Resource recommendation method and device based on large model, training method and device, equipment and medium

Abstract

The invention provides a resource recommendation method and device based on a large model, a training method and device, equipment and a medium, relates to the technical field of artificial intelligence, in particular to the technical fields of natural language processing, deep learning and the like, and can be used for application scenes such as generative retrieval, document intelligent editing, intelligent assistants, virtual assistants, intelligent e-commerce and the like. The resource recommendation method comprises the steps of obtaining a search query text of a user and user feature data; generating a user representation vector of the user based on the search query text and the user feature data; obtaining resource representation vectors of a plurality of candidate resources from a preset resource library; inputting the user representation vector and the resource representation vectors of the plurality of candidate resources into a scoring network to obtain respective comprehensive scores of the plurality of candidate resources; and sorting the plurality of candidate resources based on the comprehensive score, and displaying at least one candidate resource in the plurality of candidate resources to the user based on a sorting result.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to technical fields such as natural language processing and deep learning, and can be used in application scenarios such as generative retrieval, intelligent document editing, intelligent assistants, virtual assistants, and intelligent e-commerce. It specifically relates to a resource recommendation method based on a large model, a training method for a resource recommendation model, a resource recommendation device based on a large model, a training device for a resource recommendation model, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Artificial intelligence (AI) is the study of how computers can 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, specialized AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily encompass natural language processing, computer vision, speech recognition, machine learning / deep learning, big data processing, and knowledge graphs.

[0003] The approaches described in this section are not necessarily approaches that have been previously conceived or employed. Unless otherwise indicated, it should not be assumed that any approach described in this section is prior art simply by virtue of its inclusion in this section. Similarly, unless otherwise indicated, the issues raised in this section should not be considered as having been recognized in any prior art. Summary of the Invention

[0004] The present disclosure provides a resource recommendation method based on a large model, a training method for a resource recommendation model, a resource recommendation device based on a large model, a training device for a resource recommendation model, an electronic device, a computer-readable storage medium, and a computer program product.

[0005] According to one aspect of the present disclosure, a resource recommendation method based on a large model is provided, including: obtaining a user's search query text and user feature data; generating a user representation vector for the user based on the search query text and the user feature data; obtaining resource representation vectors of multiple candidate resources from a preset resource library; inputting the user representation vector and the resource representation vectors of the multiple candidate resources into a scoring network to obtain a comprehensive score for each of the multiple candidate resources; and sorting the multiple candidate resources based on the comprehensive score, and displaying at least one candidate resource from the multiple candidate resources to the user based on the sorting result.

[0006] According to another aspect of the present disclosure, a training method for a resource recommendation model is provided, comprising: obtaining a sample search query text and sample user feature data of a sample user; obtaining a sample resource representation vector of a plurality of sample candidate resources and a true value relevance score between the sample user and the plurality of sample candidate resources, wherein the sample resource representation vector is generated using a resource tower model; generating a sample user representation vector of the sample user based on the sample search query text and the sample user feature data using a user tower model; inputting the sample user representation vector and the sample resource representation vector of the plurality of sample candidate resources into a scoring network to obtain a predicted score for each of the plurality of sample candidate resources; and calculating a loss value based on the predicted score and the true value relevance score, and adjusting parameters of the user tower model, the resource tower model, and the scoring network based on the loss value.

[0007] According to another aspect of the present disclosure, a resource recommendation device based on a large model is provided, including: a first acquisition unit, configured to acquire a user's search query text and user feature data; a first generation unit, configured to generate a user representation vector for the user based on the search query text and the user feature data; a second acquisition unit, configured to acquire resource representation vectors of multiple candidate resources from a preset resource library; a first scoring unit, configured to input the user representation vector and the resource representation vectors of the multiple candidate resources into a scoring network to obtain a comprehensive score for each of the multiple candidate resources; and a sorting unit, configured to sort the multiple candidate resources based on the comprehensive score, and display at least one of the multiple candidate resources to the user based on the sorting result.

[0008] According to another aspect of the present disclosure, a training device for a resource recommendation model is provided, comprising: a third acquisition unit, configured to acquire a sample search query text and sample user feature data of a sample user; a fourth acquisition unit, configured to acquire a sample resource representation vector of a plurality of sample candidate resources and a true value correlation score between the sample user and the plurality of sample candidate resources, wherein the sample resource representation vector is generated using a resource tower model; a second generation unit, configured to generate a sample user representation vector of the sample user based on the sample search query text and the sample user feature data using a user tower model; a second scoring unit, configured to input the sample user representation vector and the sample resource representation vectors of the plurality of sample candidate resources into a scoring network to obtain a predicted score for each of the plurality of sample candidate resources; and a parameter adjustment unit, configured to calculate a loss value based on the predicted score and the true value correlation score, and adjust parameters of the user tower model, the resource tower model and the scoring network based on the loss value.

[0009] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the above-mentioned method.

[0010] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the above method.

[0011] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the above method when executed by a processor.

[0012] According to one or more embodiments of the present disclosure, the present disclosure generates a user representation vector based on user query text and user feature data, performs vectorized matching with a pre-generated resource representation vector, and then inputs the vector into a scoring network to calculate a comprehensive score, thereby providing high-quality resources to users in a search and recommendation fusion scenario and improving user experience.

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

[0014] The accompanying drawings illustrate exemplary embodiments and constitute a part of the specification. Together with the description of the specification, they serve to explain exemplary implementation 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 designate similar, but not necessarily identical, elements.

