Resource cold start method, device and equipment based on large model and storage medium

Abstract

The invention provides a resource cold start method and device based on a large model, equipment and a storage medium. The invention relates to the technical field of artificial intelligence, in particular to the fields of large models, natural language processing, deep learning, intelligent recommendation and the like, and can be applied to application scenes such as recommendation systems, advertisement putting, content distribution, e-commerce platforms, search engine optimization and the like. According to the specific implementation scheme, the method comprises the following steps: analyzing feature data of a target object from an online request; determining a vector of the target object based on the feature data of the target object, and obtaining a first recall result from a multi-modal vector index database based on the vector of the target object; determining a first target resource set based on the feature data of the target object, and obtaining a second recall result from a multi-modal vector index database based on a multi-modal vector corresponding to a target resource in the first target resource set; and performing integration processing based on the first recall result and the second recall result, and sending an integration result to a recommendation system.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to the fields of big models, natural language processing, deep learning and intelligent recommendation. It can be used in application scenarios such as recommendation systems, advertising, content distribution, e-commerce platforms, search engine optimization, etc., and specifically relates to a resource cold start method, device, equipment and storage medium based on a big model. Background Art

[0002] Recommendation systems are an indispensable part of modern Internet services, and personalized recommendation services provide great convenience for users. However, in cold start scenarios, since new resources have no interactive information in the recommendation system, it is difficult to make effective recommendations, and there is still a problem of poor recommendation results. Summary of the invention

[0003] The present invention provides a resource cold start method, device, equipment and storage medium based on a large model.

[0004] According to one aspect of the present disclosure, a resource cold start method based on a large model is provided, the method comprising:

[0005] Parse the characteristic data of the target object from the online request;

[0006] Determine a vector of the target object based on feature data of the target object, and obtain a first recall result from a multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores multimodal vectors of new resources, and the multimodal vectors of the new resources are extracted from the fine-tuned multimodal macro model;

[0007] Determine a first target resource set based on the feature data of the target object, and obtain a second recall result from a multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set;

[0008] An integration process is performed based on the first recall result and the second recall result, and the integration result is sent to the recommendation system, so that the recommendation system determines a new resource to be recommended based on the integration result.

[0009] According to another aspect of the present disclosure, a large model-based resource cold start device is provided, the device comprising:

[0010] A parsing module is used to parse the characteristic data of the target object from the online request;

[0011] A first determination module is used to determine a vector of a target object based on feature data of the target object, and obtain a first recall result from a multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores multimodal vectors of new resources, and the multimodal vectors of the new resources are extracted from a fine-tuned multimodal macro model;

[0012] A second determination module is used to determine a first target resource set based on the feature data of the target object, and obtain a second recall result from a multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set;

[0013] The integration and sending module is used to perform integration processing based on the first recall result and the second recall result, and send the integration result to the recommendation system, so that the recommendation system determines the new resource to be recommended based on the integration result.

[0014] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] 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 execute any method in the embodiments of the present disclosure.

[0018] 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 the computer to execute any method according to the embodiments of the present disclosure.

[0019] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements any method according to the embodiments of the present disclosure when executed by a processor.

[0020] By adopting the solution disclosed in the present invention, personalized recommendations are achieved in the absence of historical data support by combining the feature data of the target object and the multimodal vector of the new resource. This can not only improve the accuracy and diversity of the recommendations, but also accelerate the exposure and distribution of new resources and promote the effective use of new resources.

[0021] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0023] Figure 1 The following is a schematic diagram of a resource cold start method based on a large model according to an embodiment of the present disclosure. Figure 1 ;

[0024] Figure 2 The following is a schematic diagram of a resource cold start method based on a large model according to an embodiment of the present disclosure. Figure 2 ;

[0025] Figure 3 is an architecture diagram of a resource cold start based on a large model according to an embodiment of the present disclosure;

[0026] Figure 4 is a structural schematic diagram of a resource cold start device based on a large model according to an embodiment of the present disclosure;

[0027] Figure 5 is a scenario diagram of a resource cold start method based on a large model according to an embodiment of the present disclosure;

[0028] Figure 6 It is a structural schematic diagram of an electronic device used to implement the large model-based resource cold start method of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] The following is a description of exemplary embodiments of the present disclosure 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 recognized by those of ordinary skill 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, the description of well-known functions and structures is omitted in the following description.

