronic device, and storage medium.

[0004] In a first aspect, embodiments of this application provide a resource processing method, the method comprising: acquiring multiple candidate resources and acquiring object features of a target object at each of N interest characterization granularities; the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity, i∈[1,N], where N is an integer greater than 1; for each candidate resource, in ascending order of the number of interest characterization granularities, sequentially focusing and fusing the resource features of the candidate resource with the object features of the target object at the N interest characterization granularities to obtain object fusion features of the target object relative to the candidate resources, wherein the input of the focusing and fusing process of the resource features and the object features of the target object at the (i+1)-th interest characterization granularity includes an i-th level focusing and fusing feature, which is the result obtained by focusing and fusing the resource features and the object features of the target object at the i-th interest characterization granularity; determining the target object's interest in each of the candidate resources based on the object fusion features of the target object relative to each candidate resource; and determining the target resource to be pushed to the target object from multiple candidate resources based on the target object's interest in each candidate resource.

[0005] Secondly, embodiments of this application provide a resource processing apparatus, comprising: an acquisition module, configured to acquire multiple candidate resources and acquire object features of a target object at each of N interest characterization granularities; wherein the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity, i∈[1, N], and N is an integer greater than 1; and a fusion module, configured to, for each candidate resource, sequentially fuse the resource features of the candidate resource with the object features of the target object at the N interest characterization granularities in ascending order of the number of interest characterization granularities, to obtain the target object relative to the candidate resources. The input to the focused fusion process of object fusion features, resource features, and object features of the target object at the (i+1)th interest characterization granularity includes i-level focused fusion features, which are the results obtained by focusing fusion of resource features and object features of the target object at the i-th interest characterization granularity; a first determining module is used to determine the target object's interest in each of the candidate resources based on the object fusion features of the target object relative to each candidate resource; a second determining module is used to determine the target resource to be pushed to the target object from multiple candidate resources based on the target object's interest in each candidate resource.

[0006] Optionally, the fusion module is further configured to, for each candidate resource, obtain the (j-1)th level focused fusion feature of the target object relative to the candidate resource, where j is an integer and j∈[2, N]. The first-level focused fusion feature is obtained by focusing and fusing the resource features of the candidate resource with the object features of the target object at the first interest characterization granularity; the (j-1)th level focused fusion feature is fused with the object features of the target object at the j-th interest characterization granularity to obtain the j-th level intermediate fusion feature; the j-th level intermediate fusion feature is then focused and fused with the resource features of the candidate resource to obtain the j-th level focused fusion feature of the target object relative to the candidate resource; if j is less than N, j is incremented by 1, and the process returns to the step of obtaining the (j-1)th level focused fusion feature of the target object relative to the candidate resource; if j = N, the object fusion feature of the target object relative to the candidate resource is determined based on the N-th level focused fusion feature of the target object relative to the candidate resource.

[0007] Optionally, the fusion module is further configured to determine the attention score of the j-level intermediate fusion feature based on the j-level intermediate fusion feature and the resource features of the candidate resource; and to perform focused fusion based on the attention score of the j-level intermediate fusion feature and the j-level intermediate fusion feature to obtain the j-level focused fusion feature of the target object relative to the candidate resource.

[0008] Optionally, the fusion module is further configured to determine the j-level key vector and j-level value vector based on the j-level intermediate fusion features; use the resource features of the candidate resources as the query vector, determine the attention score of the j-level intermediate fusion features based on the query vector and the j-level key vector; and weight the j-level value vector based on the attention score of the j-level intermediate fusion features to obtain the j-level focused fusion features of the target object relative to the candidate resources.

[0009] Optionally, the fusion module is also used to concatenate the j-1 level focused fusion features with the object features of the target object at the j-th interest characterization granularity to obtain the j-level concatenated features; and to perform fully connected processing on the j-level concatenated features to obtain the j-level intermediate fusion features.

[0010] Optionally, the fusion module is also used to use the N-level focused fusion feature of the target object relative to the candidate resource as the object fusion feature of the target object relative to the candidate resource.

[0011] Optionally, the acquisition module is also used to acquire object information of the target object, which includes object attribute information under multiple attributes; based on the correspondence between attributes and interest characterization granularity, determine the interest characterization granularity to which each object attribute information belongs among N interest characterization granularities; for each interest characterization granularity, perform feature encoding on the object attribute information belonging to that interest characterization granularity to obtain the object features of the target object under the interest characterization granularity.

[0012] Optionally, the first determining module is further configured to input the object fusion features of the target object relative to the candidate resources into the resource prediction model for each candidate resource, so as to obtain the target object's interest in the candidate resources predicted by the resource prediction model.

[0013] Optionally, the second determining module is further configured to select a first number of candidate resources with the highest interest from multiple candidate resources as target resources to be pushed to the target object; or to select candidate resources with interest reaching a first threshold from multiple candidate resources as target resources to be pushed to the target object.

[0014] Optionally, the acquisition module is also used to determine the matching degree between each resource and the target object based on the object information of the target object and the basic resource information of each resource in the resource set; and to select multiple candidate resources for the target object from the resource set based on the matching degree.

[0015] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory; the memory stores computer-readable instructions, which, when executed by the processor, implement the above-described method.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the above-described method.

[0017] Fifthly, embodiments of this application provide a computer program product, including computer instructions, which, when executed by a processor, implement the method described above.

[0018] This application provides a resource processing method, apparatus, electronic device, and storage medium. In this application, resource features of candidate resources are sequentially fused with object features of the target object at N interest characterization granularities, in ascending order of interest characterization granularity. This results in object fusion features of the target object relative to the candidate resources. Furthermore, the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity. The result of the fusion of resource features and target object features at the i-th interest characterization granularity is also used as input for the fusion process of resource features and target object features at the (i+1)-th interest characterization granularity. This achieves dynamic perception of candidate resource features during multiple fusion processes, and the object fusion features of the target object relative to the candidate resources incorporate the dynamic perception of the candidate resources during these multiple fusion processes. Furthermore, by sequentially fusing the resource features of candidate resources with the object features of the target object at N interest characterization granularities in ascending order of interest characterization granularity, the progressive relationship between object features at different interest characterization granularities can be fully utilized. This ensures that the object fusion features of the target object relative to the candidate resources can accurately reflect the target object's interest preferences relative to the candidate resources. Subsequently, the accuracy of determining the target object's interest in each candidate resource based on the object fusion features determined by the focused fusion features is high. The interest can more accurately reflect the target object's demand for each candidate resource, ensuring that the target resources pushed to the target object can meet the target object's needs without requiring the target object to communicate with the server multiple times to obtain the required resources, thus improving the utilization rate of communication resources. Attached Figure Description

[0019] 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.

[0020] Figure 1 A schematic diagram is shown illustrating the application scenarios applicable to the embodiments of this application;

[0021] Figure 2A flowchart of a resource processing method according to an embodiment of this application is shown;

[0022] Figure 3 This illustration shows a schematic diagram of object attribute information at different interest characterization granularities in an embodiment of this application;

[0023] Figure 4 A flowchart of a resource processing method according to yet another embodiment of this application is shown;

[0024] Figure 5 A schematic diagram of a resource processing procedure in an embodiment of this application is shown;

[0025] Figure 6 A schematic diagram illustrating the process of determining an object fusion feature according to an embodiment of this application is shown;

[0026] Figure 7 A block diagram of a resource processing apparatus according to one embodiment of this application is shown;

[0027] Figure 8 A structural block diagram of an electronic device for performing a resource processing method according to an embodiment of this application is shown. Detailed Implementation

[0028] 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. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0029] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit the application. It should be noted that "multiple" as used herein refers to two or more. "And / or" describes the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0031] It should be noted that, in the embodiments of this application, the acquisition of user data such as basic attribute information under basic attributes, portrait attribute information under portrait attributes, interaction resource information under interaction attributes, and object identification information under identification attributes all require user permission or consent, and the collection, use, processing, and storage of user data such as basic attribute information under basic attributes, portrait attribute information under portrait attributes, interaction resource information under interaction attributes, and object identification information under identification attributes all need to comply with the regulations of the region.