[0015] Figure 1 A schematic diagram illustrating an exemplary system in which the various methods described herein may be implemented according to an embodiment of the present disclosure;

[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 flowchart of generating a user representation vector of a user based on search query text and user feature data according to an embodiment of the present disclosure is shown;

[0018] Figure 4 A flowchart of determining resource representation vectors of multiple candidate resources according to an embodiment of the present disclosure is shown;

[0019] Figure 5 A schematic diagram of a resource recommendation model according to an embodiment of the present disclosure is shown;

[0020] Figure 6 A flowchart of a method for training a resource recommendation model according to an embodiment of the present disclosure is shown;

[0021] Figure 7 FIG2 shows a structural block diagram of a resource recommendation device based on a large model according to an embodiment of the present disclosure;

[0022] Figure 8 A structural block diagram of a training device for a resource recommendation model according to an embodiment of the present disclosure is shown; and

[0023] Figure 9 A structural block diagram of an exemplary electronic device that can be used to implement the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0024] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0025] In this disclosure, unless otherwise specified, the use of terms such as "first" and "second" to describe various elements is not intended to limit the positional relationship, temporal relationship, or importance relationship of these elements. Such terms are only used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of the element, while in some cases, based on the context of the description, they may also refer to different instances.

[0026] The terms used in the descriptions of various examples in this disclosure are for the purpose of describing specific examples only and are not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element may be one or more. In addition, the term "and / or" used in this disclosure encompasses any one and all possible combinations of the listed items.

[0027] In related technologies, in the search recommendation fusion scenario, users are less satisfied with the recommended resources.

[0028] To solve the above problems, the present invention generates a user representation vector based on user query text and user feature data, and then vectorizes and matches it with the pre-generated resource representation vector and inputs it into the scoring network to calculate the comprehensive score, so as to provide users with high-quality resources in the search recommendation fusion scenario and improve the user experience.

[0029] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0030] Figure 1 FIG2 is a schematic diagram of an exemplary system 100 in which the various methods and apparatuses described herein may be implemented according to an embodiment of the present disclosure. 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.

[0031] In an embodiment of the present disclosure, the server 120 may run one or more services or software applications that enable execution of the methods of the present disclosure.

[0032] In some embodiments, server 120 may also provide other services or software applications, which may include non-virtualized environments and virtualized environments. In some embodiments, these services may be provided as web-based services or cloud services, such as provided to users of client devices 101, 102, 103, 104, 105, and / or 106 under a software as a service (SaaS) model.

[0033] exist Figure 1 In the configuration shown, the server 120 may include one or more components that implement the functions performed by the server 120. These components may include software components, hardware components, or a combination thereof that can be executed by one or more processors. Users operating client devices 101, 102, 103, 104, 105, and / or 106 may, in turn, utilize one or more client applications to interact with the server 120 to utilize the services provided by these components. It should be understood that a variety of different system configurations are possible, which may differ from the system 100. Therefore, Figure 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

[0034] The user can use the client device 101, 102, 103, 104, 105 and / or 106 to perform human-computer interaction. The client device can provide an interface that enables the user of the client device to interact with the client device. The client device can also output information to the user via the interface. Figure 1 Only six client devices are depicted, but one skilled in the art will appreciate that the present disclosure can support any number of client devices.

[0035] 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, etc. These computer devices may 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, tablet computers, 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 a variety of different applications, such as various internet-related applications, communication applications (such as email applications), and short message service (SMS) applications, and may use various communication protocols.

[0036] The network 110 may be any type of network known to those skilled in the art that can support data communications using any of a variety of available protocols, including but not limited to TCP / IP, SNA, IPX, etc. By way of example only, the one or more networks 110 may be a local area network (LAN), an Ethernet-based network, a token ring, 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.

[0037] Server 120 may include one or more general-purpose computers, specialized 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 virtual operating systems, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that may be virtualized to maintain a server's virtual storage device). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0038] The computing units in the server 120 may run one or more operating systems including any of the operating systems described above as well as any commercially available server operating systems. The server 120 may also run any of a variety of additional server applications and / or middle-tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, and the like.

[0039] In some implementations, server 120 may include one or more applications to analyze and consolidate 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 the data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105, and / or 106.

[0040] In some embodiments, server 120 may be a distributed system server or a server integrated with blockchain. Server 120 may also be a cloud server, or an intelligent cloud computing server or intelligent cloud host equipped with artificial intelligence technology. A cloud server is a host product within the cloud computing service system that addresses the management difficulties and poor scalability of traditional physical hosts and virtual private servers (VPS) services.

[0041] The 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. The databases 130 may reside in a variety of locations. For example, the database used by the server 120 may be local to the server 120, or may be remote from the server 120 and communicate with the server 120 via a network-based or dedicated connection. The databases 130 may be of different types. In some embodiments, the databases used by the server 120 may be, for example, relational databases. One or more of these databases may store, update, and retrieve data to and from the databases in response to commands.

[0042] In some embodiments, one or more of the databases 130 may also be used by applications to store application data. The databases used by the applications may be different types of databases, such as a key-value store, an object store, or a conventional store backed by a file system.

[0043] Figure 1 The system 100 may be configured and operated in various ways to enable application of the various methods and apparatuses described in accordance with the present disclosure.

