A multi-dimensional dynamic album recommendation method and system based on a large model

Abstract

The application relates to a multi-dimensional dynamic album recommendation method and system based on a large model and belongs to the technical field of intelligent storage management. The application comprises the following steps: constructing an album recommendation model based on an AI large model, wherein the album recommendation model comprises a user interest modeling module, a similarity calculation module, an incremental updating module and a dynamic optimization module connected in sequence; inputting a user interest feature vector and a picture feature vector into the album recommendation model; generating a user interest model according to the user interest feature vector through the user interest modeling module; calculating the similarity between the picture feature vector and the user interest model through the similarity calculation module; collecting interactive behavior data of a user on an initial recommended album list; pushing the interactive behavior data to the AI large model in real time through the incremental updating module; and reordering candidate albums in a preset picture library through the dynamic optimization module and the user interest model updated incrementally to generate a recommended album list after dynamic optimization.

Description

Technical Field

[0001] This invention belongs to the field of intelligent storage management technology, and in particular relates to a multi-dimensional dynamic photo album recommendation method and system based on a large model. Background Technology

[0002] With the continuous development of intelligent storage platforms and multimedia data management technologies, user image data is showing a trend of multi-source, multi-dimensional, and large-scale growth. Relying on large-scale artificial intelligence models for refined management of massive image resources has become an important development direction in smart storage scenarios. Building a personalized album recommendation mechanism based on user behavior and content characteristics can effectively integrate user-uploaded images, platform resources, and behavioral data, improving the retrieval efficiency and user experience of image resources.

[0003] Existing photo album recommendation technologies typically extract user behavior features and image content features to build a basic interest matching model, enabling the filtering and ranking of image collections. This type of technology can complete routine recommendation tasks under fixed preference scenarios, providing users with initial photo album aggregation services, and is of positive significance for promoting the interactive upgrade and resource optimization of intelligent storage platforms. The recommendation framework based on feature vectors and similarity calculation also lays the technical foundation for subsequent iterations of personalized recommendation technologies.

[0004] However, existing recommendation models mostly use static modeling methods, which makes it difficult to dynamically adapt to user interests based on real-time interaction behavior. They cannot respond quickly when user preferences change, and there is room for improvement in the matching accuracy between recommendation results and users' real needs. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of the invention is to provide a multi-dimensional dynamic photo album recommendation method and system based on a large model. It can integrate multi-source information such as user-uploaded image content, device binding information, operation behavior and historical interaction data to construct multi-dimensional user interest features. It can also dynamically optimize the user interest model through incremental update and interest decay-reinforcement dual mechanism to achieve real-time adaptive adjustment of recommendation results.

[0006] In a first aspect, the present invention proposes a multi-dimensional dynamic photo album recommendation method based on a large model, the method comprising: S1, acquire user data from multiple sources; S2 performs feature extraction and vectorization on multi-source data to construct user interest feature vectors and image feature vectors; S3, Construct an album recommendation model based on a large AI model. The album recommendation model includes a user interest modeling module, a similarity calculation module, an incremental update module, and a dynamic optimization module connected in sequence. S4, input the user interest feature vector and the image feature vector into the album recommendation model; S5 generates a user interest model based on the user interest feature vector through the user interest modeling module; S6, through the similarity calculation module, calculates the similarity between the image feature vector and the user interest model; S7. Based on similarity, filter from the preset image library to obtain an initial recommended album list, and collect user interaction data on the initial recommended album list; S8 uses an incremental update module to push interactive behavior data to the AI ​​big model in real time to incrementally update the user interest model. S9 uses a dynamic optimization module and an incrementally updated user interest model to reorder candidate albums in the preset image library and generate a dynamically optimized list of recommended albums.

[0007] Furthermore, multi-source data specifically includes user-uploaded image sets, user device binding information, user operation behavior data, user identity token information, and user historical interaction data.