[0032] The following is an explanation of the terms used in this application:

[0033] A thought chain refers to the process of breaking down a complex problem into subproblems by having a large model participate step by step and solving them sequentially. This can significantly improve the performance of the large model, and the intermediate steps in this series of reasoning are called thought chains (cot).

[0034] This application discloses a resource processing method, apparatus, electronic device, and storage medium, which relate to artificial intelligence technology.

[0035] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0036] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0037] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0038] Pre-trained models, also known as foundational models or large models, refer to deep neural networks (DNNs) with a large number of parameters. These DNNs are trained on massive amounts of unlabeled data. Leveraging the function approximation capabilities of large-parameter DNNs, Pre-trained Matrix Transformers (PTMs) extract common features from the data. Through fine-tuning, parameter efficient fine-tuning (PEFT), prompt-tuning, and other techniques, they are suitable for downstream tasks. Therefore, pre-trained models can achieve ideal results in scenarios with few or zero samples.

[0039] In this application, the object fusion features of the target object relative to the candidate resources can be processed by a resource prediction model based on artificial intelligence technology to obtain the interest degree of the target object relative to each candidate resource predicted by the resource prediction model, and then the target resource can be determined based on the interest degree.

[0040] See Figure 1 , Figure 1 The diagram illustrates an application scenario applicable to the embodiments of this application. The terminal device 400 connects to the server 200 via the network 300, and the server 200 connects to the database 500. The network 300 can be a wide area network or a local area network, or a combination of both.

[0041] In some embodiments, taking the electronic device as a server as an example, the resource processing provided in this application embodiment can be implemented by the server. For example, the server 200 can obtain candidate resources and object features of the target object at each of the N interest characterization granularities from the database 500. Then, the server 200 determines the target resource from the candidate resources based on the resource features of the candidate resources and the object features of the target object at each of the N interest characterization granularities. After determining the target resource, the server 200 can send the target resource to the terminal device 400 so that the terminal device 400 can display the target resource.

[0042] In some embodiments, taking the electronic device as a terminal device as an example, the resource processing provided in this application embodiment can be implemented by the terminal device. For example, the terminal device 400 can obtain candidate resources and object features of the target object at each of the N interest characterization granularities from the database 500 through the server 200. Then, the terminal device 400 determines the target resource from the candidate resources based on the resource features of the candidate resources and the object features of the target object at each of the N interest characterization granularities. After determining the target resource, the terminal device can directly display the target resource.

[0043] In some embodiments, the terminal device 400 may initiate a resource acquisition request to the server 200. Then, the server 200 obtains candidate resources and object features of the target object at each of the N interest characterization granularities from the database 500 according to the resource acquisition request. Then, based on the resource features of the candidate resources and the object features of the target object at each of the N interest characterization granularities, the server 200 determines the target resource from the multiple candidate resources and sends the determined target resource to the terminal device 400.

[0044] In some other embodiments, the terminal device 400 may initiate a data acquisition request to the database 500. Then, the database 500 returns the candidate resources and the object features of the target object at each of the N interest characterization granularities according to the data acquisition request. The terminal device 400 determines the target resource from the multiple candidate resources based on the resource features of the candidate resources and the object features of the target object at each of the N interest characterization granularities.

[0045] In some embodiments, the terminal device 400 or server 200 can implement the resource processing method provided in this application embodiment by running a computer program. For example, the computer program can be a native program or software module in an operating system; it can be a native application (APP), i.e., a program that needs to be installed in the operating system to run; it can also be a small program, i.e., a program that only needs to be downloaded to a browser environment to run; or it can be a small program that can be embedded in any APP, and the small program can be controlled by the user to run or close. In short, the above-mentioned computer program can be any form of application, module or plugin.

[0046] In some embodiments, server 200 may be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. Terminal device 400 may be a smartphone, tablet, laptop, desktop computer, smart voice interaction device, smart home appliance, vehicle terminal, aircraft, smart TV, etc., but is not limited to these. Terminal devices and servers can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment.

[0047] For ease of description, the following embodiments are illustrated using the example of a resource processing method being executed by an electronic device.

[0048] Please see Figure 2 , Figure 2 This application illustrates a flowchart of a resource processing method according to an embodiment of the present application. The method can be used in an electronic device, which may be... Figure 1 The terminal device 400 or server 200 in the middle, the method may include:

[0049] S110. Obtain multiple candidate resources and obtain the object features of the target object at each of the N interest characterization granularities.

[0050] Where the granularity of the i-th interest characterization is greater than the granularity of the (i+1)-th interest characterization, i∈[1,N], and N is an integer greater than 1.

[0051] In this application, the target object can be the object to be recommended resources, or the object can be the user of the application. Specifically, the object can be referred to by the account registered in the application. The application can be an instant messaging application, a music playback application, a news application, a lifestyle service application, a video playback application, etc., without specific limitations.

[0052] The target object can have object information, which refers to information used to describe the object's attributes. For example, object information can include the target object's object attribute information under multiple attributes.

[0053] The attributes of a target object can correspond to the granularity of interest characterization. The granularity of interest characterization of an attribute refers to the fineness with which the attribute characterizes the group characteristics of the object, or the fineness with which the attribute is used to represent the group characteristics of the object. The finer the granularity of interest characterization of an attribute, the more refined the granularity of interest characterization of the attribute; conversely, the broader the granularity of interest characterization of an attribute, the more refined the granularity of interest characterization of the attribute.

[0054] In this application, object information of the target object can be obtained; then, based on the correspondence between attributes and interest characterization granularity, the interest characterization granularity to which each object attribute information belongs is determined among N interest characterization granularities; for each interest characterization granularity, the object attribute information belonging to that interest characterization granularity is feature-encoded to obtain the object features of the target object under that interest characterization granularity.

[0055] In other words, the interest characterization granularity to which an object's attribute information belongs is the interest characterization granularity corresponding to the attribute to which that object's attribute information belongs. Specifically, the interest characterization granularity corresponding to each attribute of an object can be determined based on the level of detail in which the object's attribute information characterizes the group characteristics of the object (or the group characteristics of the object).

[0056] For example, the attributes of the target object include attribute a1, attribute a2, attribute a11, and attribute a21. The interest characterization granularity includes a first interest characterization granularity and a second interest characterization granularity, with the first interest characterization granularity being larger than the second interest characterization granularity. The object attribute information under attribute a1 and the object attribute information under attribute a2 have the same level of detail in characterizing the group characteristics of the target object. The object attribute information under attribute a11 and the object attribute information under attribute a21 have the same level of detail in characterizing the group characteristics of the target object. Furthermore, the level of detail in characterizing the group characteristics of the target object by the object attribute information under attribute a1 is higher than the level of detail in characterizing the group characteristics of the target object by the object attribute information under attribute a11. In this case, it is determined that attributes a1 and a2 correspond to the second interest characterization granularity, and attributes a11 and a21 correspond to the first interest characterization granularity.