[0044] According to one aspect of the present disclosure, a resource recommendation method based on a large model is provided. Figure 2 As shown, the method includes: step S201, obtaining the user's search query text and user feature data; step S202, generating the user representation vector of the user based on the search query text and the user feature data; step S203, obtaining the resource representation vectors of multiple candidate resources from a preset resource library; step S204, inputting the user representation vector and the resource representation vectors of the multiple candidate resources into a scoring network to obtain a comprehensive score of each of the multiple candidate resources; and step S205, sorting the multiple candidate resources based on the comprehensive score, and displaying at least one candidate resource from the multiple candidate resources to the user based on the sorting result.

[0045] Therefore, by jointly generating a user representation vector based on the user query text and user feature data, and vectorizing and matching it with the pre-generated resource representation vector, the vector is input into the scoring network to calculate the comprehensive score, so that high-quality resources can be provided to users in the search recommendation fusion scenario, thereby improving the user experience.

[0046] The above resource recommendation method can be used in the "search-after-search" scenario of search recommendation fusion, that is, after the user completes a search behavior, the system recommends relevant content in real time based on their search intention and personalized interests.

[0047] In step S201, the user enters a search query and generates a search action (e.g., browsing a search results page). The system captures the search query, user profile, and user behavior characteristics, and then enters the recall model. In some embodiments, the disclosed solution is not used to obtain search results, but rather to present recommended content on the search interface, as described below.

[0048] According to some embodiments, user feature data may include: user basic features, which are based on the user's portrait attributes; short-term interest features, which are based on the user's recent search and browsing behavior; and long-term interest features, which are based on the user's long-term search and browsing behavior.

[0049] By simultaneously incorporating three types of features—profile attributes, recent real-time behavior, and long-term historical behavior—the system can capture short-term and long-term trends based on static interest profiles, providing a granular depiction of user intent. This approach yields more accurate candidate resources, more reliable ranking, and proactively addresses potential user needs.

[0050] In some embodiments, the user's portrait attributes may include gender, age, region, device, and other attributes, and may also include other attributes.

[0051] In some embodiments, a user's search and browsing behavior may include clicks, favorites, and forwarding. "Recent" and long-term browsing behavior can be divided according to different time windows. In an exemplary embodiment, recent search and browsing behavior may refer to a sequence of resources browsed within a recent preset number of days (e.g., 7-14 days), while long-term search and browsing behavior may refer to a sequence of resources browsed within the past two years.

[0052] The resources disclosed herein may be content resources such as short videos, images, texts, audios, etc., and may also include other types of resources, which are not limited here.

[0053] According to some embodiments, Figure 3 As shown, step S202, generating a user representation vector for a user based on the search query text and the user feature data, may include: step S301, performing semantic feature extraction on the search query text after segmenting the words to obtain a query semantic vector; step S302, performing feature embedding on the user feature data to obtain a user feature vector; and step S303, combining the query semantic vector and the user feature vector and inputting them into a first neural network to obtain a user representation vector.

[0054] Therefore, through the above method, the search query text and user feature data can be effectively integrated to obtain a user representation vector that covers more comprehensive user information.

[0055] In step S301, after the search query text is segmented, a multi-layer Transformer structure may be used to extract semantic features from the segmented sequence to obtain high-dimensional semantic information, namely, a query semantic vector.

[0056] In step S302, an embedding model may be used to perform feature embedding on the user feature data to obtain a user feature vector.

[0057] In some embodiments, the user basic features, short-term interest features, and long-term interest features mentioned above may be respectively embedded to obtain feature vectors corresponding to each feature.

[0058] In step S303, the query semantic vector and the user feature vector may be concatenated to obtain a user feature vector, and then the user feature vector may be input into a first neural network (eg, MLP).

[0059] In some embodiments, the feature vectors and query semantic vectors corresponding to the user's basic features, short-term interest features, and long-term interest features can be spliced ​​and input into the MLP, so as to uniformly model the user's personalized expression under the query in the current search scenario and improve the final recommendation effect.

[0060] The operation of step S202 can be completed by using the user tower model in the data recommendation model. That is, the user tower model can complete multiple tasks such as feature embedding, semantic feature extraction, feature combination, etc., and can include a first neural network.

[0061] Back to Figure 2 In step S203, resource representation vectors of multiple candidate resources are obtained from a preset resource library.

[0062] In some embodiments, a preset resource library containing a large number of resources can be constructed before executing the method of the present disclosure. Figure 4 As shown, the resource representation vectors of multiple candidate resources are predetermined by the following operations: step S401, feature embedding the basic information of the corresponding resource to obtain a resource basic feature vector; step S402, feature embedding the multimodal data in the corresponding resource to obtain a multimodal feature vector; and step S403, combining the resource basic feature vector and the multimodal feature vector and inputting them into the second neural network to obtain the resource representation vector of the corresponding resource.

[0063] Therefore, by fusing the structured basic information and multimodal features of resources, a comprehensive resource representation vector of multidimensional signals is constructed, enabling the model to more accurately distinguish fine-grained differences between resources and explore the potential correlation between resources and users, thereby effectively improving the recommendation effect.

[0064] In some embodiments, the basic information of the resource may include structured basic information, such as title, tag, author, creation time, etc. Multimodal data may include different categories of data such as images, audio, and video.

[0065] In some embodiments, an embedding model may be used to perform feature embedding on basic information and multimodal data of a resource.

[0066] The construction operation of the preset database can be completed by using the resource tower model in the data recommendation model. That is to say, the resource tower model can complete multiple tasks such as feature embedding and feature combination.