[0008] Furthermore, S2 specifically includes: S201, integrates multi-source data uploaded by users to obtain a set of multi-dimensional user features; S202, vectorize the user's multidimensional feature set to obtain the user interest feature vector; S203, vectorize the image set in the multi-source data to obtain the image feature vector.

[0009] Furthermore, the AI ​​large model is specifically the iLoRA large model; The iLoRA large model consists of an SR-EMB feature extraction module, a gating network module, and an expert module connected in sequence.

[0010] Furthermore, S5 specifically includes: S501, input the user interest feature vector into the iLoRA large model; S502 extracts the temporal features of user behavior sequences through the SR-EMB feature extraction module; S503 uses a gated network module to assign weights to temporal features and generate a dynamic user preference distribution. S504, through the user interest modeling module, constructs a user interest model based on the dynamic distribution of user preferences.

[0011] Furthermore, the preset image library specifically refers to all image resources in the intelligent storage management system; The default image library contains a collection of user-uploaded images and image resources that the platform can recommend.

[0012] Furthermore, the interaction behavior data specifically includes the attention duration after the recommended album is exposed, the viewing operation time period, the overall browsing time of the album after clicking to enter, the dwell time of a single photo in the album, and the swiping order.

[0013] Furthermore, S8 specifically includes: S801, through the incremental update module, inputs interactive behavior data into the AI ​​big model; S802 uses a large AI model to extract user interest features based on interactive behavior data; S803 incrementally updates the user interest model based on user interest features.

[0014] Furthermore, the user interest model specifically includes an incremental learning framework and a dual mechanism of interest decay and reinforcement.

[0015] In a second aspect, the present invention proposes a multidimensional dynamic photo album recommendation system based on a large model, comprising: a memory and a processor; The memory stores an application program adapted to be executed by the processor to implement the multi-dimensional dynamic album recommendation method based on a large model as described in the first aspect.

[0016] The beneficial effects of this invention are as follows: In this embodiment of the invention, to address the problem that existing recommendation models are statically modeled and cannot dynamically adapt to user interests based on real-time interactions, a user interest model and similarity calculation basis are constructed by extracting features and vectorizing multi-source user data. An incremental update module updates the user interest model based on real-time interaction behavior, and a dynamic optimization module re-ranks candidate albums, enabling the recommendation results to quickly respond to changes in user preferences, effectively improving the matching accuracy between album recommendations and users' real needs, and achieving dynamic and accurate album recommendation results. Attached Figure Description

[0017] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. It is obvious that the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings.

[0018] Figure 1 This is a flowchart illustrating a multi-dimensional dynamic album recommendation method based on a large model provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the network architecture of a multi-dimensional dynamic photo album recommendation system based on home private cloud and distributed storage provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a multi-dimensional dynamic photo album recommendation system based on a large model, provided in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts disclosed in this invention.

[0021] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The terms "installed," "connected," and "linked" should be interpreted broadly; for example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of methods and systems consistent with some aspects of the invention as detailed in the appended claims.

[0023] This invention proposes a multidimensional dynamic photo album recommendation method and system based on a large model. It addresses the problems of cold start, data sparsity, and relatively simple recommendations in traditional recommendation methods, which lead to a decline in user experience and the effectiveness of recommendation results, thus failing to better adapt to changes in users' personalized needs.

[0024] Method Implementation Examples Reference Figure 1 The diagram illustrates a flowchart of a multi-dimensional dynamic album recommendation method based on a large model, provided by an embodiment of the present invention.

[0025] This invention provides a multi-dimensional dynamic photo album recommendation method based on a large model, the method comprising: Specifically, the method includes steps S1 to S9.

[0026] S1, acquire user data from multiple sources.

[0027] Optionally, the multi-source data specifically includes user-uploaded image sets, user device binding information, user operation behavior data, user identity token information, and user historical interaction data.

[0028] The multi-source data includes user-uploaded image sets, user device binding information, user operation behavior data, user identity token information, and user historical interaction data.

