Large model-based resource recommendation method and device, electronic equipment, medium and program product

Abstract

The present disclosure provides a large model-based resource recommendation method and device, electronic equipment, medium and program product, relates to the technical field of artificial intelligence, and in particular to the fields of large models, intelligent recommendation, agents, etc. The implementation scheme is: obtaining a plurality of content tags associated with a candidate resource; performing clustering processing on the plurality of content tags to obtain a plurality of candidate clusters; based on a large model, evaluating the association strength between the two candidate clusters of each candidate cluster pair constructed; based on the evaluation result, determining at least one first target cluster pair; determining a first candidate cluster associated with a resource to be recommended and a second candidate cluster associated with historical operation data; in response to the existence of a second target cluster pair including the first candidate cluster and the second candidate cluster in the first target cluster pair, determining the association strength between the first candidate cluster and the second candidate cluster as a target association strength; based on the target association strength, determining the ranking of the resource to be recommended for resource recommendation.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to the fields of large models, intelligent recommendation, and intelligent agents, specifically to a resource recommendation method, apparatus, electronic device, computer-readable storage medium, computer program product, and an intelligent agent based on a large model. Background Technology

[0002] Artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies mainly include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0003] In content distribution scenarios such as resource display, recommendation systems need to filter suitable content for users from massive amounts of resources. To meet users' increasingly diverse content needs, recommendation systems not only need to satisfy existing user interests and preferences but also need to expand reasonably to broader interest areas to enhance the richness of recommended resources. Compared to traditional recommendation methods based on keyword matching or semantic similarity, recommendation methods based on large models possess stronger semantic understanding and knowledge reasoning capabilities, enabling them to more deeply uncover potential connections between resources. Therefore, how to fully utilize the capabilities of large models to improve the richness of resource recommendations has become an important technological direction.

[0004] The methods described in this section are not necessarily methods that had been previously conceived or adopted. Unless otherwise specified, no method described in this section should be assumed to be prior art simply because it is included in this section. Similarly, unless otherwise specified, the issues mentioned in this section should not be considered to be accepted in any prior art. Summary of the Invention

[0005] This disclosure provides a resource recommendation method, apparatus, electronic device, computer-readable storage medium, computer program product, and intelligent agent based on a large model.

[0006] According to one aspect of this disclosure, a resource recommendation method based on a large model is provided, comprising: acquiring multiple content tags associated with candidate resources; performing clustering processing on the multiple content tags to obtain multiple candidate clusters, wherein each candidate cluster includes at least one content tag from the multiple content tags; evaluating the association strength between two candidate clusters in each candidate cluster pair among the at least one candidate cluster pair constructed based on the large model, wherein each candidate cluster pair includes any two candidate clusters from the multiple candidate clusters; determining at least one first target cluster pair that meets preset conditions from the at least one candidate cluster pair based on the evaluation results; determining the first candidate cluster to which the content tag associated with the resource to be recommended belongs and the second candidate cluster to which the content tag associated with historical operation data belongs; in response to determining that there is a second target cluster pair including the first candidate cluster and the second candidate cluster in the at least one first target cluster pair, determining the association strength between the first candidate cluster and the second candidate cluster as the target association strength; and determining the ranking of the resource to be recommended based on the target association strength for resource recommendation.

[0007] According to another aspect of this disclosure, a resource recommendation apparatus based on a large model is provided, comprising: a content tag acquisition module configured to acquire multiple content tags associated with candidate resources; a data clustering processing module configured to perform clustering processing on the multiple content tags to obtain multiple candidate clusters, wherein each candidate cluster includes at least one content tag from the multiple content tags; an association strength evaluation module configured to evaluate the association strength between two candidate clusters in each candidate cluster pair constructed in at least one candidate cluster pair based on the large model, wherein each candidate cluster pair includes any two candidate clusters from the multiple candidate clusters; and a target cluster pair determination module configured to... To determine at least one first target cluster pair that meets preset conditions from at least one candidate cluster pair based on the evaluation results; a candidate cluster determination module is configured to determine a first candidate cluster to which a content tag associated with the resource to be recommended belongs and a second candidate cluster to which a content tag associated with historical operation data belongs; a target strength determination module is configured to determine the association strength between the first candidate cluster and the second candidate cluster as a target association strength in response to determining that there is a second target cluster pair including the first candidate cluster and the second candidate cluster in at least one first target cluster pair; and a resource ranking determination module is configured to determine the ranking of the resources to be recommended based on the target association strength for resource recommendation.

[0008] According to another aspect of this disclosure, an agent is provided, wherein the agent recommends resources to a user according to the large model-based resource recommendation method described above.

[0009] According to another aspect of this disclosure, an electronic device is provided, including at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the resource recommendation method based on a large model as described above.

[0010] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the large-model-based resource recommendation method of this disclosure as described above.

[0011] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the resource recommendation method based on a large model as described above.

[0012] According to one or more embodiments of this disclosure, the richness and diversity of resource recommendations can be improved.

[0013] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0014] The accompanying drawings exemplify embodiments and form part of the specification, serving together with the textual description to explain exemplary implementations of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. Throughout the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0015] Figure 1 A schematic diagram of an exemplary system in which the various methods described herein can be implemented according to embodiments of this disclosure is shown.

[0016] Figure 2 A flowchart of a resource recommendation method based on a large model according to an embodiment of the present disclosure is shown.

[0017] Figure 3 A schematic diagram illustrating the determination of a plurality of candidate clusters according to an embodiment of the present disclosure is shown.

[0018] Figure 4 A schematic diagram illustrating the determination of at least one first target cluster pair according to an embodiment of the present disclosure is shown.

[0019] Figure 5 A schematic diagram illustrating the determination of the order of resources to be recommended according to an embodiment of the present disclosure is shown.

[0020] Figure 6A structural block diagram of a resource recommendation apparatus based on a large model according to an embodiment of the present disclosure is shown.

[0021] Figure 7 A structural block diagram of a resource recommendation apparatus based on a large model according to another embodiment of the present disclosure is shown.

[0022] Figure 8 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0023] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] In this disclosure, unless otherwise stated, the use of terms such as "first," "second," etc., to describe various elements is not intended to limit the positional, temporal, or importance relationships of these elements; such terms are merely used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of that element, while in other cases, based on the context, they may refer to different instances.

[0025] The terminology used in the description of the various examples described in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context explicitly indicates otherwise, an element may be one or more unless the number of elements is specifically limited. Furthermore, the term "and / or" as used in this disclosure covers any one of the listed items and all possible combinations thereof.