[0030] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The term "at least one" in this article means any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C. The terms "first" and "second" in this article refer to multiple similar technical terms and distinguish them. They do not mean to limit the order or to limit them to only two. For example, the first feature and the second feature refer to two types / two features. The first feature can be one or more, and the second feature can also be one or more.

[0031] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0032] Before introducing the technical solutions of the embodiments of the present disclosure, the technical terms that may be used in the present disclosure are further explained:

[0033] Resource cold start: refers to the situation when new resources (such as new products, new content, etc.) are just introduced into the system, and it is difficult to directly make personalized recommendations due to the lack of user interaction data or historical behavior data. At this time, specific strategies and methods are needed to accelerate the exposure and distribution of resources and improve the accuracy and diversity of recommendations.

[0034] Multimodal data: refers to data containing multiple types of information, such as text, images, videos, audio, etc. These data can describe the characteristics and information of an object from different angles and levels.

[0035] Multimodal vector: refers to a high-dimensional numerical representation generated by comprehensive analysis and processing of various types of data (such as text, images, videos, etc.) through a multimodal large model. It can fully reflect the characteristics and information of the data and is an important basis for the recommendation system.

[0036] Multimodal large model: refers to a deep learning model that can process multiple types of data at the same time. It can capture the intrinsic connections and interactions between different types of data, thereby more comprehensively understanding the characteristics and information of the object.

[0037] Multimodal vector index library: A database that stores vector representations of various types of resources (such as text, images, videos, etc.). These vectors extract and integrate various features of resources through advanced algorithms and can be used for tasks such as similarity calculation and cluster analysis. In resource cold start scenarios, the multimodal vector index library can provide strong support for the recommendation of new resources.

[0038] Offline logs: refers to log files that record the interaction between users and the system. These log files contain a large amount of user behavior data, which can be used to analyze user preferences, optimize recommendation algorithms, etc.

[0039] Resource cold start, as a startup or recommendation strategy, is mainly used to deal with the problem that new resources or new users lack sufficient data support in the initial stage. Most recommendation systems face problems such as data sparsity and cold start.

[0040] 1. New resources lack user feedback: Newly listed resources have just been launched and have not yet accumulated enough user reviews, ratings or purchase records, making it difficult for the system to evaluate their popularity and compatibility with users, making it difficult to make effective recommendations.

[0041] 2. Data sparsity: Even for users and resources with certain historical data, there may be data sparsity. That is, some users may have only interacted with very few resources, and some resources may have only been followed by very few users, which increases the difficulty and uncertainty of recommendation.

[0042] In related technologies, resource cold start methods mainly include the following categories:

[0043] 1. Content-based recommendation: Use the resource's own attribute information (such as product description, category, tags, etc.) to make recommendations. Even without user interaction data, similar items can be recommended based on the user's initial preferences.

[0044] 2. Graph-based recommendation: Use graph structures to model the relationship between users and resources. For example, use graph neural networks to learn the internal representation of resources and make recommendations based on similar relationships between resources.

[0045] 3. Based on active learning, taking the Exploration-Exploitation (EE) strategy as an example, when we don’t know whether an item is popular, we first tentatively expose the item to a group of seed users and observe the user’s feedback on the item. If the feedback is good, the item will be recommended to more users; otherwise, the number of recommendations will be reduced. This method can help the recommendation system quickly discover the user’s interests and optimize the recommendation effect.

[0046] However, the above cold start methods all have obvious disadvantages.

[0047] For example, content-based cold start methods are usually not customized according to the user's personalized needs, which may result in the recommended content not matching the user's interests.