[0057] After determining the corresponding interest characterization granularity for the object attribute information under all attributes of the target object, the target object has at least one attribute under each interest characterization granularity. Feature encoding is performed on the object attribute information of each attribute under the interest characterization granularity to obtain the object features of the target object under each interest characterization granularity. Here, feature encoding of object attribute information can refer to encoding the object attribute information into a vector format, that is, the obtained object features are in the format of feature vectors.

[0058] As mentioned above, the granularity of the i-th interest characterization is greater than that of the (i+1)-th interest characterization. This means that the level of detail in which attributes at the i-th interest characterization granularity represent the group characteristics of the target object (or characterize the group characteristics of the object) is lower than that at the i-th interest characterization granularity. The object features at the i-th interest characterization granularity are determined based on the object attribute information of the attributes at the i-th interest characterization granularity, and the object features at the (i+1)-th interest characterization granularity are also determined based on the object attribute information of the attributes at the i-th interest characterization granularity. Therefore, in the i-th interest... The level of detail in the group characteristics of the target object represented by the object features at the granularity of the characterization is lower than that at the (i+1)th level of interest characterization. In other words, the larger the interest characterization granularity corresponding to the object feature, the lower the level of detail in the group characteristics of the target object represented by the object feature. Conversely, the smaller the interest characterization granularity corresponding to the object feature, the higher the level of detail in the group characteristics of the target object represented by the object feature.

[0059] In some implementations, the object attribute information in this application may include at least two of the following: basic attribute information under basic attributes, image attribute information under image attributes, interactive resource information under interactive attributes, and object identification information under identification attributes (object identification information is used to indicate an identifier used to uniquely identify the target object, which may be a combination of numbers, letters, and symbols). The interactive resource information includes resource information of multiple resources that trigger interactive operations by the target object; resources may be e-books, e-papers, e-reports, articles, pictures, commodities, news, videos, audio, advertisements, etc. E-books may include novels, textbooks, essays, etc. Resource information may refer to the content of the resource, the source of the resource, the category to which the resource belongs, and the name of the resource, etc.

[0060] Specifically, the granularity of interest characterization corresponding to basic attributes is smaller than that corresponding to profile attributes; the granularity of interest characterization corresponding to profile attributes is smaller than that corresponding to interaction attributes; and the granularity of interest characterization corresponding to interaction attributes is smaller than that corresponding to identifier attributes. Triggered interactive operations can include clicks, favorites, purchases, browsing, etc.

[0061] Basic attributes refer to the attributes of an object that describe the object's basic information. The object attribute information under basic attributes is basic attribute information, which may include, for example, the user's gender, age, and location. Profile attributes refer to the tags used to reflect the object's consumption habits, social attributes, resource preferences, etc. The object attribute information under profile attributes is profile attribute information, which may include, for example, the user's occupational preferences, consumption habits, preferred products, and preferred content (such as the category of articles and videos).

[0062] Interaction attributes refer to the attributes used to describe the resources that trigger interactive operations on an object. The object attribute information under the interaction attribute is the interaction resource information. Interaction resource information may include, for example, the product identifier of the product clicked by the user, the product purchased, or the product favorited; the identifier of the article (or video, music, etc.) that triggers the user's interactive operation (interactive operation such as click, play, enter the details page, favorite, like, etc.); and the basic information of the resource that triggers the interactive operation (basic information such as the category of the resource, playback duration, reading duration). Identification attributes refer to the attributes used to uniquely identify an object. The object attribute information under the identification attribute is the object identification information. Object identification information may include, for example, a unique identifier for the user.

[0063] As mentioned earlier, attributes with a larger interest characterization granularity have finer interest characterization granularity than attributes with a smaller interest characterization granularity. Based on the previous analysis of basic attributes, profile attributes, interaction attributes, and identifier attributes, we know that the basic attribute information of an object under basic attributes simply describes the social group attributes of the object, while the profile attribute information under the object profile attributes describes the object's preferences for different categories. Compared with the basic attribute information, the profile attribute information describes the characteristics of the object more finely. Therefore, the interest characterization granularity of the object features obtained by encoding the basic attribute information is greater than the interest characterization granularity of the object features extracted from the profile attribute information. Correspondingly, it can be determined that the interest characterization granularity corresponding to the basic attribute is smaller than the interest characterization granularity corresponding to the profile attribute.

[0064] Similarly, the interaction resource information under the interaction attribute describes the resource information of the resources that trigger the interaction operation of the object. Compared with the profile attribute information, it is a more refined characterization of the object's interests and preferences. Therefore, the interest characterization granularity of the object features obtained by encoding the profile attribute information is greater than the interest characterization granularity of the object features extracted from the interaction resource information. Correspondingly, it can be determined that the interest characterization granularity corresponding to the profile attribute is less than the interest characterization granularity corresponding to the interaction attribute.

[0065] Similarly, the object identification information under the identification attribute describes the unique identifier of the object. This object identification information can uniquely correspond to the object. Therefore, compared with the interaction resource information, it reflects the characteristics that completely distinguish the object from other objects. Thus, the interest characterization granularity of the object features obtained by encoding the interaction resource information is greater than the interest characterization granularity of the object features extracted from the object identification information. Correspondingly, it can be determined that the interest characterization granularity corresponding to the interaction attribute is less than the interest characterization granularity corresponding to the identification attribute.

[0066] For example, such as Figure 3 As shown, when the target object is a user using a device with a resource processing application installed, the basic attribute information can include the user's age and gender. The profile attribute information can include the user's first-level category profile, second-level category profile, and tag profile. The first-level category profile can be the user's industry, the second-level category profile can be the user's occupation, and the tag profile can be the user's occupational tag, where the occupational tag could be that the user is a workaholic. Interactive resource information can refer to the sequence of resources that the user has clicked, the category to which the clicked resources belong, and the reading time corresponding to the clicked resources. Object identification information can refer to the identifier used to uniquely identify the target object, etc.

[0067] As mentioned above, the interest characterization granularity of basic attribute information such as age and gender is coarser than that of user profile attribute information; the interest characterization granularity of user interaction resource information is finer than that of user profile attribute information; and the interest characterization granularity of user object identification information is finer than that of user interaction resource information. Therefore, it can be determined that the user's basic attribute information corresponds to the first interest characterization granularity, the user's profile attribute information corresponds to the second interest characterization granularity, the user's interaction resource information corresponds to the third interest characterization granularity, and the object identification information corresponds to the fourth interest characterization granularity.

[0068] Then, feature encoding is performed on the user's age and gender to obtain the user's object features at the first interest characterization granularity. Feature encoding is performed on the user's first-level category profile, second-level category profile, and tag profile to obtain the user's object features at the second interest characterization granularity. Feature encoding is performed on the user's interactive resource information to obtain the user's object features at the third interest characterization granularity. Feature encoding is performed on the user's unique identifier to obtain the user's object features at the fourth interest characterization granularity.

[0069] Candidate resources can refer to resources that are recommended as candidates. Candidate resources can be e-books, e-papers, e-reports, articles, pictures, products, news, videos, audio, advertisements, etc. E-books include novels, textbooks, essays, etc.

[0070] In some embodiments, a resource set can be set, which includes multiple resources, and all resources in the resource set can be used as candidate resources.

[0071] In another embodiment, the matching degree between each resource and the target object can be determined based on the object information of the target object and the basic resource information of each resource in the resource set; based on the matching degree, multiple candidate resources are selected for the target object from the resource set. The second quantity and second threshold of resources are not limited in this application and can be set based on requirements; for example, the second quantity can be 2000 and the second threshold can be 0.5.