[0026] In related technologies, resource recommendation typically expands recommended resources based on semantic similarity or content category systems. These methods mainly establish connections between resources based on their similarity in theme, category, or textual expression. However, user interests can often be linked across different interest domains through user profiles, underlying motivations, and consumption scenarios. Relying solely on the above methods is insufficient to identify such cross-domain relationships, leading to recommended resources being concentrated in similar areas that users already have interests. This results in insufficient richness of recommendation results, thus affecting the user's long-term consumption experience.

[0027] Therefore, embodiments of this disclosure provide a resource recommendation method based on a large model.

[0028] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0029] Figure 1 A schematic diagram of an exemplary system 100 in which the various methods and apparatus described herein can be implemented according to embodiments of this disclosure is shown. Reference Figure 1 The system 100 includes one or more client devices 101, 102, 103, 104, 105 and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105 and 106 can be configured to execute one or more applications.

[0030] In embodiments of this disclosure, server 120 may be configured to perform one or more services or software applications as described in embodiments of this disclosure.

[0031] In some embodiments, server 120 may also provide other services or software applications, which may include non-virtual and virtual environments. In some embodiments, these services may be provided as web-based services or cloud services, such as to users of client devices 101, 102, 103, 104, 105 and / or 106 under a Software as a Service (SaaS) model.

[0032] exist Figure 1 In the configuration shown, server 120 may include one or more components that implement the functions performed by server 120. These components may include software components, hardware components, or combinations thereof that can be executed by one or more processors. Users operating client devices 101, 102, 103, 104, 105, and / or 106 can sequentially interact with server 120 using one or more client applications to utilize the services provided by these components. It should be understood that various different system configurations are possible and may differ from system 100. Therefore, Figure 1 This is an example of a system used to implement the various methods described herein, and is not intended to be limiting.

[0033] Users can access the recommendation system to obtain resource recommendations using client devices 101, 102, 103, 104, 105, and / or 106. The client devices can provide an interface that allows users to interact with the client devices. The client devices can also output information to users through this interface. Although... Figure 1 Only six client devices are described, but those skilled in the art will understand that this disclosure can support any number of client devices.

[0034] Client devices 101, 102, 103, 104, 105, and / or 106 may include various types of computer devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptops), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, gaming systems, thin clients, various messaging devices, sensors, or other sensing devices. These computer devices can run various types and versions of software applications and operating systems, such as Microsoft Windows, Apple iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as Google Chrome OS); or include various mobile operating systems, such as Microsoft Windows Mobile OS, iOS, Windows Phone, and Android. Portable handheld devices may include cellular phones, smartphones, tablets, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. Gaming systems may include various handheld gaming devices, internet-enabled gaming devices, etc. Client devices are capable of executing various applications, such as various internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and can use various communication protocols.

[0035] Network 110 can be any type of network well known to those skilled in the art, and can support data communication using any of a variety of available protocols (including but not limited to TCP / IP, SNA, IPX, etc.). By way of example only, one or more networks 110 can be a local area network (LAN), an Ethernet-based network, a token ring network, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a blockchain network, a public switched telephone network (PSTN), an infrared network, a wireless network (e.g., Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0036] Server 120 may include one or more general-purpose computers, special-purpose server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running a virtual operating system, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices for servers). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0037] The computing unit in server 120 can run one or more operating systems, including any of the aforementioned operating systems and any commercially available server operating system. Server 120 can also run any of a variety of additional server applications and / or middleware applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0038] In some implementations, server 120 may include one or more applications to analyze and merge data feeds and / or event updates received from users of client devices 101, 102, 103, 104, 105 and / or 106. Server 120 may also include one or more applications to display data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105 and / or 106.

[0039] In some implementations, server 120 can be a server for a distributed system or a server integrated with blockchain. Server 120 can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. A cloud server is a host product in the cloud computing service system, designed to address the shortcomings of traditional physical hosts and Virtual Private Server (VPS) services, such as high management difficulty and weak business scalability.

[0040] System 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store information such as audio files and video files. Databases 130 may reside in various locations. For example, a database used by server 120 may be local to server 120, or it may be located away from server 120 and may communicate with server 120 via a network-based or dedicated connection. Databases 130 may be of different types. In some embodiments, the database used by server 120 may be, for example, a relational database. One or more of these databases may store, update, and retrieve data from and from the databases in response to commands.

[0041] In some embodiments, one or more of the databases 130 may also be used by an application to store application data. The databases used by the application may be of different types, such as key-value stores, object stores, or regular stores supported by a file system.

[0042] Figure 1The system 100 can be configured and operated in various ways to enable the application of the various methods and apparatus described in this disclosure.

[0043] The various aspects of the large model-based resource recommendation method according to embodiments of this disclosure will be described in detail below.

[0044] Figure 2 A flowchart of a resource recommendation method 200 based on a large model according to an embodiment of the present disclosure is shown.

[0045] like Figure 2 As shown, method 200 includes steps S201, S202, S203, S204, S205, S206 and S207.

[0046] In step S201, multiple content tags associated with the candidate resources are obtained.

[0047] In the example, candidate resources can be resources that the recommendation system has displayed during historical periods, such as videos, images, and other types of content available for recommendation. Content tags are used to describe the style, theme, points of interest, consumption scenarios, or content formats of these candidate resources, and can be generated by a content understanding system, a manual annotation system, or a recommendation feature system. For example, the style field of displayed resources can be read from the historical display logs of the recommendation system, and cleaned and standardized by removing null values, outliers, and duplicate values ​​to obtain multiple content tags associated with candidate resources. Since content tags come from real display records, they can reflect the actual usage of resources. Therefore, obtaining and processing such content tags can provide a data foundation for subsequent resource relationship mining.

[0048] In step S202, multiple content tags are clustered to obtain multiple candidate clusters, wherein each candidate cluster includes at least one content tag from the multiple content tags.

[0049] In the example, clustering refers to the process of grouping content tags describing similar interests into the same set based on their semantic similarity. After clustering, multiple candidate clusters are obtained. Therefore, a candidate cluster can be a set of tags composed of several content tags. Content tags within the same candidate cluster share certain commonalities in terms of theme, points of interest, or consumption scenarios, and can be considered a relatively independent unit of interest. Clustering is performed first because individual content tags are granular and numerous. Directly processing content tags would require a large amount of data, while using candidate clusters as units can reduce the scale of subsequent processing while preserving the expressive power of interests.

[0050] In step S203, based on the large model, the correlation strength between the two candidate clusters in each candidate cluster pair of at least one constructed candidate cluster pair is evaluated, wherein each candidate cluster pair includes any two candidate clusters from a plurality of candidate clusters.