[0029] For example, user behavior data may include album clicks, photo swiping, and favorites, while user history interaction data may include browsing history and dwell time of previously recommended albums.

[0030] Furthermore, the multi-source data also includes low-frequency photo metadata such as the location and time of uploads to the user's album. This data will be incorporated into the feature system to improve the accuracy of user interest profiling and scene adaptation capabilities.

[0031] Specifically, the user identity token information generates a unique and encrypted special credential for different terminal devices of the same user, which is used to achieve precise binding between device identity and user behavior.

[0032] For example, the location a user uploads may include their home address, travel destination, etc., the upload time may include a specific date and time period, and the user's terminal device may include a mobile phone, tablet, PC, smart photo album, etc.

[0033] Reference Figure 2 The diagram illustrates a network architecture of a multi-dimensional dynamic photo album recommendation system based on a home private cloud and distributed storage, as provided in an embodiment of the present invention.

[0034] Specifically, Figure 2The user terminals include ordinary WiFi terminals, private network WiFi terminals, and 4G / 5G mobile terminals, which connect to the edge nodes via ordinary WiFi access, private network access, and mobile network access, respectively. The edge nodes include a home private cloud (with built-in object storage agent) and edge analytics services, connected to the central storage node and service nodes via DCI (Data Center Interconnect). The central storage node uses distributed object storage to provide user data space. The service nodes are divided into a smart storage platform and a private cloud business platform. The smart storage platform provides smart file services, variable data services, and smart photo album services, and stores an index library, tag library, media asset library, and vector library. The private cloud business platform provides device management, file management, and home management functions, and stores file, device, home, and data-related information. The two platforms interact through data synchronization.

[0035] Furthermore, in terms of inclusion relationships, the system consists of a user terminal layer, an edge node layer, a central storage node layer, and a service node layer. The edge node layer includes a home private cloud and edge analytics services; the home private cloud has a built-in object storage agent. The service node layer includes an intelligent storage platform and a private cloud business platform, each with its own built-in services and database. In terms of connectivity, user terminals connect to edge nodes, which in turn connect to the central storage node and service nodes via DCI. The central storage node and service nodes interact bidirectionally, and the intelligent storage platform and private cloud business platform are connected through data synchronization.

[0036] It should be noted that, Figure 2 By employing multi-terminal access methods, a collaborative architecture between edge nodes and central nodes, and a dual-platform data synchronization mechanism, the system achieves secure storage of user data, efficient interaction, and personalized intelligent album recommendations, thereby enhancing the system's scenario adaptability and service stability.

[0037] In this embodiment of the invention, multi-terminal access is used to acquire multi-source user data, which can comprehensively collect images, devices, behaviors, identities and historical interaction information, providing a rich and reliable data foundation for subsequent accurate modeling, while ensuring the stability and security of data collection.

[0038] S2 performs feature extraction and vectorization on multi-source data to construct user interest feature vectors and image feature vectors.

[0039] In one possible implementation, S2 specifically includes: S201 integrates multi-source data uploaded by users to obtain a set of multi-dimensional user features.

[0040] Among them, the user multidimensional feature set is a multidimensional representation of user features formed by integrating multi-source data.

[0041] Specifically, the user multidimensional feature set integrates user image attributes, device information, behavior records, identity information, and historical interaction information to form a unified feature expression.

[0042] Furthermore, the construction of the user's multi-dimensional feature set adopts multi-source data fusion technology and feature engineering optimization, and incorporates low-frequency photo metadata to effectively expand the user's interest reference dimensions.

[0043] It should be noted that low-frequency photo metadata includes the user's photo album upload location, upload time, etc. This data is rarely used in traditional algorithms, and this embodiment innovatively incorporates it into the feature system.

[0044] S202, the user's multidimensional feature set is vectorized and encoded to obtain the user interest feature vector.

[0045] Among them, the user interest feature vector is a numerical representation of the user's multidimensional feature set after vectorization encoding.

[0046] Furthermore, vectorized encoding maps discrete and continuous features into fixed-dimensional vectors, which facilitates model computation and matching.