[0067] According to some embodiments, step S203, obtaining resource representation vectors of multiple candidate resources from a preset resource library may include: calculating the similarity between the resource representation vector of each resource in the preset resource library and the user representation vector; and determining multiple resources with the highest similarity to the user representation vector as multiple candidate resources.

[0068] Thus, through the above method, it is possible to quickly filter out some content that users may be interested in from a large amount of resources. These resources can then be used as candidate resources and refined scoring and sorting can be performed, thereby effectively improving recommendation efficiency.

[0069] In some embodiments, cosine similarity or other metrics may be used to calculate the similarity between the user representation vector and the resource representation vectors of each resource in the preset resource library.

[0070] In step S204, the user representation vector and the resource representation vectors of the multiple candidate resources are input into a scoring network to obtain comprehensive scores of the multiple candidate resources.

[0071] In some embodiments, a pre-trained scoring network can be used to score candidate resources to indicate the user's interest in the resource. The scoring network can be part of the recommended resource model. An exemplary training method for the scoring network (and the entire recommended resource model) is described below.

[0072] According to some embodiments, the scoring network may include a relevance scoring network and a consumption index prediction network. Step S204, inputting the user representation vector and the resource representation vectors of the multiple candidate resources into the scoring network to obtain comprehensive scores for each of the multiple candidate resources, may include: utilizing the relevance scoring network to obtain relevance scores for the multiple candidate resources, where the relevance scores represent the relevance between the corresponding candidate resources and the user; utilizing the consumption index prediction network to obtain consumption indexes for the multiple candidate resources, where the consumption indexes represent the user's expected consumption behavior for the corresponding candidate resources; and obtaining comprehensive scores for the multiple candidate resources based on the relevance scores and consumption indexes of the multiple candidate resources.

[0073] Therefore, by simultaneously incorporating relevance scores and consumption indicator predictions into a two-dimensional assessment of candidate resources, we can more comprehensively characterize user interests and consumption tendencies. Deep semantic matching based on relevance, combined with anticipated consumer behaviors such as clicks, stays, and completions, allows for a refined understanding of user preferences. This approach ensures that recommendations are closely aligned with potential user needs, effectively improving recommendation effectiveness.

[0074] In some embodiments, the relevance scoring network serves as a subnetwork of the scoring network, and its input includes a user representation vector, a resource representation vector of a candidate resource, and optional other features (e.g., contextual features). After being processed by a multi-layer neural network, it outputs a relevance score between the corresponding resource and the user. The relevance score can be used to characterize the degree of match between the candidate resource and the user's search intent or potential interest. The score reflects the degree of match between the candidate resource and the user's search intent and historical behavior. Unlike the coarse recall using similarity in the previous step, the relevance scoring network can fuse multi-dimensional features and capture deep semantic associations through learnable nonlinear mapping. The use of two successive metrics (i.e., similarity and relevance scoring) can take into account both the rapid screening of massive resources and high-precision sorting within multiple candidate resources, taking into account both system performance and recommendation quality.

[0075] In some embodiments, the consumption index prediction network serves as another subnetwork of the scoring network. Its input includes the user representation vector, the resource representation vector of the candidate resource, and other optional features (e.g., contextual features). After being processed by a multi-layer neural network, it outputs the prediction results of the user's behavior on the resource.

[0076] According to some embodiments, the consumption indicator prediction network may include at least one of a click-through rate prediction network, a dwell time prediction network, and a completion rate prediction network. The consumption indicator may correspondingly include at least one of click-through rate, estimated dwell time, and completion rate. These indicators can more intuitively reflect user preference for resources, thereby improving recommendation conversion rates.

[0077] In some embodiments, the consumption index prediction network may include multiple output branch networks. Specifically, the click-through rate prediction network may use a fully connected layer and Sigmoid activation to output predicted click probability; the dwell time prediction network may use a fully connected layer and linear activation to output predicted dwell time; and the completion rate prediction network may use a fully connected layer and Sigmoid activation to output predicted completion probability.

[0078] These output branch networks can use corresponding loss functions for supervised learning, such as cross-entropy loss for the click-through rate branch, mean squared error loss for the dwell time branch, and cross-entropy loss for the completion rate branch. By modeling the complex interactions between user historical behavior and resource representation vectors, the consumption index prediction network can characterize users' actual consumption tendencies for candidate resources, providing rich behavioral prediction signals for comprehensive scoring. This complements the deep semantic matching results of the relevance scoring network, jointly driving the search recommendation system to improve user click and dwell rates while ensuring recall accuracy.

[0079] It is understandable that, in addition to the above three consumption indicators and the corresponding output branch networks, output branch networks for predicting other consumption indicators, such as attention, collection, etc., can also be used.

[0080] Figure 5 FIG. 1 shows a schematic diagram of a resource recommendation model according to an exemplary embodiment of the present disclosure. Figure 5 As shown, the input of the user tower model includes basic user features (i.e., the basic user features mentioned above), user interest features (including the short-term interest features and long-term interest features mentioned above), and query segmentation (i.e., the segmentation results of the search query text mentioned above). Among them, query segmentation is processed by a Transformer structure model and then spliced ​​with the basic user features and user interest features. The spliced ​​result is processed by MLP to obtain a user representation vector. The input of the resource tower model includes basic resource features (i.e., the resource basic feature vector mentioned above) and multimodal resource features (i.e., the multimodal feature vector mentioned above). After splicing, the two are input and processed by MLP to obtain a resource representation vector. The user representation vector and the resource representation vector of each candidate resource are spliced ​​and input into the multi-objective score (Score) network (i.e., the scoring network mentioned above) to obtain a comprehensive score.