[0048] For example, other cold start methods based on personalized models do not fully utilize resource attribute information, lack multimodal information fusion, and have limited ability to extract resource-related information. Taking e-commerce recommendations as an example, product image information generally uses traditional algorithms such as Scale-Invariant Feature Transform (SIFT) and Speeded Up Robust Features (SURF) to extract key points in the image, and deep learning models such as Convolutional Neural Networks (CNN) to extract high-level feature representations from the image; extract text content information such as product titles and product descriptions, using bag-of-words models such as Term Frequency-Inverse Document Frequency (TF-IDF) and word embedding methods to convert words in the text into fixed-length vector representations. Image and text information are extracted and modeled separately, lacking the fusion of different modal information, and cannot fully characterize the attribute information of a product.

[0049] Therefore, in the cold start phase, the interaction behavior data between new resources and users is sparse, and the model trained based on user feedback signals lacks sufficient input. Model training cannot converge to a confident result, and the recommendation system relies more on the attribute information of the new resource itself. Due to the lack of fusion of different modal information, the attribute characteristics of a new resource cannot be effectively characterized, which makes the cold start problem more serious.

[0050] In order to at least partially solve one or more of the above problems and other potential problems, the present disclosure proposes a resource cold start method based on a large model, which realizes personalized recommendation in the absence of historical data support by combining the feature data of the target object and the multimodal vector of the new resource. This method can not only improve the accuracy and diversity of recommendations, but also accelerate the exposure and distribution of new resources, promote the effective use of new resources and improve the satisfaction of the target object.

[0051] The present disclosure provides a resource cold start method based on a large model. Figure 1It is a flow chart of a resource cold start method based on a big model according to an embodiment of the present disclosure. The resource cold start method based on a big model can be applied to a resource cold start device based on a big model. The resource cold start device based on a big model is located in an electronic device. The electronic device includes but is not limited to fixed devices and / or mobile devices. For example, fixed devices include but are not limited to servers, and servers can be cloud servers or ordinary servers. For example, mobile devices include but are not limited to: mobile phones, tablet computers. In some possible implementations, the resource cold start method based on a big model can also be implemented by a processor calling computer-readable instructions stored in a memory. For example Figure 1 As shown, the resource cold start method based on the large model includes:

[0052] S101, parsing characteristic data of a target object from an online request;

[0053] S102, determining a vector of the target object based on feature data of the target object, and obtaining a first recall result from a multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores multimodal vectors of new resources, and the multimodal vectors of the new resources are extracted from the fine-tuned multimodal macro model;

[0054] S103, determining a first target resource set based on the feature data of the target object, and obtaining a second recall result from a multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set;

[0055] S104: Perform integration processing based on the first recall result and the second recall result, and send the integration result to the recommendation system, so that the recommendation system determines new resources to be recommended based on the integration result.

[0056] In the embodiment of the present disclosure, the target object refers to the subject that needs resource recommendation or service. In the personalized recommendation system, the target object can be understood as the user who needs recommended resources.

[0057] In the disclosed embodiments, an online request refers to a request initiated by a target object to the system in some way (such as web browsing, application use, search query, etc.). The online request contains some basic information, behavior data or clear query requirements of the target object. By parsing these online requests, the system can understand the current status, needs and preferences of the target object, thereby providing personalized recommendation services for the target object.

[0058] In the disclosed embodiment, the characteristic data of the target object may include basic information authorized to be disclosed by the target object (such as nickname, geographic location), hobbies, historical behavior records, etc. By analyzing these characteristic data, the needs and preferences of the target object can be preliminarily understood, and then subsequent steps such as vector generation, similarity calculation and recommendation can be performed.

[0059] In the disclosed embodiment, the multimodal vector index library stores the multimodal vectors of the new resources. The multimodal vectors are vectors that extract and integrate multiple features of the new resources.

[0060] In some embodiments, a vector representation of the target object is generated using a machine learning or deep learning algorithm based on the feature data of the target object. This vector can capture the key features of the target object and is used to perform similarity calculations in subsequent steps. Then, this vector is compared with the new resource vector in the multimodal vector index library, and the new resources ranked in the top K (K is an integer greater than 1) in similarity are retrieved as the first recall result.