[0047] S203, vectorize the image set in the multi-source data to obtain the image feature vector.

[0048] Among them, the image feature vector is the numerical representation of the image set after vectorization encoding.

[0049] It should be noted that the image feature vector is jointly encoded by image content features, metadata features, and tag features.

[0050] Specifically, image content features are extracted from photos using computer vision technology, including feature tags such as scenes and people.

[0051] For example, scene features can be divided into landscapes, portraits, food, etc., while person features can identify family members, friends, etc. in the photo, and tag features can include the shooting time and location.

[0052] In this embodiment of the invention, fusing and vectorizing multi-source data can unify heterogeneous data into computable feature vectors, expand the dimension of interest reference, improve feature expression capabilities, and lay a data foundation for efficient matching and recommendation.

[0053] S3 constructs an album recommendation model based on a large AI model. The album recommendation model includes a user interest modeling module, a similarity calculation module, an incremental update module, and a dynamic optimization module connected in sequence.

[0054] Optionally, the AI ​​large model is specifically the iLoRA large model.

[0055] The iLoRA large model consists of an SR-EMB feature extraction module, a gating network module, and an expert module connected in sequence.

[0056] Among them, the iLoRA large model focuses on sequence recommendation, using instantiated parameters and gating mechanisms to capture users' dynamic preferences. The SREMB feature extraction module is used to extract user behavior sequence features, the gating network module dynamically allocates weights according to sequence embedding, and the expert module is used to generate personalized contribution distributions for multiple experts, generating a customized LoRA parameter matrix.

[0057] Specifically, the gating network module dynamically adjusts the weights of each feature based on the temporal characteristics, and the expert module outputs a personalized parameter matrix adapted to the current user.

[0058] Furthermore, the user interest modeling module innovatively introduces a real-time behavior flow-driven incremental learning framework and an interest decay-reinforcement dual mechanism, breaking the limitations of traditional recommendation models that rely on static training data and fixed interest characterization.

[0059] In this embodiment of the invention, an AI album recommendation model containing multiple modules is constructed, which can realize the complete process from interest modeling to dynamic optimization. Relying on the iLoRA large model, the sequential feature learning ability is improved, making the recommendation system more in line with the user's dynamic preferences.

[0060] S4. Input the user interest feature vector and the image feature vector into the album recommendation model.

[0061] Furthermore, the two types of vectors represent user preferences and image attributes, respectively, providing basic data for subsequent similarity calculation and recommendation ranking.

[0062] It should be noted that the construction of user interest feature vectors incorporates low-frequency photo metadata, while the construction of image feature vectors incorporates content features extracted using computer vision technology, together improving recommendation accuracy.

[0063] In this embodiment of the invention, inputting two types of feature vectors into the recommendation model enables unified input of user preferences and image attributes, providing standardized data support for similarity calculation and personalized recommendation, and improving model processing efficiency.

[0064] S5 generates a user interest model based on user interest feature vectors through the user interest modeling module.

[0065] Among them, the user interest model is used to characterize users' preferences for album content.

[0066] Furthermore, the user interest model achieves dynamic adaptation to changes in user preferences through an incremental learning framework driven by real-time behavior flow and a dual mechanism of interest decay and reinforcement, thus constructing a real-time feedback closed loop of user behavior, interest model, and recommendation results.

[0067] In one possible implementation, S5 specifically includes: S501 inputs the user interest feature vector into the iLoRA large model.

[0068] Specifically, user interest feature vectors are used as input to the iLoRA large model to initiate the feature mining process of user behavior sequences.

[0069] S502 extracts the temporal features of user behavior sequences through the SR-EMB feature extraction module.

[0070] Among them, temporal features reflect the changing patterns of user behavior over time.

[0071] For example, time-series characteristics may include the frequency of users browsing photo albums at different times, changes in photo viewing time, and patterns in operation intervals.

[0072] S503 uses a gated network module to assign weights to time-series features and generate a dynamic user preference distribution.