[0051] In the example, a candidate cluster pair can be formed by any two candidate clusters from multiple candidate clusters, that is, by combining each candidate cluster in pairs to obtain at least one candidate cluster pair. Association strength can be used to characterize the degree of interest association between the two candidate clusters in a candidate cluster pair; the higher the value, the stronger the connection between the interest content corresponding to the two candidate clusters. This interest association can be generated not only by semantics but also by the people browsing the content corresponding to the two candidate clusters, their browsing motivations, the context in which they browse, or the interest expansion path extending from one type of content to another. These associations are often not reflected in the literal expression or semantic distance of the content tags but are implicit in the correspondence between content and people / contexts, requiring the semantic understanding and knowledge reasoning capabilities of a large model for identification. Therefore, relevant information about each candidate cluster can be provided to the large model, which can then evaluate the association strength between the two candidate clusters in each candidate cluster pair based on the above-mentioned factors, and the evaluation result can be reflected as a corresponding score.

[0052] In step S204, based on the evaluation results, at least one first target cluster pair that meets the preset conditions is determined from at least one candidate cluster pair.

[0053] In the example, the number of candidate cluster pairs obtained by combining each candidate cluster in pairs is large. Some of these candidate cluster pairs do not have any actual interest association between the two candidate clusters, so they can be filtered based on the evaluation results. The evaluation results can be the association strength assessment given by the large model for each candidate cluster pair, such as the corresponding score. A higher score indicates a stronger interest association between the two candidate clusters. Preset conditions are used to determine the degree of association strength that can be considered as an interest association between two candidate clusters. For example, a strength threshold can be set, and candidate cluster pairs whose association strength reaches this threshold are considered to meet the preset conditions. The specific value of this threshold can be adjusted based on the results of manual sampling evaluation; this disclosure does not limit this. After the above processing, candidate cluster pairs whose association strength does not meet the preset conditions are excluded, and the remaining candidate cluster pairs are retained as the first target cluster pairs.

[0054] In step S205, a first candidate cluster to which the content tags associated with the resource to be recommended belong and a second candidate cluster to which the content tags associated with historical operation data belong are determined.

[0055] In the example, the resource to be recommended refers to the resource whose ranking needs to be determined during the recommendation phase, and historical operation data refers to data such as clicks and views generated by the user over a previous period. This step maps the resource to be recommended and the historical operation data to corresponding candidate clusters. For the resource to be recommended, its content tags can be obtained, and based on the correspondence between content tags and candidate clusters formed by the previous clustering process, the candidate cluster to which the content tag belongs is determined, which is the first candidate cluster. For the historical operation data, the content tags corresponding to resources previously clicked by the user can be obtained, and their candidate clusters can be determined in the same way, which is the second candidate cluster.

[0056] In step S206, in response to determining that there is a second target cluster pair including a first candidate cluster and a second candidate cluster in at least one first target cluster pair, the association strength between the first candidate cluster and the second candidate cluster is determined as the target association strength.

[0057] In the example, the second target cluster pair can refer to the first target cluster pair consisting of a first candidate cluster and a second candidate cluster within at least one first target cluster pair; the target association strength can refer to the association strength between the two candidate clusters in the second target cluster pair, used to characterize the degree of interest association between the resource to be recommended and the content previously viewed by the user. Firstly, each first target cluster pair and its association strength can be stored. Then, during the recommendation stage, the first candidate cluster and the second candidate cluster are used as query conditions. If a corresponding first target cluster pair is found, it is treated as the second target cluster pair, and its association strength is read as the target association strength; if not found, it indicates that there is no interest association between the first candidate cluster and the second candidate cluster. For example, among the first target cluster pairs retained after the aforementioned processing, there is a first target cluster pair consisting of a candidate cluster of handmade art creations and a candidate cluster of traditional Chinese aesthetic culture, and the association strength between the two can be 0.78. During the recommendation phase, if a resource to be recommended is Hanfu (traditional Han clothing) styling content, and its content tag belongs to the first candidate cluster of the "Traditional Chinese Style Aesthetics and Culture" candidate cluster, while the second candidate cluster determined based on historical operational data is the "Handmade Art Creation" candidate cluster, then searching using these two candidate clusters will find the aforementioned first target cluster pair, indicating the existence of a second target cluster pair. Therefore, 0.78 is determined as the target association strength. Conversely, if the second candidate cluster is the "Food Tutorials" candidate cluster, and it does not form a first target cluster pair with the "Traditional Chinese Style Aesthetics and Culture" candidate cluster, then no second target cluster pair will be found, indicating that there is no interest association between the "Food Tutorials" candidate cluster and the "Traditional Chinese Style Aesthetics and Culture" candidate cluster.

[0058] In step S207, the ranking of resources to be recommended is determined based on the target association strength in order to make resource recommendations.

[0059] In the example, ranking refers to the position of the resource to be recommended in the recommendation results. The higher the position, the greater the chance that the resource will be shown to the user. Since the target association strength characterizes the degree of interest connection between the resource to be recommended and the content the user has previously viewed, the higher the strength, the greater the likelihood that the user's previous interests will extend to the content related to the resource to be recommended. Appropriately adjusting the ranking accordingly can ensure that the recommendation results not only match the user's existing interests but also cover other related interests. Therefore, the original ranking of the resources to be recommended can be adjusted based on the target association strength, so that resources with higher target association strength are ranked higher, thereby improving the diversity of the recommendation results.

[0060] Therefore, by clustering content tags and using a large model to evaluate the association strength between two candidate clusters, we can identify relationships where content tags are not literally similar but can be linked through shared browsing groups, motivations, scenarios, or interests. Furthermore, using candidate clusters as processing units significantly reduces the amount of data that needs to be processed. By identifying the first target cluster pair that meets preset conditions, candidate cluster pairs with insufficient association strength can be eliminated, improving the reliability of the association. Further, by mapping the resources to be recommended and historical operation data to the first and second candidate clusters respectively, and determining the ranking of the resources to be recommended based on the found target association strength, the recommendation results can not only match the user's existing interests but also cover other related content, thereby improving the richness and diversity of resource recommendations.

[0061] The collection, storage, use, processing, transmission, provision, and disclosure of any type of information, such as user personal information, in this technical solution comply with relevant laws and regulations and do not violate public order and good morals.

[0062] In some embodiments, such as in combination Figure 2 Step S202 involves clustering multiple content tags to obtain multiple candidate clusters, including: converting multiple content tags into multiple corresponding feature vectors; clustering based on the multiple feature vectors to obtain multiple initial clusters, wherein each initial cluster includes at least one content tag; and filtering the multiple initial clusters based on at least one content tag in the multiple initial clusters to obtain multiple candidate clusters.