[0061] In some embodiments, the first target resource set is a set of target resources (e.g., the posterior index is greater than the first threshold) extracted from the historical consumption resource set of the target object. Based on the multimodal vectors corresponding to the target resources in this set, the top M (M is an integer greater than 1) new resources with similarity ranking are retrieved from the multimodal vector index library as the second recall result. Here, the values ​​of M and K can be the same or different.

[0062] In some embodiments, the first recall result and the second recall result are integrated. The integration process may include operations such as deduplication and sorting to ensure that the final recommended resource set is both rich and accurate. The integrated results will be sent to the recommendation system, which will determine the final new resources to be recommended based on the integrated results and the specific needs of the user.

[0063] For example, suppose an e-commerce platform encounters the problem of resource cold start when promoting a newly launched smart watch. Through this method, the e-commerce platform can parse the user's nickname, shopping preferences and other feature data from the user's online request, and generate a vector representation of the user. Then, the multimodal vectors of the smart watches in the multimodal vector index library are used to calculate the similarity, and the top K smart watches with similarity ranking are obtained as the first recall result. At the same time, a first target resource set (such as electronic products purchased by the user before) is determined based on the user's shopping history record, and new resources with top M similarity rankings with the set are obtained from the multimodal vector index library as the second recall result. Finally, the two recall results are integrated and recommended to users to improve user experience.

[0064] It should be noted that the present disclosure does not limit the execution order of S102 and S103. The two can be executed at the same time, and S102 can be executed before S103 or after S103.

[0065] The technical solution of the disclosed embodiment realizes personalized recommendation in the absence of historical data support by combining the feature data of the target object and the multimodal vector of the new resource, which not only improves the accuracy and diversity of the recommendation, but also accelerates the exposure and distribution of new resources and promotes the effective use of new resources.

[0066] In some embodiments, the resource cold start method based on the large model further includes:

[0067] Extract multi-modal data from multiple datasets corresponding to each published resource;

[0068] Input the multi-modal data corresponding to each published resource into the multi-modal big model to obtain the multi-modal vector of each published resource;

[0069] The multimodal large model is fine-tuned using the interaction data between the published resources and the objects mined from the first offline log to obtain a fine-tuned multimodal large model.

[0070] In the disclosed embodiments, the data set may include text descriptions, images, videos, audios, etc., covering different aspects and features of the resources.

[0071] In some embodiments, the multimodal data corresponding to each published resource is input into a multimodal macromodel. This multimodal macromodel can process multiple types of data simultaneously and capture the intrinsic connections and interactions between them. Through the calculation of the model, a multimodal vector of each published resource is obtained. This vector is a high-dimensional numerical representation that can fully reflect the characteristics and information of the resource.

[0072] In some embodiments, in order to further improve the accuracy and adaptability of the model, the multimodal large model is fine-tuned using the interaction data between the published resources and the objects mined from the first offline log. These interaction data record the actual interaction behaviors between the objects and the resources, such as clicks, browsing, purchases, etc., and reflect the preferences and needs of the objects for the resources. Through the fine-tuning process, the model can learn more object behavior patterns and resource characteristics, thereby more accurately predicting the object's interest and preference for new resources.

[0073] For example, suppose an e-commerce platform uses the above resource cold start method when promoting a newly launched smart speaker. First, extract multiple modal data such as text descriptions, product images, and usage videos from multiple data sets of smart speakers. Then, input these data into the multimodal large model to obtain the multimodal vector of the smart speaker. Next, use the interaction data between users and smart speaker products in offline logs to fine-tune the model. Finally, in the resource cold start stage, use the fine-tuned multimodal large model and multimodal vector index library to quickly find target users who are interested in smart speakers and recommend the product to them.

[0074] In this way, by extracting multi-modal data and using a multimodal large model, we can capture the characteristics and information of resources more comprehensively, thereby improving the accuracy of recommendations. We use the interaction data in offline logs to fine-tune the model so that it can better adapt to user behavior and demand changes in actual scenarios. In the resource cold start phase, by using the multimodal vector index library and the fine-tuned multimodal large model, we can quickly find similar target objects for new resources, thereby accelerating the exposure and distribution of resources.