[0073] Among them, the dynamic user preference distribution represents the degree of user preference for different types of content and their changing trends.

[0074] It should be noted that the weighting directly determines the degree of influence of each time-series feature on the final preference distribution.

[0075] S504, through the user interest modeling module, constructs a user interest model based on the dynamic distribution of user preferences.

[0076] Furthermore, the user interest model is constructed based on a dynamic preference distribution and is used for subsequent similarity matching with image feature vectors.

[0077] Specifically, user interest models are built by combining low-frequency photo metadata, which can aggregate and generate themed photo albums that meet users' personalized needs.

[0078] For example, themed photo albums may include travel albums, family gathering collections, etc., which are automatically generated based on metadata such as the time and location of user uploads.

[0079] In this embodiment of the invention, a user interest model is constructed based on temporal features and dynamic weight allocation, which can accurately depict the changing patterns of user preferences. Combined with incremental learning and a dual mechanism, dynamic iteration is achieved, making the interest characterization more in line with real needs.

[0080] S6 calculates the similarity between the image feature vector and the user interest model through the similarity calculation module.

[0081] Similarity is used to measure the degree to which recommended images match user interests.

[0082] Specifically, similarity calculation uses a vector space matching method, and the higher the value, the more the image content matches the user's current preferences.

[0083] Furthermore, similarity calculation combines image content features, user interest features, and low-frequency secondary data to improve matching accuracy and scenario adaptability.

[0084] In this embodiment of the invention, by calculating the similarity between images and user interests, the degree of content matching can be objectively quantified, and the matching accuracy can be improved by combining multi-dimensional features, providing a reliable basis for the initial album screening.

[0085] S7. Based on similarity, filter from the preset image library to obtain an initial recommended album list, and collect user interaction data on the initial recommended album list.

[0086] Optionally, the preset image library specifically refers to the full image resources in the intelligent storage management system.

[0087] The default image library contains a collection of user-uploaded images and image resources that the platform can recommend.

[0088] Furthermore, user-uploaded images in the preset image library are stored in multiple copies through the intelligent storage platform. The platform adopts a multi-site active-active architecture, storing photo copies on server nodes in different regions to prevent data loss.

[0089] For example, server nodes in different regions can include nodes in Beijing, Shanghai, Guangzhou, etc., to ensure the security of photo data.

[0090] Optionally, the interaction behavior data specifically includes the attention duration after the recommended album is exposed, the viewing operation time period, the overall browsing time of the album after clicking to enter, the dwell time of a single photo in the album, and the swiping order.

[0091] For example, the attention duration can be 3 seconds, 5 seconds, 10 seconds, etc., the overall album browsing time can be 15 seconds, 30 seconds, 60 seconds, etc., and the single photo viewing time can be 1 second, 2 seconds, 3 seconds, etc.

[0092] Specifically, interaction behavior data also includes users' actions of saving photos. This data will serve as real-time feedback data for incremental updates to the user interest model.

[0093] In this embodiment of the invention, the recommendation albums are filtered and recommended based on similarity and interaction behavior is collected. Real-time feedback data can be obtained while completing the initial recommendation, providing a basis for subsequent model updates and forming a recommendation closed loop.

[0094] The S8 uses an incremental update module to push interactive behavior data to the AI ​​big model in real time, and incrementally updates the user interest model.

[0095] Incremental update refers to dynamically adjusting the existing model based on newly added interactive data.

[0096] Specifically, incremental updates rely on an incremental learning framework driven by real-time behavior flow, combined with an interest decay-reinforcement dual mechanism, to achieve continuous model iteration.

[0097] In one possible implementation, S8 specifically includes: The S801 uses an incremental update module to input interactive behavior data into the AI ​​big model.

[0098] Specifically, interactive behavior data is used as real-time feedback data input into the model to update the representation of user interests.