[0072] Basic resource information can include the resource's basic attribute information, which may include the resource's category, the target audience, etc. For example, if the resource is a commodity, the basic attribute information may include the commodity category, the commodity's place of origin, the commodity's price, and the gender to which the commodity is applicable. Similarly, if the resource is news, the basic attribute information may include the news's category, the news's publication time, and the news's publishing platform.

[0073] In this application, object attribute information under multiple attributes included in the object information of the target object is feature-encoded to obtain object attribute features. Simultaneously, the basic resource information of each resource is encoded to obtain basic resource features. Then, the similarity between the basic resource features and the object attribute features is determined as the matching degree between the resource and the target object. Specifically, cosine similarity, Euclidean distance, etc., between the basic resource features and the object attribute features can be used as the similarity between the basic resource features and the object attribute features.

[0074] In some implementations, for each resource, object attribute information and basic resource information can be input into a coarse-ranked resource prediction model. The interest score output by the coarse-ranked resource prediction model is used as the matching degree between the resource and the target object. A higher interest score indicates a higher interest in the resource from the target object, resulting in a higher matching degree; conversely, a lower interest score indicates a lower interest in the resource from the target object, resulting in a lower matching degree.

[0075] In this embodiment, the coarse-ranked resource prediction model can be obtained by training a neural network model with initialized parameters based on the object attribute information of the sample object and the basic resource information of the sample resources recommended for the sample object. The neural network model can include recurrent neural networks, regression models, and convolutional neural networks, etc.

[0076] After determining the matching degree between each resource and the target object, candidate resources can be filtered based on the matching degree. For example, a second number of resources can be selected from multiple initial resources according to the matching degree from high to low, or resources with a matching degree reaching a second threshold can be selected as candidate resources.

[0077] S120. For each candidate resource, in order of increasing interest characterization granularity, the resource features of the candidate resource are sequentially fused with the object features of the target object at N interest characterization granularities to obtain the object fusion features of the target object relative to the candidate resources.

[0078] The input to the focusing fusion process of resource features and target object features at the (i+1)th interest characterization granularity includes i-level focusing fusion features, which are the results obtained by focusing fusion of resource features and target object features at the i-th interest characterization granularity.

[0079] In this application, focused fusion refers to using candidate resources as observers (targets) and focusing the attention of the target object's object features at different interest characterization granularities onto the candidate resources in order to extract interest features related to the candidate resources from the target object's object features. In this way, the object fusion features of the target object relative to the candidate resources obtained through focused fusion incorporate the perception of the candidate resources, so that the object fusion features of the target object relative to the candidate resources can reflect the target object's interest preferences related to the candidate resources.

[0080] For example, if the resource is an article, and the candidate resource is candidate entertainment information, then focusing on fusing the features of the candidate entertainment information with the object features of the target object can make the object fusion features of the target object relative to the candidate entertainment information reflect the target object's interest preferences related to the candidate entertainment information. That is, the object fusion features more reflect the target object's interest preferences in entertainment. If the candidate resource is candidate social information, focusing on fusing the features of the candidate social information with the object features of the target object can make the object fusion features of the target object relative to the social information reflect the target object's interest preferences related to the social information. This achieves the integration of the object fusion features of the target object relative to the candidate resource into the perception of the candidate resource.

[0081] Furthermore, in this application, the resource features of the candidate resources are sequentially fused with the object features of the target object at N interest characterization granularities, and the i-th level focused fusion feature obtained from the previous level of fusion is used as the input for the focused fusion process of the resource features and the object features of the target object at the (i+1)-th interest characterization granularity. In this way, the features of the candidate resources are dynamically perceived during multiple focused fusion processes, and the dynamic perception of the candidate resources during multiple focused fusion processes is incorporated into the object fusion features of the target object relative to the candidate resources.

[0082] Furthermore, the resource features of the candidate resources are sequentially fused with the object features of the target object at N interest characterization granularities in order of increasing interest characterization granularity. This fully utilizes the progressive relationship between object features at different interest characterization granularities, so that the object fusion features of the target object relative to the candidate resources can accurately reflect the target object's interest preference relative to the candidate resources.

[0083] In some embodiments, the basic information of the candidate resources (e.g., if the resource is a video, the basic information of the video may include the video name, video category, video publishing platform, video duration, video description, etc.) can be feature-encoded to obtain the resource features of the candidate resources. Then, for each candidate resource, the interest characterization granularity is traversed from small to large. For each current interest characterization granularity in the current traversal, if the current interest characterization granularity in the current traversal is level 1, the object features of the target object at level i can be focused and fused with the resource features of the candidate resources to obtain level 1 focused fused features; if the current interest characterization granularity is not level 1... The process involves fusing the object features of the target object at the current interest characterization granularity with the resource features of the candidate resources to obtain intermediate features at the current interest characterization granularity. Then, the intermediate features at the current interest characterization granularity are fused with the focused fusion features of the previous interest characterization granularity. Through these two fusions, the resource features of the candidate resources and the object features of the target object at the current interest characterization granularity are focusedly fused to obtain focused fusion features at the current interest characterization granularity. After obtaining the focused fusion features at each interest characterization granularity, the object fusion features of the target object relative to the candidate resources are determined based on the focused fusion features at each interest characterization granularity.

[0084] For example, when the current interest characterization granularity is level 1, the object features of the target object at level 1 and the resource features of the candidate resources can be concatenated and summed to achieve focused fusion and obtain level 1 focused fusion features. When the current interest characterization granularity is not level 1, the object features of the target object at the current interest characterization granularity and the resource features of the candidate resources can be concatenated and summed to achieve one fusion and obtain intermediate features at the current interest characterization granularity. Then, the intermediate features at the current interest characterization granularity are concatenated and summed with the focused fusion features of the previous interest characterization granularity to achieve one fusion, thereby achieving focused fusion and obtaining the focused fusion features at the current interest characterization granularity.

[0085] In some embodiments, the object features of the target object at the first interest characterization granularity can be first focused and fused with the resource features of the candidate resources to obtain level 1 focused fused features. For example, the object features of the target object at the first interest characterization granularity and the resource features of the candidate resources can be fully connected to obtain level 1 focused fused features. Alternatively, the object features of the target object at the first interest characterization granularity can be fully connected first, and then the fully connected features can be focused and fused with the resource features of the candidate resources based on an attention mechanism to obtain level 1 focused fused features. Subsequently, for the object features of the target object at the j-th interest characterization granularity (where j is an integer and j∈[2,N]), the j-1 level focused fused features are fused with the object features of the target object at the j-th interest characterization granularity to obtain j-th intermediate fused features; the j-th intermediate fused features are then focused and fused with the resource features of the candidate resources to obtain j-th level focused fused features of the target object relative to the candidate resources; finally, based on the level 1 to N level focused fused features of the target object relative to the candidate resources, the object fused features of the target object relative to the candidate resources are determined.

[0086] S130. Determine the target object's interest in each candidate resource based on the object fusion characteristics of the target object relative to each candidate resource.

[0087] For each candidate resource, analysis can be performed based on the object fusion features of the target object relative to that candidate resource to obtain the target object's interest level relative to that candidate resource. The interest level of the candidate resource is used to indicate the degree of interest of the target object in the candidate resource. The higher the interest level of the candidate resource, the higher the degree of interest of the target object in the candidate resource, and the lower the interest level of the candidate resource, the lower the degree of interest of the target object in the candidate resource.

[0088] In some embodiments, for each candidate resource, the object fusion features of the target object relative to the candidate resource can be input into the resource prediction model to obtain the interest degree of the target object relative to each candidate resource predicted by the resource prediction model.