[0075] In some embodiments, the resource cold start method based on the large model further includes:

[0076] Extract multimodal vectors of new resources using the fine-tuned multimodal macro model;

[0077] Build a multimodal vector index library based on the multimodal vectors of new resources.

[0078] In some embodiments, after fine-tuning the multimodal large model, the model is used to extract multimodal vectors for new resources. New resources also contain multimodal data, such as text descriptions, images, videos, etc. After preprocessing, these data are input into the fine-tuned multimodal large model. Based on the learned knowledge, the model conducts a comprehensive analysis of the multimodal data of the new resource and generates a multimodal vector that can fully reflect the characteristics and information of the new resource. This vector will serve as an important basis for the new resource in the subsequent recommendation process.

[0079] In some embodiments, after obtaining the multimodal vectors of new resources, it is necessary to build a multimodal vector index library to store these vectors. The index library is a key component in the recommendation system, which allows the system to quickly retrieve and match similar resources or users. When building the index library, efficient indexing algorithms and data structures can be used to optimize query performance. At the same time, in order to ensure the timeliness and accuracy of the index library, the vector information in the index library needs to be updated regularly to reflect the latest status of new resources.

[0080] For example, suppose a video platform has just launched a new documentary that contains rich text descriptions, high-quality trailers, and detailed introductions to the production team. During the resource cold start phase, the platform can use the fine-tuned multimodal large model to extract the multimodal vectors of the documentary and add them to the multimodal vector index library. Later, when a user visits the platform, the recommendation system can query the index library to find other documentaries or related content similar to the documentary and recommend them to the user. This not only increases the exposure of new documentaries, but also increases the user's viewing experience and satisfaction.

[0081] In this way, by extracting the multimodal vectors of new resources in real time and building an index library, the recommendation system can quickly include new resources in the recommendation scope and improve the real-time performance of the recommendation. At the same time, the introduction of multimodal vectors makes the recommendation process more comprehensive and accurate, and can better meet the personalized needs of users. In the cold start stage of resources, due to the lack of user interaction data, traditional recommendation algorithms are often difficult to work. However, by using a fine-tuned multimodal large model to extract the multimodal vectors of new resources and building an index library for similarity matching, the resource cold start problem can be effectively alleviated and the exposure and click-through rate of new resources can be improved. With the continuous release of new resources and the continuous changes in user needs, the recommendation system needs to have good scalability and flexibility. By building a multimodal vector index library and continuously optimizing and updating the index algorithm and data structure, it can be ensured that the recommendation system can adapt to these changes and maintain efficient and stable operation.

[0082] In some embodiments, fine-tuning the multimodal large model using the interaction data between the published resources and the objects mined from the first offline log includes:

[0083] A content-based collaborative filtering (CB2CF) model is adopted to fine-tune the multimodal large model using the interaction data of published resources and objects mined from the first offline log.

[0084] In the disclosed embodiment, CB2CF is a recommendation algorithm that combines the ideas of content filtering and collaborative filtering. It uses the content characteristics of resources to predict user preferences and optimizes the recommendation results based on the user's historical interaction behavior. The CB2CF model can simultaneously consider the intrinsic attributes of resources and the external behavioral characteristics of users, thereby providing a more comprehensive and accurate recommendation service.

[0085] In some implementations, the interaction data between the published resources and the objects are mined from the first offline log. Such data may include user behavior records such as clicks, browsing, and comments, as well as multimodal information such as text descriptions, images, and videos related to the resources. The mined data is cleaned and preprocessed to remove noise and outliers to ensure data quality and consistency.

[0086] In some implementations, the CB2CF model is used as the basis for fine-tuning the multimodal large model. The CB2CF model combines the advantages of content filtering and collaborative filtering, and can optimize the recommendation effect by utilizing the content features of resources and the interactive behavior of users.

[0087] In some embodiments, the preprocessed interaction data is used as a training set and input into a multimodal large model. During the training process, the model learns the correlation between the multimodal features of resources and user behavior. Using the mechanism of the CB2CF model, the user's preference for resources is predicted through content features (such as text descriptions, image features, etc.), and the idea of ​​collaborative filtering is combined to optimize these prediction results. Specifically, the model will try to find other resources similar to the user's historical interaction resources and recommend these resources to the user. During the fine-tuning process, the model parameters and training strategies, such as learning rate, batch size, training rounds, etc., can be adjusted according to the actual situation to obtain better recommendation effects.