[0063] In the example, a feature vector can refer to a vectorized representation that characterizes the semantic information of a content tag. Feature vectors corresponding to semantically similar content tags are also closer in distance within the vector space. An initial cluster can refer to a set of content tags aggregated solely based on the distance between their feature vectors, with each initial cluster containing at least one content tag. A semantic vector model can be used to vectorize each content tag separately to obtain its corresponding feature vector. Then, based on the distance between these feature vectors, content tags with closer distances are aggregated into the same set, thus obtaining multiple initial clusters.

[0064] In the example, since the feature vectors reflect the semantic similarity of content tags, some initial clusters may contain content tags that are similar in form but represent different topics of interest. There may also be initial clusters where the content tags as a whole lack practical reference value. If such initial clusters are directly used as units for subsequent processing, the accuracy of the results will be affected. Therefore, based on the specific content tags contained in each initial cluster, further filtering can be performed on multiple initial clusters to ensure that the content tags contained in each candidate cluster are more consistent in terms of topics of interest.

[0065] Therefore, by filtering multiple initial clusters based on the content tags contained in the initial clusters, the preliminary aggregation results can be corrected, thereby improving the accuracy and reliability of the candidate clusters.

[0066] In some embodiments, multiple initial clusters are filtered based on at least one content tag in multiple initial clusters to obtain multiple candidate clusters, including: for each initial cluster, in response to the number of at least one content tag in the initial cluster being greater than a preset threshold, constructing a first prompt word to provide to a large model, wherein the first prompt word is used to guide the large model to divide at least one content tag in the initial cluster into multiple content tag subsets based on the interest topics represented by at least one content tag in the initial cluster, wherein each content tag subset is a middle cluster; in response to the number of at least one content tag in the initial cluster being less than or equal to a preset threshold, the initial cluster is designated as a middle cluster; and multiple candidate clusters are selected from the multiple middle clusters based on historical statistical data associated with at least one content tag in the multiple initial clusters.

[0067] In the example, the initial clusters are obtained by aggregating the distances between feature vectors. Feature vectors reflect the similarity of content tags at the literal level. While some content tags may contain the same terms representing the content format, their represented interest topics differ. Such content tags are easily aggregated into the same initial cluster. The more content tags in an initial cluster, the greater the possibility of multiple different interest topics being mixed within it. If such initial clusters are directly used as units for subsequent processing, their represented interest topics are not clearly defined, affecting the accuracy of subsequent processing results. Therefore, for initial clusters with more than a preset threshold of content tags, a larger model can be used to further divide them based on the interest topics represented by the content tags, making the resulting intermediate clusters more consistent in their interest topics. Specifically, for initial clusters with more than a preset threshold of content tags, all content tags and the first prompt word can be provided to the larger model to guide it in identifying the interest topics represented by each content tag. Content tags with consistent interest topics are grouped into the same content tag subset, and each content tag subset becomes an intermediate cluster. In this process, the first prompt word can instruct the large model to remove terms representing content form, such as analysis, explanation, popular science, review, tutorial, and evaluation, from the content tags, and then divide them according to the interest themes represented by the remaining parts. For example, although TCM analysis and film analysis use the same formal terminology, they represent different interest themes and should be classified into different content tag subsets. For the initial cluster with the number of content tags less than or equal to the preset threshold, the number of content tags is already limited, and the interest themes are usually relatively concentrated, so it can be directly used as an intermediate cluster.

[0068] In the example, historical statistics can refer to statistical information such as the display status of content tags over a previous period. Since the historical display data for some content tags is limited and lacks sufficient reference value, if such content tags account for a large proportion in a certain intermediate cluster, the interest topics represented by that intermediate cluster will be correspondingly less stable. Therefore, based on the historical statistics of the content tags contained in each intermediate cluster, multiple intermediate clusters can be further filtered, eliminating those with fewer historical statistics for their content tags, leaving the remaining intermediate clusters as candidate clusters.

[0069] Therefore, by using large models to process initial clusters of different sizes and combining them with the interest topics represented by content tags and historical statistical data for screening, we can reduce the impact of excessively large or unstable initial clusters on subsequent processing while maintaining the consistency of interest topics, thus providing a more reliable data foundation for evaluating the strength of associations between subsequent clusters.

[0070] In some embodiments, based on historical statistics associated with at least one content tag in a plurality of initial clusters, multiple candidate clusters are selected from a plurality of intermediate clusters, including: for each intermediate cluster, determining the proportion of the number of content tags in the intermediate cluster whose historical display count is less than or equal to a preset number to the total number of content tags in the intermediate cluster; and in response to the proportion being greater than the preset proportion, removing the intermediate cluster from the plurality of intermediate clusters to obtain a plurality of candidate clusters.

[0071] In the example, historical impression counts refer to the number of times the resource corresponding to the content tag was displayed within a previous period, which can be obtained from the historical impression logs of the recommendation system. The preset number of impressions can be used to determine whether the historical impression count of a content tag meets the requirements. For content tags with a historical impression count less than or equal to the preset number of impressions, it can be considered that the content tag lacks sufficient reference value; for example, the preset number of impressions could be 1. The preset proportion can be used to determine the proportion of such content tags in an intermediate cluster. Further, for each intermediate cluster, the number of content tags with a historical impression count less than or equal to the preset number of impressions can be counted separately, and this number can be compared with the total number of content tags in that intermediate cluster to obtain the corresponding proportion. If the proportion corresponding to a certain intermediate cluster is greater than the preset proportion, it indicates that content tags lacking reference value occupy a large proportion in that intermediate cluster, and the interest topics they represent are not stable enough. Therefore, this intermediate cluster is removed from multiple intermediate clusters, and the intermediate clusters that are not removed are considered as multiple candidate clusters. The specific values ​​of the preset number of impressions and the preset proportion can be set according to actual needs, and this disclosure does not limit them.

[0072] Therefore, by filtering intermediate clusters based on historical statistical data of content tags, it is possible to identify and remove intermediate clusters with a high proportion of low-frequency content tags, reduce the impact of low-frequency content tags on candidate clusters, and thus improve the effectiveness of candidate clusters.

[0073] Figure 3 A schematic diagram illustrating the determination of a plurality of candidate clusters according to an embodiment of the present disclosure is shown.