[0081] According to some embodiments, sorting multiple candidate resources based on comprehensive scores and displaying at least one of the multiple candidate resources to the user based on the sorting result may include: displaying at least one candidate resource in the first screen position in the information flow area of ​​the search results page.

[0082] In some embodiments, for the search recommendation fusion scenario, an information flow area can be presented in the search results page, so that users can consume information flow resources while obtaining search results, meeting the user's various needs. The resource recommendation method proposed in the present disclosure can be used in the recall stage. In addition to the at least one candidate resource obtained in step S205, multiple other resources obtained by other recall paths can also be obtained. Since the resource recommendation method proposed in the present disclosure fully considers individual difference factors such as user interests and preferences, and integrates the user's search query text in the search scenario, and then uses relevance scores and consumption indicators for scoring during sorting, it can ultimately recall resources that accurately hit the user's interests and needs. Placing these resources on the first screen in the information flow area for display to the user can further enhance the user experience.

[0083] According to some embodiments, the resource recommendation method may further include: recording user behavior of at least one candidate resource, wherein the user behavior may include at least one of clicks, dwell time, and completion of broadcast, and the user behavior data is used to optimize the consumption index prediction network.

[0084] After displaying resources to users, user behavior data such as clicks and stays can be recorded and used as part of the next round of training samples for the model or neural network, achieving feedback-driven end-to-end optimization. This approach allows the model or neural network to automatically learn matching logic and continuously improve recommendation results.

[0085] According to another aspect of the present disclosure, a method for training a resource recommendation model is provided. Figure 6 As shown, the training method includes: step S601, obtaining a sample search query text and sample user feature data of a sample user; step S602, obtaining a sample resource representation vector of multiple sample candidate resources and a true value correlation score between the sample user and the multiple sample candidate resources, wherein the sample resource representation vector is generated using a resource tower model; step S603, using a user tower model, based on the sample search query text and the sample user feature data, generating a sample user representation vector of the sample user; step S604, inputting the sample user representation vector and the sample resource representation vector of the multiple sample candidate resources into a scoring network to obtain a predicted score for each of the multiple sample candidate resources; and step S605, calculating a loss value based on the predicted score and the true value correlation score, and adjusting the parameters of the user tower model, the resource tower model and the scoring network based on the loss value.

[0086] It is understandable that Figure 6 Part of the operations of step S601 to step S605 can refer to the above description of Figure 2 The description of steps S201 to S205 in the process will not be repeated here.

[0087] In some embodiments, the relevant content and ground truth results of the sample user and multiple sample candidate resources obtained in steps S601-S602 can be collectively referred to as sample data. Sample data can include positive samples and negative samples. Positive samples can be constructed based on the actual consumption behavior of the sample user. For example, resources clicked by the sample user can be used as candidate resources, and a supervision signal can be generated based on the actual browsing behavior (e.g., length of stay, whether the broadcast is completed, etc.).

[0088] The true value results of the samples (for example, the true value relevance score) can be obtained through manual labeling, or can be obtained by statistics based on indicators such as user click behavior, bounce rate, and stay time in historical logs, or can be obtained through other methods, which are not limited here.

[0089] In some embodiments, the operations of step S603 and step S604 may refer to the above description of the operations of step S203 and step S204.

[0090] According to some embodiments, the scoring network may include a relevance scoring network and a consumption index prediction network. The training method may further include: obtaining the true user behavior of the sample user on multiple sample candidate resources. Step S604, inputting the sample user representation vector and the sample resource representation vectors of the multiple sample candidate resources into the scoring network, and obtaining the predicted scores of the multiple sample candidate resources may include: using the relevance scoring network to obtain the predicted relevance scores of the multiple sample candidate resources, the predicted relevance scores representing the relevance between the corresponding sample candidate resources and the sample user; and using the consumption index prediction network to obtain the consumption index of the multiple sample candidate resources, the consumption index representing the expected consumption behavior of the sample user on the corresponding sample candidate resources.

[0091] Therefore, by evaluating from two aspects, relevance and consumption indicators, the model's perception of the user's personalized information can be improved, so that the model can recommend resources that the user is more interested in and improve the user experience.

[0092] According to some embodiments, step S605, calculating the loss value based on the predicted score and the true value relevance score, and adjusting the parameters of the scoring network based on the loss value may include: calculating a first loss value based on the relevance score and the true value relevance score; calculating a second loss value based on the consumption index and the true value user behavior; and adjusting the parameters of the user tower model, the resource tower model and the scoring network based on the first loss value and the second loss value.

[0093] Therefore, by incorporating users' real consumption behavior into model training, the model's ability to identify users' interests and potential needs can be significantly enhanced.

[0094] In some embodiments, the first loss value can be calculated based on the predicted relevance score output by the relevance scoring network for the i-th candidate resource and its corresponding true value relevance score using the cross entropy loss function. Then, for the consumption index prediction network, the second loss value can be calculated using the cross entropy or mean square error as the loss function. If the consumption index prediction network includes multiple branch networks, the loss of each branch can be calculated separately and summarized according to the preset weights. Furthermore, the first loss value and the second loss value can be combined into a total loss according to the weighted coefficient, and the gradients of all learnable parameters in the user tower, resource tower and scoring network are calculated simultaneously through the back propagation algorithm, and an optimizer such as Adam (which can set group learning rates for different sub-networks) is used for a one-time synchronous update to achieve end-to-end joint optimization.