[0089] The resource prediction model can be trained on a neural network model with initialized parameters, based on the object fusion features of the sample object relative to each sample resource and the label information of each sample resource. The label information of the sample resource can be used to indicate whether the sample object is interested in or not interested in the sample resource (e.g., a label of 1 indicates interest in the sample resource, and a label of 0 indicates disinterest). The object fusion features of the sample object relative to the sample resource can be determined based on the object features of the sample object at each of the N interest characterization granularities and the resource features of the sample resource. The process of determining the object fusion features of the sample object relative to the sample resource is similar to the process of determining the object fusion features of the target object relative to the candidate resource, and will not be repeated here.

[0090] In this application, the neural network model used to train the resource prediction model can be a logistic regression model, a recurrent neural network model, a long short-term memory network model, a factorization machine, etc.

[0091] S140. Based on the target object's interest in each candidate resource, determine the target resource to be pushed to the target object from multiple candidate resources.

[0092] After obtaining the interest level of each candidate resource, a first number of candidate resources can be selected as the target resource from multiple candidate resources according to the interest level from high to low, or, the candidate resources whose interest level reaches a first threshold can be selected as the target resource. The first number and the first threshold can be set based on requirements; the first number can be less than a second number, for example, the first number is 20 and the first threshold is 0.8.

[0093] In some implementations, candidate resources with interest reaching a first threshold may be selected as intermediate candidate resources, and then a third number of intermediate candidate resources may be randomly selected from the intermediate candidate resources as the target resource.

[0094] In this embodiment, following the order of increasing interest characterization granularity, the resource features of candidate resources are sequentially fused with the object features of the target object at N interest characterization granularities to obtain the object fusion features of the target object relative to the candidate resources. Furthermore, the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity. The result of the fusion of resource features and the object features of the target object at the i-th interest characterization granularity is also used as input for the fusion process of resource features and the object features of the target object at the (i+1)-th interest characterization granularity. This achieves dynamic perception of candidate resource features during multiple fusion processes. The object fusion features of the target object relative to the candidate resources incorporate the dynamic perception of candidate resources during multiple fusion processes. Moreover, following the order of increasing interest characterization granularity... The resource features of candidate resources are sequentially fused with the object features of the target object at N interest characterization granularities in ascending order. This fully utilizes the progressive relationship between object features at different interest characterization granularities, ensuring that the object fusion features of the target object relative to the candidate resources accurately reflect the target object's interest preferences. Subsequently, the accuracy of determining the target object's interest in each candidate resource based on the object fusion features determined by the fused features is high. The interest can more accurately reflect the target object's demand for each candidate resource, ensuring that the target resources pushed to the target object meet the target object's needs without requiring the target object to communicate with the server multiple times to obtain the required resources, thus improving the utilization rate of communication resources.

[0095] Please see Figure 4 , Figure 4 This application illustrates a flowchart of a resource processing method according to another embodiment of the present application. The method can be used in an electronic device, which may be... Figure 1 The terminal device 400 or server 200 in the middle, the method may include:

[0096] S210. Obtain multiple candidate resources and obtain the object features of the target object at each of the N interest characterization granularities.

[0097] The description of S210 is the same as that of S110 above, and will not be repeated here.

[0098] S220. For each candidate resource, based on the resource features of the candidate resource and the object features of the target object at the first interest characterization granularity, determine the attention score of the first-level object features; perform focus fusion based on the attention score of the first-level object features and the first-level object features to obtain the first-level focus fusion features of the target object relative to the candidate resources.

[0099] In some implementations, determining the attention score of a Level 1 object feature based on the resource features of candidate resources and the object features of the target object at the first interest characterization granularity includes: performing fully connected processing on the object features of the target object at the first interest characterization granularity to obtain Level 1 intermediate fusion features, and then determining Level 1 key vectors and Level 1 value vectors based on the Level 1 intermediate fusion features; using the resource features of candidate resources as query vectors, determining the attention score of the Level 1 object feature based on the query vectors and the Level 1 key vectors; and performing focused fusion based on the attention score of the Level 1 object feature and the Level 1 object feature to obtain the Level 1 focused fusion feature of the target object relative to the candidate resources, including: weighting the Level 1 value vector based on the attention score of the Level 1 object feature to obtain the Level 1 focused fusion feature of the target object relative to the candidate resources. The fully connected layer used for fully connected processing of the object features of the target object at the first interest characterization granularity can be one layer or multiple layers.

[0100] In some embodiments, the first-level intermediate fusion feature of the target object is a feature matrix that can convert the object features of the target object at the first level into two sparse matrices, which serve as the first-level key vector and the first-level value vector, respectively.

[0101] In some embodiments, two weight matrices W can also be obtained. k and W v Multiply the object features of the target object at level 1 by W. k This yields a Level 1 key vector, which is then multiplied by W based on the object features of the target object at Level 1. v This yields the j-th level value vector. The weight matrix W... k and W v It can be the weight matrix in the trained attention mechanism network, which can be a self-attention mechanism network, a multi-head self-attention mechanism network, etc.

[0102] After obtaining the Level 1 key vector, the attention score can be calculated using Formula 1, as follows:

[0103]

[0104] Where attn is the attention score of the level 1 object feature, Q is the resource feature of the level 1 candidate resource, and K is the resource feature of the level 1 candidate resource. T Let d be the transpose of the first-order bond space vector K. k Let be the dimension of the first-level bond space vector, and softmax be the activation function.

[0105] Accordingly, after obtaining the attention scores of the Level 1 object features, the Level 1 focus fusion features of the candidate resources can be determined according to Formula 2, as follows:

[0106]

[0107] Where output is the Level 1 focused fusion feature and V is the Level 1 value vector.

[0108] In another embodiment, determining the attention score of the level 1 object feature based on the resource features of the candidate resource and the object features of the target object at the first interest characterization granularity may include: performing fully connected processing on the object features of the target object at the first interest characterization granularity to obtain level 1 intermediate fusion features, and then determining the level 1 key vector and level 1 value vector based on the level 1 intermediate fusion features; using the resource features of the candidate resource as the query vector, and determining the attention score of the level 1 object feature based on the query vector, the level 1 key vector, and the level 1 value vector.

[0109] For example, the attention score can be calculated using Formula 3, which is as follows:

[0110] attn = softmax(V T tanh(W k K+W Q Q)) (3)

[0111] Among them, W Q W represents the weights corresponding to the resource features of the level 1 candidate resources. Q Let be the weight matrix in the trained attention mechanism network, and tanh be the activation function.

[0112] S230. For each candidate resource, obtain the j-1 level focusing fusion feature of the target object relative to the candidate resource.

[0113] Where j is an integer and j∈[2,N], and as mentioned above, the level 1 focused fusion feature is obtained by focusing and fusing the resource features of the candidate resource with the object features of the target object at the first interest characterization granularity.

[0114] S240. The j-1 level focused fusion feature is fused with the object feature of the target object at the j-th interest characterization granularity to obtain the j-level intermediate fusion feature; the j-level intermediate fusion feature is focused fused with the resource feature of the candidate resource to obtain the j-level focused fusion feature of the target object relative to the candidate resource.

[0115] For the first interest characterization granularity, the resource features of the candidate resource and the object features of the target object under the first interest characterization granularity are directly focused and fused to obtain the first-level focused fusion feature. For the j-th interest characterization granularity, the focused fusion feature of the previous interest characterization granularity is obtained, and the focused fusion feature of the previous interest characterization granularity and the object features of the target object under the interest characterization granularity are fused to obtain the j-th level intermediate fusion feature. Then, based on the j-th level intermediate fusion feature and the resource features of the candidate resource, the focused fusion feature of the interest characterization granularity is further determined.