[0074] like Figure 3As shown, after clustering multiple content tags, a first initial cluster 310 and a second initial cluster 320 are obtained. For the first initial cluster 310, the number of content tags is less than or equal to a preset threshold, so no further subdivision is needed, and the first initial cluster 310 can be directly used as the first intermediate cluster 311. For the second initial cluster 320, the number of content tags is greater than the preset threshold. A first prompt word can be constructed and provided to the large model. The large model then divides the content tags in the second initial cluster 320 into multiple content tag subsets according to the interest topics represented by the content tags in the second initial cluster 320. Each content tag subset serves as an intermediate cluster, thus obtaining the second intermediate cluster 321 and the third intermediate cluster 322. Subsequently, based on historical statistical data associated with the content tags, the multiple intermediate clusters are further filtered. Specifically, for each intermediate cluster, the proportion of content tags with a historical display frequency less than or equal to a preset frequency can be determined relative to the total number of content tags in that intermediate cluster, and this proportion is compared with a preset proportion. The proportions corresponding to the first intermediate cluster 311 and the third intermediate cluster 322 are both less than the preset proportion, so they are retained. The proportion corresponding to the second intermediate cluster 321 is greater than the preset proportion, indicating that content tags with low historical display frequency account for a large proportion, so the second intermediate cluster 321 is removed from the multiple intermediate clusters. The retained first intermediate cluster 311 and third intermediate cluster 322 are used as multiple candidate clusters 330 for subsequent processing.

[0075] Understandable Figure 3 The illustration uses only two initial clusters as an example; however, the embodiments disclosed herein are not limited to this, and the number of initial clusters may not be limited to two.

[0076] In some embodiments, the input of the large model includes a second prompt word and a third prompt word, wherein the second prompt word is used to guide the large model to generate a topic description corresponding to each candidate cluster based on at least one content tag in the candidate cluster, and the third prompt word is used to guide the large model to evaluate the correlation strength between any two candidate clusters based on the topic description corresponding to each candidate cluster.

[0077] In the example, since a candidate cluster often contains a large number of content tags, providing each content tag to the large model during evaluation would require a significant amount of content. Therefore, it is necessary to extract a topic description for each candidate cluster. A topic description can be a short text summarizing the common interests pointed to by all content tags in a candidate cluster, used to extract the shared interest meaning behind the content tags. The second cue word is used to explain the generation requirements of the topic description to the large model. For example, the large model can be asked to summarize the topic description into expressions such as lifestyle, aesthetic preferences, knowledge direction, entertainment type, skill learning, or consumption scenarios, and avoid using overly broad terms such as general content, popular videos, or daily life. It can also guide the large model to find commonalities in terms of shared groups, scenarios, or motivations. For example, for a candidate cluster containing content tags such as paper cutting, origami, handicrafts, and painting, its topic description could be "handmade art creation."

[0078] In the example, the third cue word can be used to explain the evaluation requirements for the association strength to the large model, based on the topic descriptions corresponding to each candidate cluster. The topic descriptions of two candidate clusters can be provided to the large model, which is then asked to first extract the core interest topics from the topic descriptions, removing terms representing content forms such as explanation, documentary, popular science, tutorial, evaluation, review, sharing, and record. Then, the model should evaluate the following aspects in turn: whether the users browsing the content of both are likely to be the same group of people; whether their browsing motivations are similar, such as creative expression, aesthetic appreciation, knowledge exploration, practical improvement, or emotional satisfaction; whether the scenarios are consistent, such as leisure, learning, hands-on practice, or self-improvement; whether there is a natural interest extension path from one to the other, such as adjacent interest content, continuation of lifestyle, or transfer of skills; and whether there is a semantic relationship between the two, such as hierarchical, parallel, compositional, or tool-goal relationship. Based on this, the large model can be required to provide a corresponding correlation strength based on the evaluation of the above aspects, for example, expressed as a score between 0 and 1, with a higher score indicating a stronger interest correlation between the two. For two candidate clusters that only share the same terminology representing content form but differ in their core interest themes, or two candidate clusters belonging to unrelated domains and having no common ground in terms of demographics, motivations, scenarios, or interest expansion paths, a lower correlation strength should be assigned. Furthermore, the large model can be required to adopt a more concise output format, such as only outputting the candidate cluster pair's index, correlation strength, and a brief evaluation rationale.

[0079] Therefore, by guiding the large model to generate topic descriptions from the content tags in the candidate clusters and evaluating them accordingly, the large model can more accurately infer the association strength between the candidate clusters, thereby improving the accuracy of association strength evaluation.

[0080] In some embodiments, the large model includes a first large model and a second large model, and the association strength includes a first association strength and a second association strength. Based on the large model, the association strength between the two candidate clusters in each candidate cluster pair of at least one constructed candidate cluster pair is evaluated, including: based on the first large model, performing a preliminary evaluation of the first association strength between the two candidate clusters in each candidate cluster pair; based on the preliminary evaluation result, determining at least one cluster pair to be verified from at least one candidate cluster pair; and based on the second large model, performing a secondary evaluation of the second association strength between the two candidate clusters in each cluster pair to be verified.

[0081] In the example, since the number of candidate cluster pairs obtained by combining each candidate cluster is large, and only a portion of these pairs actually have an interest correlation, the accuracy of the correlation strength obtained by using only one large model for evaluation is insufficient. Therefore, a first large model can be used to perform a preliminary evaluation on all candidate cluster pairs, and then a second large model can be used for a secondary evaluation. The second large model can be the same as the first large model, or it can be a more powerful large model, thereby improving the accuracy of the correlation strength.

[0082] In the example, the first association strength refers to the association strength obtained by the first large model through preliminary evaluation, and the second association strength refers to the association strength obtained by the second large model through secondary evaluation. Both are used to characterize the degree of interest association between two candidate clusters. Specifically, the first large model can first evaluate the first association strength between the two candidate clusters in each candidate cluster pair, and determine the cluster pairs to be verified based on the obtained first association strength. The cluster pairs to be verified can refer to the candidate cluster pairs whose first association strength between the two candidate clusters reaches a certain level, thus requiring a second evaluation by the second large model. Subsequently, the second large model can evaluate the second association strength between the two candidate clusters in each cluster pair to be verified. After the above processing, the second association strength given by the second large model is used as the association strength between the two candidate clusters in the cluster pair to be verified for subsequent processing.

[0083] Therefore, by conducting a phased association strength assessment of candidate cluster pairs, first determining the cluster pairs to be verified based on the preliminary assessment results, and then conducting a secondary assessment of the cluster pairs to be verified, the reliability of the association strength assessment between candidate clusters can be improved.

[0084] In some embodiments, the preset conditions include a first preset condition and a second preset condition. Based on the evaluation results, determining at least one first target cluster pair that meets the preset conditions from at least one candidate cluster pair includes: based on the secondary evaluation results, selecting at least one first target cluster pair whose second association strength meets the second preset condition from at least one cluster pair to be verified, wherein at least one cluster pair to be verified is a candidate cluster pair whose first association strength meets the first preset condition from at least one candidate cluster pair.