[0095] According to some embodiments, the consumption indicator prediction network includes at least one of a click-through rate prediction network, a dwell time prediction network, and a completion rate prediction network, and the true value user behavior correspondingly includes at least one of a true value click behavior, a true value dwell time, and a true value completion behavior.

[0096] In some embodiments, each output branch network of the consumption indicator prediction network calculates a cross-entropy loss or a mean squared error loss based on the corresponding true value user behavior label (e.g., a binary label of click / no click, a real value of dwell time, and a binary label of completed / incomplete). Specifically, the click-through rate prediction network and the completion rate prediction network can use a cross-entropy loss, and the dwell time prediction network can use a mean squared error loss.

[0097] In some embodiments, the true value user behavior can be determined in the following ways. For the true value click behavior, if the user clicks on the resource within a preset time limit (for example, 10 seconds) after the resource is displayed to the user, it is marked as "1", otherwise it is marked as "0". For the true value stay time, the cumulative number of seconds from the time the user clicks to enter the resource details page (or starts playing) to the time the user leaves the details page is the true value stay time. For the true value completion behavior, when the user's playback progress reaches the preset end point percentage (for example, ≥90%), it is marked as "1" (completed); otherwise it is marked as "0" (not completed). In the above manner, clicks, stay time and completion labels can be accurately extracted from offline logs or online buried points to form a reliable true value supervision signal for consumption indicator prediction branches.

[0098] In some embodiments, the output branch networks of the consumption index prediction network may be trained using the same loss function, which is as follows:

[0099]

[0100] Among them, K means that the consumption index prediction network has K training targets (i.e., K consumption indicators or K branches), and its corresponding weight coefficient is α k The loss value calculated using this loss function can be the second loss value mentioned above. The weight coefficients corresponding to each target can be dynamically adjusted through grid search or based on gradient normalization (such as GradNorm) to balance the contribution of each target to the final model performance.

[0101] According to some embodiments, step S601, obtaining sample search query text and sample user feature data of sample users may include: sampling target candidate resources from resources displayed to sample users; using a content understanding model to extract keywords from the target candidate resources to obtain multiple keywords and their scores; sorting the multiple keywords based on the scores of the multiple keywords, and determining one or more target keywords from the multiple keywords based on the sorting results; and constructing a sample search query text based on the one or more target keywords.

[0102] In some embodiments, the system first randomly extracts resources from the historical display data of the information flow scene (for example, the main feed flow) as target candidate resources. Subsequently, the content understanding model is called to perform semantic analysis on the title and text of each target candidate resource and output a set of keywords and their matching scores. The content understanding model can use a BERT encoder, a TextRank algorithm, other models, or any combination of these models. The system sorts the keyword set in descending order based on the matching score, and selects the top N keywords (or interest tags, attention words) as target keywords based on a preset threshold. Finally, the target keywords are spliced ​​into a word sequence in a predetermined order to construct a pseudo query paired with the corresponding resource, that is, a sample search query text. By incorporating this pseudo query into training together with the real search query, the query sample size can be significantly expanded and the representation learning between the query and the resource can be enhanced.

[0103] In some embodiments, a hierarchical strategy can be used to construct negative samples during the training phase. First, for prediction tasks targeting user consumption behaviors (e.g., clicks, stays, or completions), the system automatically identifies consumption negative samples based on actual behavior labels: when a user does not perform the target behavior on a resource, the resource is labeled as a consumption negative sample for that user. Second, to enhance the ability to discriminate relevance in retrieval, the system further generates two types of relevance negative samples within each training batch: in-batch negative samples and hard negative samples.

[0104] An exemplary method for obtaining negative samples within a batch is: in the same training batch, when a sample is considered a positive sample, the remaining samples serve as negative samples corresponding to the positive sample, thereby significantly expanding the number of negative samples without the need for additional calculations.

[0105] An exemplary method for obtaining hard negative samples is to first determine a set of similar resources based on the type of positive sample resources (e.g., celebrities, sports, music, etc.). From this set, the system then randomly selects a one-to-one correspondence equal to the number of positive samples to form hard negative samples, thereby increasing the difficulty of sample differentiation. By collaboratively designing real-world negative samples, within-batch negative sampling, and hard negative sampling, the model obtains a rich and moderately difficult negative control in both user consumption prediction and relevance matching, effectively improving the final discrimination accuracy and robustness.

[0106] According to another aspect of the present disclosure, a resource recommendation device based on a large model is provided. Figure 7 As shown, the device 700 includes: a first acquisition unit 710, configured to acquire a user's search query text and user feature data; a first generation unit 720, configured to generate a user representation vector for the user based on the search query text and the user feature data; a second acquisition unit 730, configured to acquire resource representation vectors of multiple candidate resources from a preset resource library; a first scoring unit 740, configured to input the user representation vector and the resource representation vectors of the multiple candidate resources into a scoring network to obtain a comprehensive score for each of the multiple candidate resources; and a sorting unit 750, configured to sort the multiple candidate resources based on the comprehensive score, and display at least one of the multiple candidate resources to the user based on the sorting result.

[0107] It can be understood that the operations and effects of units 710 to 750 in the device 700 can refer to the above description of steps S201 to S205, and are not repeated here.