[0116] In this application, the j-1 level focused fusion feature can be concatenated with the object feature of the target object at the j-th interest characterization granularity to obtain the j-level concatenated feature; the j-level concatenated feature is then subjected to fully connected processing to obtain the j-level intermediate fusion feature. The fully connected layer used for the fully connected processing can be a single fully connected layer or multiple fully connected layers.

[0117] After determining the j-th level intermediate fusion feature, the process of focusing and fusing the j-th level intermediate fusion feature with the resource features of the candidate resources to obtain the j-th level focused fusion feature of the target object relative to the candidate resources includes: determining the attention score of the j-th level intermediate fusion feature based on the j-th level intermediate fusion feature and the resource features of the candidate resources; and performing focused fusion based on the attention score of the j-th level intermediate fusion feature and the j-th level intermediate fusion feature to obtain the j-th level focused fusion feature of the target object relative to the candidate resources.

[0118] In other words, by using the attention mechanism, based on the j-level intermediate fusion features and the resource features of the candidate resources, the j-level focused fusion features of the target object relative to the candidate resources are determined.

[0119] In some embodiments, determining the attention score of the j-level intermediate fusion feature based on the j-level intermediate fusion feature and the resource features of the candidate resource includes: determining the j-level key vector and the j-level value vector based on the j-level intermediate fusion feature; using the resource features of the candidate resource as a query vector, determining the attention score of the j-level intermediate fusion feature based on the query vector and the j-level key vector; and performing focused fusion based on the attention score and the j-level intermediate fusion feature to obtain the j-level focused fusion feature of the target object relative to the candidate resource, including: weighting the j-level value vector based on the attention score of the j-level intermediate fusion feature to obtain the j-level focused fusion feature of the target object relative to the candidate resource.

[0120] In some embodiments, the j-th level intermediate fusion feature is a feature matrix, which can be converted into two sparse matrices, serving as the j-th level key space vector and the j-th level value vector, respectively.

[0121] In some embodiments, two weight matrices W can also be obtained. k and W v Multiply the j-th level intermediate fusion feature by W k This yields the j-th level key space vector, and the j-th level intermediate fusion feature is multiplied by W. v This yields the j-th level value space vector. The weight matrix W... k and W v It can be the weight matrix in the trained attention mechanism network, which can be a self-attention mechanism network, a multi-head self-attention mechanism network, etc.

[0122] After determining the j-level key space vector and the j-level value vector, the j-level intermediate fusion feature can be used as Q in Formula 1 (or Formula 3) and Formula 2. The j-level focusing fusion feature of the target object relative to the candidate resource can be determined in the manner of Formula 1 (or Formula 3) and Formula 2.

[0123] S250. Determine if j equals N. If not, it means that the determination of all interest characterization granularity focusing fusion features has not yet been completed. Increment j by 1 and return to step S230. If yes, proceed to step S260.

[0124] S260. Based on the N-level focusing fusion characteristics of the target object relative to the candidate resources, determine the object fusion characteristics of the target object relative to the candidate resources.

[0125] In some implementations, for each candidate resource, the focus fusion features of the target object relative to the candidate resource at multiple interest characterization granularities can be weighted and summed. The sum is then used as the object fusion feature of the target object relative to the candidate resource. The larger the number of interest characterization granularities, the higher the weight of the focus fusion feature at that interest characterization granularity. Since, for any layer, the larger the number of interest characterization granularities, the more accurate the focus fusion feature at that interest characterization granularity is, and therefore, it has a higher weight.

[0126] In other implementations, for each candidate resource, the N-level focus fusion feature of the target object relative to the candidate resource can also be used as the object fusion feature of the target object relative to the candidate resource. This N-level focus fusion feature fuses the focus fusion features of all interest characterization granularities of the previous N-1 interest characterization granularities, as well as the N and object features, resulting in a high fusion accuracy for the N-level focus fusion feature.

[0127] S270. Based on the object fusion characteristics of the target object relative to each candidate resource, determine the target object's interest in each candidate resource; based on the target object's interest in each candidate resource, determine the target resource to be pushed to the target object from multiple candidate resources.

[0128] The description of S270 is the same as that of S130-S140 above, and will not be repeated here.

[0129] In this embodiment, the attention mechanism is unidirectional: the focused fusion features of the previous interest characterization granularity based on the attention mechanism are passed to the next interest characterization granularity for attention mechanism operation. This ensures that the learning of the next interest characterization granularity is a progressive relearning based on the learning results of the previous interest characterization granularity. On the one hand, it fully utilizes the progressive relationship between object features of different interest characterization granularities. On the other hand, by using candidate resources as observers, the resource features of candidate resources are sequentially focused and fused with the object features of the target object under N interest characterization granularities in order of increasing interest characterization granularity. This achieves a progressive step-by-step process, like a thought chain. The focused learning approach integrates effective features from the coarser feature expression granularity of the object features at a more refined interest characterization level into the object features at a finer interest characterization level. This makes better use of object features at different granularities, enabling a more accurate characterization of the interest representations related to the target object and candidate resources. It effectively improves the accuracy of focused fusion features at each interest characterization level, making the object fusion features determined based on the focused fusion features at each interest characterization level more accurate. This improves the accuracy of the interest level determined based on the object fusion features, making the target resources selected based on the interest level more in line with the interests and needs of the target object.

[0130] Meanwhile, for each interest characterization granularity, the resource features of candidate resources are fused based on the attention mechanism, which effectively enables dynamic perception of candidate resources and achieves better identification of the target object's interest preferences for candidate resources. This makes the focused fusion features more accurately indicate the target object's interest in candidate resources, thereby improving the accuracy of the target resources determined by the focused fusion features based on each interest characterization granularity.

[0131] To more clearly explain the resource processing method of this application, a specific example is used below to illustrate the method.

[0132] like Figure 5As shown, in response to a refresh operation on the news recommendation page of user client A, the terminal device generates a news retrieval request and sends it to client A's server. The server responds to the news request by retrieving 5000 initial news items from the news content pool, along with the user's attribute information. This user attribute information includes basic attribute information, profile attribute information, interactive resource information, and user identification information. The basic attribute information includes age and gender; the profile attribute information includes whether the user is a programmer and likes anime / manga; the interactive resource information includes the user's 3 hours of browsing technology news and the list of technology news items the user has viewed; and the user identification information includes a unique identifier (ay) set by the server for the user.

[0133] In the coarse ranking phase, a coarse ranking resource prediction model is used to determine the user's interest level relative to each initial news item based on the user's age and gender. Initial news items with an interest level higher than 0.5 are then selected as candidate news items. At this point, 1000 candidate news items are identified.

[0134] In the fine-tuning stage, based on basic attribute information, profile attribute information, interactive resource information, and identifier attribute information, according to... Figure 6 The process shown is used to obtain object fusion features, which are then used to determine the target news.

[0135] For any candidate news item, such as Figure 6 As shown, feature extraction is performed on the basic attribute information to obtain the first feature. Then, the first feature is processed by a fully connected layer to obtain the first-level intermediate fusion feature. Then, the first-level key vector and the first-level value vector are determined based on the first-level intermediate fusion feature. The news features of the candidate news are used as the query vector. Attention operation is performed through the first-level key vector and the first-level value vector to obtain the first-level focused fusion feature.