[0085] In the example, the first and second preset conditions can be corresponding preset conditions related to the intensity threshold. After the initial evaluation, candidate cluster pairs with a first association strength greater than the intensity threshold corresponding to the first preset condition can be identified as cluster pairs to be verified, while candidate cluster pairs with an intensity strength less than or equal to the intensity threshold are not further processed. After the second evaluation, the cluster pairs to be verified with a second association strength greater than the intensity threshold corresponding to the second preset condition are identified as first target cluster pairs, thus obtaining at least one first target cluster pair. The thresholds corresponding to the above two preset conditions can be set separately, with the threshold corresponding to the first preset condition being relatively lower to avoid prematurely excluding candidate cluster pairs of interest.

[0086] Therefore, by setting first and second preset conditions, candidate cluster pairs can be screened in stages, and the first target cluster pair can be determined based on the results of the second evaluation, which can improve the accuracy of screening the correlation between candidate clusters.

[0087] Figure 4 A schematic diagram illustrating the determination of at least one first target cluster pair according to an embodiment of the present disclosure is shown.

[0088] like Figure 4 As shown, firstly, for each candidate cluster, a topic description corresponding to that candidate cluster can be generated based on at least one content tag in that candidate cluster. This can be used to construct a second prompt word and provide it to the large model, which then summarizes the common interest meaning pointed to by the content tags in that candidate cluster. Next, based on the first large model, a preliminary evaluation of the first association strength between the two candidate clusters in each candidate cluster pair is performed. This can be used to construct a third prompt word and provide it to the large model. Then, based on the preliminary evaluation results, at least one cluster pair to be verified is selected from at least one candidate cluster pair whose first association strength meets the first preset condition. Next, based on the second large model, a second evaluation of the second association strength between the two candidate clusters in each cluster pair to be verified is performed, again using the third prompt word. Finally, based on the second evaluation results, at least one first target cluster pair is selected from at least one cluster pair to be verified whose second association strength meets the second preset condition.

[0089] In some embodiments, determining the ranking of resources to be recommended based on the target association strength for resource recommendation includes: determining the weighting score corresponding to the resource to be recommended based on the target association strength; and adjusting the ranking score of the resource to be recommended based on the weighting score to determine the ranking of the resource to be recommended.

[0090] In the example, the weighted score can refer to the numerical value used to adjust the ranking score of the resource to be recommended, obtained based on the target association strength. The ranking score can refer to the score used to determine the order of the resource to be recommended in the recommendation results. For example, the target association strength can be multiplied by the user's interest intensity, time decay factor, and weighting coefficient of the content corresponding to the second candidate cluster, and the product is the weighted score corresponding to the resource to be recommended. Among them, the interest intensity can be calculated based on browsing, dwell time, and interaction data in historical operation data, the time decay factor is used to reflect the distance of the corresponding historical operation from the current time, and the weighting coefficient is used to control the magnitude of the adjustment. When there are multiple second candidate clusters that have interest associations with the first candidate cluster to which the resource to be recommended belongs, the corresponding values ​​of each can be obtained in the above manner, and then these values ​​can be weighted and summed. The result is the weighted score corresponding to the resource to be recommended, where the weight can be taken as the proportion of the corresponding second candidate cluster in the historical operation data.

[0091] In the example, after obtaining the weighted score, the ranking score of the resource to be recommended can be adjusted accordingly. Multiple weighting methods can be set in parallel during the recommendation phase. Besides the weighting score based on the target association strength, methods based on resource release time, resource quality, etc., can also be used. Each method provides a corresponding value for the same resource to be recommended. The ranking score of the resource to be recommended can be set to a base value of 1. This base value is multiplied sequentially by the weighted score and the values ​​given by the other weighting methods; the product is the adjusted ranking score of the resource. For example, if the weighted score of a resource to be recommended is 0.8, and the values ​​given by the other two weighting methods are 2 and 5 respectively, then its adjusted ranking score is 8. Subsequently, the ranking of each resource to be recommended is determined in descending order of its adjusted ranking score, and the recommendation results are generated and displayed to the user. Furthermore, the method of adjusting the weighted score is not limited to the multiplication method mentioned above; addition or other methods can also be used to adjust the ranking score.

[0092] Therefore, by determining the weighting score based on the target association strength and using this weighting score to adjust the ranking score of the recommended resources, the degree of association between resources can be reflected in the resource ranking process, realizing dynamic adjustment of the ranking of recommended resources. This allows for a moderate boosting of recommended resources with interest associations in the recommendation results, balancing the richness and stability of the recommendation results.

[0093] Figure 5 A schematic diagram illustrating the determination of the order of resources to be recommended according to an embodiment of the present disclosure is shown.

[0094] like Figure 5 As shown, on the one hand, the first candidate cluster to which the content tag associated with the resource to be recommended belongs can be determined first. That is, the content tag of the resource to be recommended can be obtained from the resource library, and then the candidate cluster to which it belongs can be found according to the correspondence between the content tag and the candidate cluster. On the other hand, the second candidate cluster to which the content tag associated with historical operation data belongs can be determined. Then, it is determined whether there is a second target cluster pair including the first candidate cluster and the second candidate cluster in at least one first target cluster pair. If there is, the association strength between the first candidate cluster and the second candidate cluster is determined as the target association strength. Then, the weighting score corresponding to the resource to be recommended is determined based on the target association strength. Finally, the ranking score of the resource to be recommended is adjusted based on the weighting score, and the order of each resource to be recommended is determined according to the order of the adjusted ranking scores from high to low. Based on this, the recommendation results are generated and displayed to the user.

[0095] In some embodiments, an agent is also provided, which recommends resources to the user according to the large model-based resource recommendation method described above.

[0096] In this example, the agent can be deployed in a resource recommendation system, integrating a processing module for executing resource recommendation methods based on large models. Upon receiving a resource recommendation request, the agent, based on previously mined pairs of first target clusters and their association strengths, determines the target association strength between the first candidate cluster to which the resource to be recommended belongs and the second candidate cluster corresponding to historical operation data. Based on this, it determines the ranking of the resource to be recommended and generates and outputs recommendation results according to the determined ranking. This ensures that the recommendation results not only match the user's existing interests but also cover other related interests, improving the richness and diversity of the recommendation results.

[0097] Figure 6 A structural block diagram of a resource recommendation device 600 based on a large model according to an embodiment of the present disclosure is shown.