[0088] In some embodiments, appropriate evaluation indicators (such as accuracy, recall, etc.) are used to evaluate the performance of the fine-tuned multimodal large model. The model is further optimized and adjusted based on the evaluation results to improve the recommendation effect and user satisfaction.

[0089] For example, suppose an e-commerce platform has newly launched a smartwatch with multiple functions and features, and provides detailed text descriptions and high-definition pictures on the product details page. By mining the user's interaction data with similar smartwatches (such as purchase records, browsing time, etc.) from offline logs, and using the CB2CF model to fine-tune the multimodal large model. During the fine-tuning process, the model will learn the correlation between the text description and image features of the smartwatch and the user's purchasing behavior. When a new user visits the platform and browses this smartwatch, the model can predict the user's preference for the watch based on its multimodal features and the user's potential needs, and recommend it to the user to improve the user experience.

[0090] In this way, by combining content features and user behavior data, the fine-tuned multimodal large model can more accurately predict user preferences and needs, thereby providing more personalized recommendation services. For newly released resources or new users, traditional recommendation algorithms often fail to work due to the lack of sufficient interaction data. The content-based collaborative filtering model can use the content features of resources to make recommendations, thereby alleviating the cold start problem to a certain extent. With the continuous release of new resources and the continuous changes in user behavior, the fine-tuned multimodal large model can continuously learn and optimize its own recommendation capabilities to adapt to these changes and maintain efficient recommendation results.

[0091] In some embodiments, determining the vector of the target object based on the feature data of the target object includes: using a collaborative filtering (CF) model of a CB2CF model to determine the vector of the target object based on the feature data of the target object.

[0092] Here, the CF model is the CF model of the CB2CF model used when fine-tuning the multimodal large model.

[0093] In the disclosed embodiments, CF model: a recommendation algorithm that discovers user interests or similarities between items by analyzing the interaction data between users and items (such as ratings, purchase records, etc.), and makes recommendations based on this.

[0094] In some embodiments, a suitable CF model is selected according to the application scenario and data characteristics. Common CF models include user-based collaborative filtering (User-based CF) and item-based collaborative filtering (Item-based CF). User-based collaborative filtering focuses on finding other users with similar interests to the target user and recommending items that these users like; while item-based collaborative filtering focuses on finding other items similar to the target item and recommending them to users who like these similar items.

[0095] In some implementations, interactive data related to the target object is extracted, including user ratings, purchase records, browsing history, etc. The data is preprocessed, including removing noise, filling missing values, normalizing, etc., to ensure data quality.

[0096] In some embodiments, under the framework of the CB2CF model, the result output by the CF model can be used as part or all of the target object vector. For item-based CF, the target object vector can be directly obtained by weighted average of the feature vectors of similar items. For user-based CF, if user features need to be included in the target object vector, the target object vector can be constructed by calculating the similarity between the target user and other users and combining the feature vectors of these users.

[0097] In some embodiments, considering that the CB2CF model also combines content features (Content-Based, CB), the vector of the target object can be further optimized by fusing the content features and the collaborative filtering results. For example, a deep learning model (such as a neural network) can be used to take the content features and the collaborative filtering results as input to generate a more accurate target object vector by training the model.

[0098] For example, suppose an e-commerce platform is using the CB2CF model for product recommendations. For a newly listed product A, we can determine its vector by following the steps below: Use the item-based collaborative filtering model to find other products similar to product A (such as product B, product C, etc.). Extract the feature vectors of these similar products (such as price, brand, category, user rating, etc.). According to the similarity between product A and these similar products, perform a weighted average of these feature vectors to obtain the preliminary vector of product A. Combined with the content features of product A (such as product description, pictures, videos, etc.), use a deep learning model to optimize and adjust the preliminary vector to obtain the final vector of product A.