[0136] The second feature obtained by encoding the portrait attribute information is concatenated with the first-level focused fusion feature to obtain the first concatenation result. The first concatenation result is then subjected to full connection processing to obtain the second-level intermediate fusion feature. Then, the second-level key vector and the second-level value vector are determined based on the second-level intermediate fusion feature. The news features of the candidate news are used as the query vector. Attention operation is performed through the second-level key vector and the second-level value vector to obtain the second-level focused fusion feature.

[0137] The third feature after encoding the interactive resource information is concatenated with the second-level focused fusion feature to obtain the second concatenation result. The second concatenation result is then processed by a fully connected layer to obtain the third-level intermediate fusion feature. The third-level key vector and the third-level value vector are then determined based on the third-level intermediate fusion feature. The news features of the candidate news are used as the query vector. Attention is then performed on the third-level key vector and the third-level value vector to obtain the third-level focused fusion feature.

[0138] The fourth feature after encoding the object identification information is concatenated with the third-level focused fusion feature to obtain the third concatenation result. The third concatenation result is then fully connected to obtain the fourth-level intermediate fusion feature. The fourth-level key vector and the fourth-level value vector are then determined based on the fourth-level intermediate fusion feature. The news features of the candidate news are used as the query vector. Attention is then performed on the fourth-level key vector and the fourth-level value vector to obtain the fourth-level focused fusion feature.

[0139] Subsequently, the four-focused fusion features of the user relative to the candidate news are obtained as the object fusion features of the user relative to the candidate news. Based on the object fusion features of the user relative to the candidate news, the user's interest in the candidate news is determined by the resource prediction model.

[0140] The process described above is used to iterate through each candidate news item and obtain the interest level of each candidate news item. 100 candidate news items are selected as target news items according to the interest level from high to low. Then, the 100 target news items are sorted according to the interest level from high to low to obtain a news sequence. The news sequence is sent to the terminal device so that the terminal device can display the news sequence through client A.

[0141] Please see Figure 7 , Figure 7 This illustration shows a block diagram of a resource processing apparatus according to an embodiment of the present application. The resource processing apparatus 1500 includes:

[0142] The acquisition module 1510 is used to acquire multiple candidate resources and acquire the object features of the target object at each of the N interest characterization granularities; the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity, i∈[1,N], and N is an integer greater than 1;

[0143] The fusion module 1520 is used to perform focused fusion of the resource features of each candidate resource with the object features of the target object at N interest characterization granularities in order of increasing interest characterization granularity, in order of increasing interest characterization granularity, to obtain the object fusion features of the target object relative to the candidate resources. The input of the focused fusion process of the resource features and the object features of the target object at the (i+1)th interest characterization granularity includes the i-th level focused fusion feature, which is the result of the focused fusion of the resource features and the object features of the target object at the i-th interest characterization granularity.

[0144] The first determining module 1530 is used to determine the interest degree of the target object in each of the candidate resources based on the object fusion characteristics of the target object relative to each candidate resource;

[0145] The second determining module 1540 is used to determine the target resource to be pushed to the target object from multiple candidate resources based on the target object's interest in each candidate resource.

[0146] Optionally, the fusion module 1520 is further configured to, for each candidate resource, obtain the j-1 level focused fusion feature of the target object relative to the candidate resource, where j is an integer and j∈[2, N]. The level 1 focused fusion feature is obtained by focusing and fusing the resource features of the candidate resource with the object features of the target object at the first interest characterization granularity; the j-1 level focused fusion feature is fused with the object features of the target object at the j interest characterization granularity to obtain the j-level intermediate fusion feature; the j-level intermediate fusion feature is then focused and fused with the resource features of the candidate resource to obtain the j-level focused fusion feature of the target object relative to the candidate resource; if j is less than N, j is incremented by 1, and the process returns to the step of obtaining the j-1 level focused fusion feature of the target object relative to the candidate resource; if j = N, the object fusion feature of the target object relative to the candidate resource is determined based on the N-level focused fusion feature of the target object relative to the candidate resource.

[0147] Optionally, the fusion module 1520 is further configured to determine the attention score of the j-level intermediate fusion feature based on the j-level intermediate fusion feature and the resource features of the candidate resource; and to perform focused fusion based on the attention score of the j-level intermediate fusion feature and the j-level intermediate fusion feature to obtain the j-level focused fusion feature of the target object relative to the candidate resource.

[0148] Optionally, the fusion module 1520 is further configured to determine the j-level key vector and the j-level value vector based on the j-level intermediate fusion features; use the resource features of the candidate resources as the query vector, determine the attention score of the j-level intermediate fusion features based on the query vector and the j-level key vector; and weight the j-level value vector based on the attention score of the j-level intermediate fusion features to obtain the j-level focused fusion features of the target object relative to the candidate resources.

[0149] Optionally, the fusion module 1520 is also used to concatenate the j-1 level focused fusion feature with the object feature of the target object at the j-th interest characterization granularity to obtain the j-level concatenated feature; and to perform fully connected processing on the j-level concatenated feature to obtain the j-level intermediate fusion feature.

[0150] Optionally, the fusion module 1520 is further configured to use the N-level focused fusion feature of the target object relative to the candidate resource as the object fusion feature of the target object relative to the candidate resource.

[0151] Optionally, the acquisition module 1510 is further configured to acquire object information of the target object, including object attribute information under multiple attributes; based on the correspondence between attributes and interest characterization granularity, determine the interest characterization granularity to which each object attribute information belongs among N interest characterization granularities; for each interest characterization granularity, perform feature encoding on the object attribute information belonging to that interest characterization granularity to obtain the object features of the target object under the interest characterization granularity.

[0152] Optionally, the first determining module 1530 is further configured to input the object fusion features of the target object relative to the candidate resources into the resource prediction model for each candidate resource, so as to obtain the target object's interest in the candidate resources predicted by the resource prediction model.

[0153] Optionally, the second determining module 1540 is further configured to select a first number of candidate resources with the highest interest from multiple candidate resources as target resources to be pushed to the target object; or select candidate resources with interest reaching a first threshold from multiple candidate resources as target resources to be pushed to the target object.

[0154] Optionally, the acquisition module 1510 is further configured to determine the matching degree between each resource and the target object based on the object information of the target object and the basic resource information of each resource in the resource set; and select multiple candidate resources for the target object from the resource set based on the matching degree.

[0155] It should be noted that the device embodiments in this application correspond to the aforementioned method embodiments. The specific principles in the device embodiments can be found in the content of the aforementioned method embodiments, and will not be repeated here.

[0156] Figure 8 A structural block diagram of an electronic device for performing a resource processing method according to an embodiment of this application is shown. The electronic device may be... Figure 1 The server 200 or terminal device 400, etc., should be noted. Figure 8 The computer system 1200 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0157] like Figure 8 As shown, the computer system 1200 includes a Central Processing Unit (CPU) 1201, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on programs stored in Read-Only Memory (ROM) 1202 or programs loaded from storage portion 1208 into Random Access Memory (RAM) 1203. The RAM 1203 also stores various programs and data required for system operation. The CPU 1201, ROM 1202, and RAM 1203 are interconnected via a bus 1204. An Input / Output (I / O) interface 1205 is also connected to the bus 1204.

[0158] The following components are connected to I / O interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to I / O interface 1205 as needed. Removable media 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1210 as needed so that computer programs read from them can be installed into storage section 1208 as needed.

[0159] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1209, and / or installed from removable medium 1211. When the computer program is executed by central processing unit (CPU) 1201, it performs various functions defined in the system of this application.

[0160] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0161] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0162] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0163] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium carries computer-readable instructions that, when executed by a processor, implement the methods in any of the above embodiments.