[0098] like Figure 6As shown, the device 600 includes a content tag acquisition module 601, a data clustering processing module 602, an association strength evaluation module 603, a target cluster pair determination module 604, a candidate cluster determination module 605, a target strength determination module 606, and a resource ranking determination module 607. The content tag acquisition module 601 can be configured to acquire multiple content tags associated with candidate resources. The data clustering processing module 602 can be configured to perform clustering processing on the multiple content tags to obtain multiple candidate clusters, wherein each candidate cluster includes at least one content tag from the multiple content tags. The association strength evaluation module 603 can be configured to evaluate the association strength between the two candidate clusters in each candidate cluster pair among the at least one constructed candidate cluster pair based on a large model, wherein each candidate cluster pair includes any two candidate clusters from the multiple candidate clusters. The target cluster pair determination module 604 can be configured to determine at least one first target cluster pair that meets preset conditions from the at least one candidate cluster pair based on the evaluation results. The candidate cluster determination module 605 can be configured to determine a first candidate cluster to which the content tags associated with the resource to be recommended belong and a second candidate cluster to which the content tags associated with historical operation data belong. The target strength determination module 606 can be configured to, in response to determining that at least one first target cluster pair contains a second target cluster pair including the first and second candidate clusters, determine the association strength between the first and second candidate clusters as the target association strength. The resource ranking determination module 607 can be configured to determine the ranking of the resources to be recommended based on the target association strength for resource recommendation.

[0099] The operations of the above-mentioned content tag acquisition module 601, data cluster processing module 602, association strength evaluation module 603, target cluster pair determination module 604, candidate cluster determination module 605, target strength determination module 606, and resource ranking determination module 607 can respectively correspond to the following... Figure 2 The operations of steps S201, S202, S203, S204, S205, S206, and S207 are shown. Therefore, the details of each aspect will not be elaborated here.

[0100] Figure 7 A structural block diagram of a large-model-based resource recommendation device 700 according to another embodiment of the present disclosure is shown.

[0101] like Figure 7As shown, the device 700 may include a content tag acquisition module 701, a data clustering processing module 702, a correlation strength evaluation module 703, a target cluster pair determination module 704, a candidate cluster determination module 705, a target strength determination module 706, and a resource ranking determination module 707. The operations of the content tag acquisition module 701, data clustering processing module 702, correlation strength evaluation module 703, target cluster pair determination module 704, candidate cluster determination module 705, target strength determination module 706, and resource ranking determination module 707 can be coordinated with... Figure 6 The operations of the content tag acquisition module 601, data cluster processing module 602, association strength evaluation module 603, target cluster pair determination module 604, candidate cluster determination module 605, target strength determination module 606, and resource sorting determination module 607 are the same.

[0102] In some embodiments, the data clustering processing module 702 may include a feature vector conversion module 7021, an initial cluster determination module 7022, and an initial cluster filtering module 7023. The feature vector conversion module 7021 may be configured to convert multiple content tags into corresponding feature vectors. The initial cluster determination module 7022 may be configured to perform clustering based on the multiple feature vectors to obtain multiple initial clusters, wherein each initial cluster includes at least one content tag. The initial cluster filtering module 7023 may be configured to filter the multiple initial clusters based on at least one content tag from the multiple initial clusters to obtain multiple candidate clusters.

[0103] In some embodiments, the initial cluster filtering module 7023 may include a first intermediate cluster determination module 7023a, a second intermediate cluster determination module 7023b, and an intermediate cluster filtering module 7023c. The first intermediate cluster determination module 7023a may be configured to, for each initial cluster, construct a first prompt word to provide to a large model in response to the number of at least one content tag in the initial cluster being greater than a preset threshold. The first prompt word guides the large model to divide the at least one content tag in the initial cluster into multiple content tag subsets based on the interest topics represented by the at least one content tag in the initial cluster, wherein each content tag subset is considered an intermediate cluster. The second intermediate cluster determination module 7023b may be configured to, in response to the number of at least one content tag in the initial cluster being less than or equal to a preset threshold, designate the initial cluster as an intermediate cluster. The intermediate cluster filtering module 7023c may be configured to filter multiple candidate clusters from multiple intermediate clusters based on historical statistical data associated with at least one content tag in multiple initial clusters.

[0104] In some embodiments, the intermediate cluster filtering module 7023c may include a proportion determination module 7023c-1 and an intermediate cluster removal module 7023c-2. The proportion determination module 7023c-1 may be configured to, for each intermediate cluster, determine the proportion of content tags in that intermediate cluster whose historical display count is less than or equal to a preset number of times, relative to the total number of content tags in that intermediate cluster. The intermediate cluster removal module 7023c-2 may be configured to, in response to a proportion greater than a preset proportion, remove the intermediate cluster from multiple intermediate clusters to obtain multiple candidate clusters.

[0105] In some embodiments, the large model includes a first large model and a second large model, and the association strength includes a first association strength and a second association strength. The association strength evaluation module 703 may include a preliminary evaluation module 7031, a cluster pair determination module 7032, and a secondary evaluation module 7033. The preliminary evaluation module 7031 may be configured to perform a preliminary evaluation of the first association strength between two candidate clusters in each candidate cluster pair based on the first large model. The cluster pair determination module 7032 may be configured to determine at least one cluster pair to be verified from at least one candidate cluster pair based on the preliminary evaluation results. The secondary evaluation module 7033 may be configured to perform a secondary evaluation of the second association strength between two candidate clusters in each cluster pair to be verified based on the second large model.

[0106] In some embodiments, the preset conditions include a first preset condition and a second preset condition, and the target cluster pair determination module 704 may include a target cluster pair filtering module 7041. The target cluster pair filtering module 7041 may be configured to filter at least one first target cluster pair from at least one cluster pair to be verified based on the results of a secondary evaluation, wherein the at least one cluster pair to be verified is a candidate cluster pair among at least one candidate cluster pair whose first association strength meets the first preset condition.

[0107] In some embodiments, the resource ranking determination module 707 may include a weighting score determination module 7071 and a ranking score determination module 7072. The weighting score determination module 7071 may be configured to determine the weighting score corresponding to the resource to be recommended based on the target association strength. The ranking score determination module 7072 may be configured to adjust the ranking score of the resource to be recommended based on the weighting score, thereby determining the ranking of the resource to be recommended.

[0108] According to embodiments of this disclosure, an electronic device is also provided, including at least one processor; and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0109] According to embodiments of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is also provided, wherein the computer instructions are used to cause a computer to perform the methods described above.

[0110] According to embodiments of this disclosure, a computer program product is also provided, including a computer program, wherein the computer program implements the method described above when executed by a processor.

[0111] refer to Figure 8 The present invention describes a structural block diagram of an electronic device 800 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0112] like Figure 8 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0113] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, output unit 807, storage unit 808, and communication unit 809. Input unit 806 can be any type of device capable of inputting information to electronic device 800. Input unit 806 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device, and can include, but is not limited to, a mouse, keyboard, touchscreen, trackpad, trackball, joystick, microphone, and / or remote control. Output unit 807 can be any type of device capable of presenting information, and can include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 808 can include, but is not limited to, disk and optical disk. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and can include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth devices, 802.11 devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0114] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above. For example, in some embodiments, the method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform methods by any other suitable means (e.g., by means of firmware).