[0108] According to another aspect of the present disclosure, a training device for a resource recommendation model is provided. Figure 8As shown, the device 800 includes: a third acquisition unit 810, configured to obtain a sample search query text and sample user feature data of a sample user; a fourth acquisition unit 820, configured to obtain a sample resource representation vector of multiple sample candidate resources and a true value correlation score between the sample user and the multiple sample candidate resources, wherein the sample resource representation vector is generated using a resource tower model; a second generation unit 830, configured to generate a sample user representation vector of the sample user based on the sample search query text and the sample user feature data using a user tower model; a second scoring unit 840, configured to input the sample user representation vector and the sample resource representation vector of the multiple sample candidate resources into a scoring network to obtain a predicted score for each of the multiple sample candidate resources; and a parameter adjustment unit 850, configured to calculate a loss value based on the predicted score and the true value correlation score, and adjust the parameters of the user tower model, the resource tower model and the scoring network based on the loss value.

[0109] It can be understood that the operations and effects of units 810 to 850 in the apparatus 800 may refer to the above description of steps S601 to S605 .

[0110] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0111] According to an embodiment of the present disclosure, an electronic device, a readable storage medium, and a computer program product are also provided.

[0112] refer to Figure 9 , a block diagram of an electronic device 900 that can serve as a server or client of the present disclosure will now be described, 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 processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0113] like Figure 9As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the electronic device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0114] Multiple components within electronic device 900 are connected to I / O interface 905, including an input unit 906, an output unit 907, a storage unit 908, and a communication unit 909. Input unit 906 can be any type of device capable of inputting information into electronic device 900. Input unit 906 can receive input numeric or character information and generate key signal input related to user settings and / or function control of the electronic device. It may include, but is not limited to, a mouse, keyboard, touch screen, trackpad, trackball, joystick, microphone, and / or remote control. Output unit 907 can be any type of device capable of presenting information, and may include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. Storage unit 908 may include, but is not limited to, a magnetic disk or an optical disk. Communication unit 909 allows electronic device 900 to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks. It may include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or chipset, such as a Bluetooth device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0115] The computing unit 901 can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods, processes, and / or processes described above. For example, in some embodiments, these methods, processes, and / or processes can be implemented as computer software programs that are tangibly contained in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the methods, processes, and / or processes described above can be performed. Alternatively, in other embodiments, the computing unit 901 can be configured to perform these methods, processes, and / or processes in any other appropriate manner (e.g., by means of firmware).

[0116] Various embodiments of the systems and techniques described above 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), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0117] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0118] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0119] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0120] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0121] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. The client-server relationship is established by computer programs running on the respective computers and establishing a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service system that addresses the management difficulties and poor business scalability of traditional physical hosts and VPS services ("Virtual Private Servers," or simply "VPS"). The server may also be a server in a distributed system or a server integrated with blockchain.

[0122] It should be understood that the various forms of the processes shown above can be used to reorder, 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 results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0123] Although the embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above-mentioned methods, systems and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but is only limited by the claims after authorization and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. In addition, the steps may be performed in an order different from that described in this disclosure. Further, the various elements in the embodiments or examples may be combined in various ways. It is important that as technology evolves, many of the elements described herein may be replaced by equivalent elements that appear after this disclosure.

Claims

1. A resource recommendation method based on a large model, comprising: Obtain the user's search query text and user feature data; generating a user representation vector for the user based on the search query text and the user feature data; Obtain resource representation vectors of multiple candidate resources from a preset resource library; Inputting the user representation vector and the resource representation vectors of the multiple candidate resources into a scoring network to obtain a comprehensive score for each of the multiple candidate resources; as well as The plurality of candidate resources are sorted based on the comprehensive scores, and at least one candidate resource among the plurality of candidate resources is displayed to the user based on the sorting result.

2. The method according to claim 1, wherein Generating a user representation vector of the user based on the search query text and the user feature data includes: After segmenting the search query text, perform semantic feature extraction to obtain a query semantic vector; Embedding the user feature data to obtain a user feature vector; and The query semantic vector and the user feature vector are combined and input into a first neural network to obtain the user representation vector.

3. The method according to claim 2, wherein: The user characteristic data includes: User basic features, which are based on the user's profile attributes; Short-term interest features, the short-term interest features are based on the user's recent search and browsing behavior; and Long-term interest features, where the long-term interest features are based on the user's long-term search and browsing behavior.

4. The method according to any one of claims 1 to 3, wherein The scoring network includes a relevance scoring network and a consumption index prediction network. Inputting the user representation vector and the resource representation vectors of the plurality of candidate resources into the scoring network to obtain the comprehensive scores of the plurality of candidate resources includes: Obtaining relevance scores of the plurality of candidate resources using the relevance scoring network, wherein the relevance scores represent relevance between the corresponding candidate resources and the user; Obtaining consumption indices of the plurality of candidate resources using the consumption indices prediction network, wherein the consumption indices represent the user's expected consumption behavior for the corresponding candidate resources; and Based on the relevance scores and consumption indicators of the multiple candidate resources, a comprehensive score of the multiple candidate resources is obtained.

5. The method according to claim 4, wherein The consumption index prediction network includes at least one of a click rate prediction network, a stay time prediction network and a completion rate prediction network, and the consumption index correspondingly includes at least one of a click rate, an estimated stay time and a completion rate.