[0099] Furthermore, the transmission of interactive behavior data can rely on relevant protocols to ensure efficiency. When a user uploads a new photo that matches the theme of a certain album, it can be synchronized to the terminal APP in real time.

[0100] S802 uses a large AI model to extract user interest features based on interactive behavior data.

[0101] Among them, user interest characteristics include photo album preference time periods, photo attention, content preference intensity, browsing habits, and interest decay patterns.

[0102] For example, the preferred time period for albums can be from 7 pm to 10 pm, the attention level of photos can be divided into three levels: high, medium and low, and the content preference intensity can be represented by a value between 0 and 1.

[0103] S803 incrementally updates the user interest model based on user interest features.

[0104] Among them, the incremental learning framework is used to support the model to continuously iterate based on new data, and the interest decay-reinforcement dual mechanism is used to realize the dynamic adaptation of the model to changes in user preferences, and to build a real-time feedback closed loop of user behavior-interest model-recommendation results.

[0105] It should be noted that the interest decay-reinforcement dual mechanism achieves dynamic model adaptation by weakening old preferences and strengthening new preferences.

[0106] Specifically, incremental updates enable the real-time inclusion of newly uploaded photos by users. When a user uploads a new photo that matches a certain themed album, the platform automatically includes it and synchronizes it to the terminal APP.

[0107] Optionally, the user interest model specifically includes an incremental learning framework and an interest decay-reinforcement dual mechanism.

[0108] In this embodiment of the invention, incremental updates to the interest model based on interactive data can quickly respond to changes in user preferences, continuously optimize model expression, achieve dynamic adaptation and real-time iteration, and improve the timeliness of recommendations.

[0109] S9 uses a dynamic optimization module and an incrementally updated user interest model to reorder candidate albums in the preset image library and generate a dynamically optimized list of recommended albums.

[0110] Among them, the dynamically optimized recommended album list refers to the recommendation results that have been reordered to better match the user's real-time interests.

[0111] Furthermore, the re-ranking is performed based on the updated user interest model, ensuring that the recommended album list aligns with users' actual needs.

[0112] Specifically, the dynamically optimized recommended album list can be quickly loaded via the terminal app based on relevant protocols, ensuring that users can view the latest smart album collection at any time.

[0113] For example, when a user accesses the Smart Storage platform through the Gallery app, the app uses the HTTP / 2 protocol to quickly load album data, improving the user's viewing experience.

[0114] In this embodiment of the invention, the album is reordered using the updated model, which can output a recommendation list that is more in line with the user's real-time interests, improve recommendation accuracy and user experience, and realize dynamic and personalized album recommendations.

[0115] The beneficial effects of this invention are as follows: In this embodiment of the invention, to address the problem that existing recommendation models are statically modeled and cannot dynamically adapt to user interests based on real-time interactions, a user interest model and similarity calculation basis are constructed by extracting features and vectorizing multi-source user data. An incremental update module updates the user interest model based on real-time interaction behavior, and a dynamic optimization module re-ranks candidate albums, enabling the recommendation results to quickly respond to changes in user preferences, effectively improving the matching accuracy between album recommendations and users' real needs, and achieving dynamic and accurate album recommendation results.

[0116] System Implementation Examples Reference manual attached Figure 3The diagram shows a structural schematic of a multi-dimensional dynamic photo album recommendation system based on a large model provided by an embodiment of the present invention.

[0117] This invention proposes a multi-dimensional dynamic photo album recommendation system 30 based on a large model, comprising: a memory 303 and a processor 301.

[0118] The memory 303 stores an application program adapted to be executed by the processor 301 to implement the big data-based oil equipment fault prediction method of the method embodiment.

[0119] The big data-based oil equipment fault prediction system 30 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302.

[0120] The structure of the multi-dimensional dynamic album recommendation system 30 based on a large model does not constitute a limitation on the embodiments of the present invention.

[0121] Processor 301 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0122] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI bus or an EISA bus, etc. Bus 302 may be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus.