[0164] According to one aspect of the embodiments of this application, a computer program product is provided, the computer program product including 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 methods of any of the above embodiments.

[0165] In the embodiments of this application, 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. It can be implemented wholly or partially using software, hardware (e.g., 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 functions as a whole.

[0166] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0167] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause an electronic device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0168] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A resource processing method, characterized in that, The method includes: Obtain multiple candidate resources and obtain the object features of the target object under each of the N interest characterization granularities; the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity, i∈[1,N], and N is an integer greater than 1; For each candidate resource, the resource features of the candidate resource are sequentially fused with the object features of the target object at N interest characterization granularities in order of increasing interest characterization granularity, to obtain the object fusion features of the target object relative to the candidate resources. The input of the fusion process of the resource features and the object features of the target object at the (i+1)th interest characterization granularity includes the i-th level fusion feature, which is the result of the fusion of the resource features and the object features of the target object at the i-th interest characterization granularity. Based on the object fusion characteristics of the target object relative to each of the candidate resources, the interest degree of the target object in each of the candidate resources is determined; Based on the target object's interest in each of the candidate resources, a target resource is determined from the plurality of candidate resources and pushed to the target object.

2. The method according to claim 1, characterized in that, For each of the candidate resources, the resource features of the candidate resources are sequentially fused with the object features of the target object at N interest characterization granularities, in ascending order of interest characterization granularity, to obtain the object fusion features of the target object relative to the candidate resources, including: For each of the candidate resources, the j-1 level focused fusion feature of the target object relative to the candidate resource is obtained, where j is an integer and j∈[2,N]. The level 1 focused fusion feature is obtained by focusing and fusing the resource features of the candidate resource with the object features of the target object at the first interest characterization granularity. The j-1 level focused fusion feature is fused with the object feature of the target object at the j-th interest characterization granularity to obtain the j-th level intermediate fusion feature; The j-th level intermediate fusion feature is focused and fused with the resource feature of the candidate resource to obtain the j-th level focused fusion feature of the target object relative to the candidate resource; If j is less than N, increment j by 1 and return to the step of obtaining the j-1 level focusing fusion feature of the target object relative to the candidate resource; If j = N, the object fusion feature of the target object relative to the candidate resource is determined based on the N-level focusing fusion feature of the target object relative to the candidate resource.

3. The method according to claim 2, characterized in that, The step of focusing and fusing the j-th level intermediate fusion feature with the resource features of the candidate resource to obtain the j-th level focused fusion feature of the target object relative to the candidate resource includes: Based on the j-th level intermediate fusion feature and the resource features of the candidate resource, determine the attention score of the j-th level intermediate fusion feature; Focused fusion is performed based on the attention score of the j-th level intermediate fusion feature and the j-th level intermediate fusion feature to obtain the j-th level focused fusion feature of the target object relative to the candidate resource.

4. The method according to claim 3, characterized in that, The step of determining the attention score of the j-th level intermediate fusion feature based on the j-th level intermediate fusion feature and the resource features of the candidate resource includes: Based on the j-th level intermediate fusion features, determine the j-th level key vector and the j-th level value vector; Using the resource features of the candidate resources as the query vector, the attention score of the j-level intermediate fusion feature is determined based on the query vector and the j-level key vector. The step of performing focused fusion based on the attention score of the j-th level intermediate fusion feature and the j-th level intermediate fusion feature to obtain the j-th level focused fusion feature of the target object relative to the candidate resource includes: The j-level value vector is weighted based on the attention score of the j-level intermediate fusion feature to obtain the j-level focused fusion feature of the target object relative to the candidate resource.

5. The method according to claim 2, characterized in that, The step of fusing the j-1 level focused fusion feature with the object feature of the target object at the j-th interest characterization granularity to obtain the j-th level intermediate fusion feature includes: The j-1 level focused fusion feature is concatenated with the object feature of the target object at the j-th interest characterization granularity to obtain the j-th level concatenated feature; The j-level splicing features are fully connected to obtain j-level intermediate fusion features.

6. The method according to claim 2, characterized in that, The step of determining the object fusion feature of the target object relative to the candidate resource based on the N-level focusing fusion feature of the target object relative to the candidate resource includes: The N-level focus fusion feature of the target object relative to the candidate resource is used as the object fusion feature of the target object relative to the candidate resource.

7. The method according to claim 1, characterized in that, The acquisition of object features of the target object at each of the N interest characterization granularities includes: Obtain object information of the target object, the object information including object attribute information under multiple attributes; Based on the correspondence between attributes and interest characterization granularity, the interest characterization granularity to which each object attribute information belongs is determined among the N interest characterization granularities. For each interest characterization granularity, the object attribute information belonging to that interest characterization granularity is feature-encoded to obtain the object features of the target object under the interest characterization granularity.

8. The method according to claim 7, characterized in that, The object information includes at least two of the following: basic attribute information under basic attributes, portrait attribute information under portrait attributes, interaction resource information under interaction attributes, and object identification information under identification attributes. The interaction resource information includes resource information of multiple resources that trigger the interaction operation of the target object. The interest characterization granularity corresponding to the basic attribute is smaller than that corresponding to the profile attribute, the interest characterization granularity corresponding to the profile attribute is smaller than that corresponding to the interaction attribute, and the interest characterization granularity corresponding to the interaction attribute is smaller than that corresponding to the identification attribute.

9. The method according to any one of claims 1 to 8, characterized in that, The step of determining the target object's interest in each of the candidate resources based on the object fusion features of the target object relative to each of the candidate resources includes: For each of the candidate resources, the object fusion features of the target object relative to the candidate resource are input into the resource prediction model to obtain the interest degree of the target object in the candidate resource predicted by the resource prediction model.

10. The method according to any one of claims 1 to 8, characterized in that, The step of determining the target resource to be pushed to the target object from the plurality of candidate resources based on the target object's interest in each of the candidate resources includes: Select the first number of candidate resources with the highest interest from the plurality of candidate resources as the target resources to be pushed to the target object; or From the plurality of candidate resources, the candidate resources whose interest level reaches a first threshold are selected as the target resources to be pushed to the target object.

11. The method according to any one of claims 1 to 8, characterized in that, The acquisition of multiple candidate resources includes: Based on the object information of the target object and the basic resource information of each resource in the resource set, determine the matching degree between each resource and the target object; Based on the matching degree, multiple candidate resources are selected from the resource set for the target object.

12. A resource processing device, characterized in that, The device includes: The acquisition module is used to acquire multiple candidate resources and acquire the object features of the target object at each of the N interest characterization granularities; the i-th interest characterization granularity is greater than the (i+1)-th interest characterization granularity, i∈[1,N], and N is an integer greater than 1; The fusion module is used to perform focused fusion of the resource features of each candidate resource with the object features of the target object at N interest characterization granularities in an order of increasing interest characterization granularity, to obtain the object fusion features of the target object relative to the candidate resources. The input of the focused fusion process of the resource features and the object features of the target object at the (i+1)th interest characterization granularity includes the i-th level focused fusion feature, which is the result obtained by focusing fusion of the resource features and the object features of the target object at the i-th interest characterization granularity. The first determining module is used to determine the target object's interest in each of the candidate resources based on the object fusion features of the target object relative to each of the candidate resources; The second determining module is used to determine the target resource to be pushed to the target object from the plurality of candidate resources based on the target object's interest in each of the candidate resources.

13. An electronic device, characterized in that, include: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, It stores computer-readable instructions that, when executed by a processor, implement the method as described in any one of claims 1-11.

15. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the method of any one of claims 1-11.

Images