6. The method according to claim 4, wherein: The step of obtaining resource representation vectors of multiple candidate resources from a preset resource library includes: Calculating the similarity between the resource representation vector of each resource in the preset resource library and the user representation vector; and A plurality of resources having the highest similarity to the user representation vector are determined as the plurality of candidate resources.

7. The method according to any one of claims 1 to 3, wherein The resource representation vectors of the plurality of candidate resources are predetermined by the following operations: Embed the basic information of the corresponding resources into features to obtain the resource basic feature vector; Performing feature embedding on the multimodal data in the corresponding resource to obtain a multimodal feature vector; as well as The resource basic feature vector and the multimodal feature vector are combined and input into a second neural network to obtain a resource representation vector of the corresponding resource.

8. The method according to any one of claims 1 to 3, wherein The sorting of the plurality of candidate resources based on the comprehensive scores, and presenting at least one of the plurality of candidate resources to the user based on the sorting result includes: In the information flow area of ​​the search results page, the at least one candidate resource is displayed at the home screen position.

9. The method according to claim 5, further comprising: The user behavior of the user on the at least one candidate resource is recorded, wherein the user behavior includes at least one of clicks, stay time, and completion of broadcasting, and the user behavior data is used to optimize the consumption index prediction network.

10. A method for training a resource recommendation model, comprising: Obtaining sample search query text and sample user feature data of sample users; Obtaining sample resource representation vectors of a plurality of sample candidate resources and true value relevance scores between the sample user and the plurality of sample candidate resources, wherein the sample resource representation vectors are generated using a resource tower model; generating a sample user representation vector of the sample user based on the sample search query text and the sample user feature data using a user tower model; Inputting the sample user representation vector and the sample resource representation vectors of the multiple sample candidate resources into a scoring network to obtain a predicted score for each of the multiple sample candidate resources; as well as A loss value is calculated based on the predicted score and the true value relevance score, and parameters of the user tower model, the resource tower model, and the scoring network are adjusted based on the loss value.

11. The method according to claim 10, wherein: The scoring network includes a relevance scoring network and a consumption index prediction network, and the method further includes: Obtaining the true user behavior of the sample user on the multiple sample candidate resources, The step of inputting the sample user representation vector and the sample resource representation vectors of the plurality of sample candidate resources into a scoring network to obtain the predicted scores of the plurality of sample candidate resources comprises: Obtaining predicted relevance scores of the plurality of sample candidate resources using the relevance scoring network, the predicted relevance scores representing relevance between corresponding sample candidate resources and the sample user; and The consumption index prediction network is used to obtain consumption indexes of the plurality of sample candidate resources, where the consumption indexes represent the expected consumption behaviors of the sample users on the corresponding sample candidate resources.

12. The method according to claim 11, wherein Calculating a loss value based on the predicted score and the true value relevance score, and adjusting parameters of the scoring network based on the loss value includes: Calculating a first loss value based on the relevance score and the true value relevance score; Calculating a second loss value based on the consumption index and the true user behavior; and Based on the first loss value and the second loss value, parameters of the user tower model, the resource tower model, and the scoring network are adjusted.

13. The method according to claim 11, wherein The consumption index prediction network includes at least one of a click rate prediction network, a stay time prediction network and a completion rate prediction network, and the true value user behavior correspondingly includes at least one of a true value click behavior, a true value stay time and a true value completion behavior.

14. The method according to any one of claims 10 to 13, wherein: The obtaining of sample search query texts and sample user feature data of sample users includes: Sampling target candidate resources from the resources displayed to the sample user; Using a content understanding model to extract keywords from the target candidate resources, obtaining multiple keywords and their scores; sorting the plurality of keywords based on the scores of the plurality of keywords, and determining one or more target keywords from the plurality of keywords based on the sorting result; and The sample search query text is constructed based on the one or more target keywords.

15. A resource recommendation device based on a large model, comprising: A first acquisition unit is configured to acquire a user's search query text and user feature data; a first generating unit configured to generate a user representation vector of the user based on the search query text and the user feature data; A second acquisition unit is configured to acquire resource representation vectors of a plurality of candidate resources from a preset resource library; a first scoring unit configured to input the user representation vector and resource representation vectors of multiple candidate resources into a scoring network to obtain a comprehensive score for each of the multiple candidate resources; as well as A sorting unit is configured to sort the multiple candidate resources based on the comprehensive score, and present at least one candidate resource from the multiple candidate resources to the user based on the sorting result.

16. A training device for a resource recommendation model, comprising: a third acquiring unit, configured to acquire a sample search query text and sample user feature data of a sample user; a fourth acquisition unit configured to acquire sample resource representation vectors of a plurality of sample candidate resources and true value relevance scores between the sample user and the plurality of sample candidate resources, wherein the sample resource representation vectors are generated using a resource tower model; a second generating unit configured to generate a sample user representation vector of the sample user based on the sample search query text and the sample user feature data by using a user tower model; A second scoring unit is configured to input the sample user representation vector and the sample resource representation vectors of the multiple sample candidate resources into a scoring network to obtain a predicted score for each of the multiple sample candidate resources; and A parameter adjustment unit is configured to calculate a loss value based on the predicted score and the true value relevance score, and adjust parameters of the user tower model, the resource tower model, and the scoring network based on the loss value.

17. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 14.

18. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-14.

19. A computer program product comprising a computer program, wherein When the computer program is executed by a processor, the method according to any one of claims 1 to 14 is implemented.

Images