[0123] The memory 303 may be a ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0124] Computer-readable storage medium embodiments This invention proposes a computer-readable storage medium storing a computer program that can be loaded and executed by a processor for a first aspect of a multi-dimensional dynamic photo album recommendation method based on a large model.

[0125] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-dimensional dynamic photo album recommendation method based on a large model, characterized in that, The multi-dimensional dynamic album recommendation method based on a large model includes: S1, acquire user data from multiple sources; S2, perform feature extraction and vectorization processing on the multi-source data to construct user interest feature vectors and image feature vectors; S3, Construct an album recommendation model based on an AI large model. The album recommendation model includes a user interest modeling module, a similarity calculation module, an incremental update module, and a dynamic optimization module connected in sequence. S4, input the user interest feature vector and the image feature vector into the album recommendation model; S5, The user interest model is generated by the user interest modeling module based on the user interest feature vector; S6, The similarity calculation module calculates the similarity between the image feature vector and the user interest model; S7. Based on the similarity, filter from the preset image library to obtain an initial recommended album list, and collect user interaction data on the initial recommended album list; S8, The interactive behavior data is pushed to the AI ​​big model in real time through the incremental update module to incrementally update the user interest model; S9. Using the dynamic optimization module and the incrementally updated user interest model, the candidate albums in the preset image library are reordered to generate a dynamically optimized list of recommended albums.

2. The multi-dimensional dynamic album recommendation method based on a large model according to claim 1, characterized in that, The multi-source data specifically includes user-uploaded image sets, user device binding information, user operation behavior data, user identity token information, and user historical interaction data.

3. The multi-dimensional dynamic album recommendation method based on a large model according to claim 1, characterized in that, S2 specifically includes: S201, integrate the multi-source data uploaded by the user to obtain a multi-dimensional feature set of the user; S202, the user's multidimensional feature set is vectorized and encoded to obtain the user's interest feature vector; S203, the image set in the multi-source data is vectorized and encoded to obtain the image feature vector.

4. The multi-dimensional dynamic album recommendation method based on a large model according to claim 1, characterized in that, The AI ​​large model is specifically the iLoRA large model; The iLoRA large model includes an SR-EMB feature extraction module, a gating network module, and an expert module connected in sequence.

5. The multi-dimensional dynamic album recommendation method based on a large model according to claim 4, characterized in that, S5 specifically includes: S501, Input the user interest feature vector into the iLoRA large model; S502, the temporal features of the user behavior sequence are extracted through the SR-EMB feature extraction module; S503, The time-series features are weighted and distributed through the gating network module to generate a dynamic user preference distribution; S504, The user interest model is constructed based on the user dynamic preference distribution through the user interest modeling module.

6. The multi-dimensional dynamic album recommendation method based on a large model according to claim 1, characterized in that, The preset image library specifically refers to the full image resources in the intelligent storage management system; The preset image library includes a collection of user-uploaded images and image resources that the platform can recommend.

7. The multi-dimensional dynamic album recommendation method based on a large model according to claim 1, characterized in that, The interactive behavior data specifically includes the attention duration after the recommended album is exposed, the viewing operation time period, the overall browsing time of the album after clicking to enter, the dwell time of a single photo in the album, and the swiping order.

8. The multi-dimensional dynamic album recommendation method based on a large model according to claim 1, characterized in that, S8 specifically includes: S801, The interactive behavior data is input into the AI ​​big model through the incremental update module; S802, using the aforementioned AI big model, extracts user interest features based on interactive behavior data; S803, Based on the user interest features, the user interest model is incrementally updated.

9. The multi-dimensional dynamic photo album recommendation method based on a large model according to claim 1, characterized in that, The user interest model specifically includes an incremental learning framework and an interest decay-reinforcement dual mechanism.

10. A multi-dimensional dynamic photo album recommendation system based on a large model, characterized in that, include: Memory and processor; The memory stores an application program adapted to be executed by the processor to implement the multi-dimensional dynamic album recommendation method based on a large model as described in any one of claims 1 to 9.

Images