[0115] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0116] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0118] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0119] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

[0120] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0121] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0122] While embodiments or examples of this disclosure have been described with reference to the accompanying drawings, it should be understood that the methods, systems, and devices described above are merely exemplary embodiments or examples, and the scope of this disclosure is not limited by these embodiments or examples, but only by the granted claims and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. Furthermore, the steps may be performed in a different order than that described in this disclosure. Further, various elements in the embodiments or examples may be combined in various ways. Importantly, as technology evolves, many elements described herein can be replaced by equivalents that appear after this disclosure.

Claims

1. A resource recommendation method based on a large model, comprising: Retrieve multiple content tags associated with candidate resources; The plurality of content tags are clustered to obtain a plurality of candidate clusters, wherein each candidate cluster includes at least one content tag from the plurality of content tags; Based on the large model, the correlation strength between the two candidate clusters in each candidate cluster pair in at least one constructed candidate cluster pair is evaluated, wherein each candidate cluster pair includes any two candidate clusters from the plurality of candidate clusters; Based on the evaluation results, at least one first target cluster pair that meets the preset conditions is determined from the at least one candidate cluster pair; Determine the first candidate cluster to which the content tags associated with the resource to be recommended belong, and the second candidate cluster to which the content tags associated with historical operation data belong; In response to determining that a second target cluster pair including the first candidate cluster and the second candidate cluster exists in the at least one first target cluster pair, the association strength between the first candidate cluster and the second candidate cluster is determined as the target association strength; and Based on the target association strength, the ranking of the resources to be recommended is determined for resource recommendation.

2. The method according to claim 1, wherein, The clustering process of the multiple content tags to obtain multiple candidate clusters includes: The multiple content tags are converted into multiple corresponding feature vectors; Clustering is performed based on the multiple feature vectors to obtain multiple initial clusters, wherein each initial cluster includes at least one content tag; and Based on at least one content tag from the plurality of initial clusters, the plurality of initial clusters are filtered to obtain the plurality of candidate clusters.

3. The method according to claim 2, wherein, The step of filtering the initial clusters based on at least one content tag from the plurality of initial clusters to obtain the plurality of candidate clusters includes: For each initial cluster, in response to the number of at least one content tag in the initial cluster being greater than a preset threshold, a first prompt word is constructed to be provided to the large model, wherein the first prompt word is used to guide the large model to divide at least one content tag in the initial cluster into multiple content tag subsets based on the interest topics represented by at least one content tag in the initial cluster, wherein each content tag subset is an intermediate cluster; In response to the condition that the number of at least one content tag in the initial cluster is less than or equal to the preset threshold, the initial cluster is designated as an intermediate cluster; and The candidate clusters are selected from the intermediate clusters based on historical statistics associated with at least one content tag in the plurality of initial clusters.

4. The method according to claim 3, wherein, The step of filtering the multiple candidate clusters from multiple intermediate clusters based on historical statistical data associated with at least one content tag in the multiple initial clusters includes: For each intermediate cluster, determine the proportion of content tags in that intermediate cluster with a historical display count less than or equal to a preset count to the total number of content tags in that intermediate cluster; and In response to the ratio being greater than a preset ratio, the intermediate cluster is removed from the plurality of intermediate clusters to obtain the plurality of candidate clusters.

5. The method according to any one of claims 1 to 4, wherein, The input of the large model includes a second prompt word and a third prompt word. The second prompt word is used to guide the large model to generate a topic description corresponding to each candidate cluster based on at least one content tag in the candidate cluster. The third prompt word is used to guide the large model to evaluate the correlation strength between any two candidate clusters based on the topic description corresponding to each candidate cluster.

6. The method according to any one of claims 1 to 5, wherein, The large model includes a first large model and a second large model, and the association strength includes a first association strength and a second association strength. The evaluation of the association strength between the two candidate clusters in each of the at least one candidate cluster pair constructed based on the large model includes: Based on the first large model, a preliminary assessment of the first association strength between the two candidate clusters in each candidate cluster pair is performed; Based on the preliminary evaluation results, at least one cluster pair to be verified is determined from the at least one candidate cluster pair; and Based on the second major model, a second evaluation is performed on the second association strength between the two candidate clusters in each cluster pair to be verified.

7. The method according to claim 6, wherein, The preset conditions include a first preset condition and a second preset condition. The step of determining at least one first target cluster pair that meets the preset conditions from the at least one candidate cluster pair based on the evaluation results includes: Based on the secondary evaluation results, at least one first target cluster pair that meets the second preset condition is selected from the at least one cluster pair to be verified, wherein the at least one cluster pair to be verified is a candidate cluster pair among the at least one candidate cluster pair whose first association strength meets the first preset condition.

8. The method according to any one of claims 1 to 7, wherein, The step of determining the ranking of the resources to be recommended based on the target association strength for resource recommendation includes: Based on the target association strength, determine the weighting score corresponding to the resource to be recommended; and The ranking score of the resource to be recommended is adjusted based on the weighted score to determine the ranking of the resource to be recommended.

9. An intelligent agent that recommends resources to a user according to any one of claims 1 to 8.

10. A resource recommendation device based on a large model, comprising: The content tag acquisition module is configured to acquire multiple content tags associated with candidate resources; The data clustering processing module is configured to perform clustering processing on the plurality of content tags to obtain a plurality of candidate clusters, wherein each candidate cluster includes at least one content tag among the plurality of content tags; The association strength evaluation module is configured to evaluate the association strength between the two candidate clusters in each candidate cluster pair in at least one constructed candidate cluster pair based on a large model, wherein each candidate cluster pair includes any two candidate clusters from the plurality of candidate clusters; The target cluster pair determination module is configured to determine at least one first target cluster pair that meets preset conditions from the at least one candidate cluster pair based on the evaluation results; The candidate cluster determination module is configured to determine the first candidate cluster to which the content tags associated with the resource to be recommended belong and the second candidate cluster to which the content tags associated with historical operation data belong; The target strength determination module is configured to, in response to determining that a second target cluster pair including the first candidate cluster and the second candidate cluster exists in the at least one first target cluster pair, determine the association strength between the first candidate cluster and the second candidate cluster as the target association strength; and The resource ranking determination module is configured to determine the ranking of the resources to be recommended based on the target association strength in order to make resource recommendations.

11. An electronic device, comprising: At least one processor; as well as A memory that is communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the method according to any one of claims 1-8.

12. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-8.

13. A computer program product comprising a computer program, wherein, The computer program, when executed by a processor, implements the method according to any one of claims 1-8.

Images