[0099] In this way, by combining collaborative filtering and content features, user interests and item attributes can be captured more comprehensively, thereby improving the accuracy of recommendations. For new users or new items, even if there is a lack of interaction data, their vectors can be constructed through content features and preliminary recommendations can be made. The collaborative filtering model can explain the reasons behind the recommendation results (such as similar users or similar items), thereby improving the interpretability of the model and user trust.

[0100] In some embodiments, determining the first target resource set based on the characteristic data of the target object includes:

[0101] Based on the characteristic data of the target object, obtaining a list of published resources consumed by the target object from the second offline log;

[0102] Published resources whose a posteriori indicators are greater than a first threshold are screened out from the published resource list to form a first target resource set.

[0103] In some embodiments, obtaining the list of consumed resources from the second offline log includes: using the unique identifier of the target object (such as a user identification (ID)) to retrieve the list of published resources consumed by the target object in the second offline log. The second offline log is usually a large database that stores user behavior data (such as clicks, browsing, purchases, etc.), which is used for subsequent analysis and recommendation.

[0104] In some implementations, screening resources whose a posteriori index is greater than the first threshold includes: for a list of published resources obtained from the log, it is necessary to calculate or extract the a posteriori index of each resource. The posteriori index is a quantitative standard for measuring resource quality or user satisfaction, and may include user ratings, number of comments, conversion rate, repurchase rate, etc. The published resources whose a posteriori index is greater than the first threshold are screened out, and these resources will be included in the first target resource set.

[0105] In the embodiment of the present disclosure, the first threshold is a judgment standard set in advance based on factors such as business requirements, data distribution and actual effects.

[0106] For example, suppose an e-commerce platform wants to recommend high-quality products to its users. First, the platform extracts feature data from the user's profile and historical purchase records. Then, it searches the offline log for a list of products that the user has purchased. Next, the platform calculates the posterior indicators of each product, such as user ratings, number of reviews, and conversion rate. Suppose the first threshold is set to 4.5 points (out of 5 points). Finally, the platform screens out all products with user ratings greater than 4.5 points to form the first target resource set.

[0107] In this way, by screening resources with excellent a posteriori indicators, we can ensure that the content recommended to users is more in line with their needs and interests, thereby improving the accuracy of recommendations and user satisfaction. Under limited resource conditions, allocating more exposure opportunities and resources to verified, high-quality resources will help improve the overall content quality and user experience. Through automated screening and recommendation processes, the cost of cold start of resources can be reduced.

[0108] In some embodiments, the large model-based resource cold start method further includes:

[0109] Based on the first target resource set, a multimodal vector storage system is requested to obtain a multimodal vector corresponding to a target resource in the first target resource set. The multimodal vector storage system stores multimodal vectors of published resources and multimodal vectors of new resources.

[0110] In the disclosed embodiment, the multimodal vector storage system is a system capable of storing and retrieving multimodal vectors of published resources and new resources. The system is usually equipped with an efficient indexing and query mechanism to quickly respond to requests from the recommendation system.

[0111] In the resource cold start method, requesting the multimodal vector storage system based on the first target resource set to obtain the multimodal vectors corresponding to the target resources is an important step to improve the recommendation system's ability to understand and process new resources.

[0112] In some embodiments, a request is constructed to be sent to a multimodal vector storage system based on a resource ID or unique identifier in the first target resource set. The type of multimodal vector to be obtained is determined, such as a text vector, an image vector, an audio vector, etc., as well as any additional parameters that may be required (such as vector dimensions, similarity measurement methods, etc.). The prepared request is sent to a multimodal vector storage system. This system should have pre-stored multimodal vectors for published resources and new resources.

[0113] In some embodiments, a response is received from a multimodal vector storage system, the response including a multimodal vector of a resource specified in the request. The received multimodal vector is processed as necessary, such as format conversion, normalization, etc., for subsequent use in the recommendation system. The processed multimodal vector is integrated into relevant modules of the recommendation system, such as a feature extraction module, a similarity calculation module, etc. These multimodal vectors are used to enhance the recommendation system's understanding and recommendation capabilities for new resources.

[0114] For example, suppose a video recommendation platform wants to recommend newly released movies to its users. The platform first determines the first target resource set, that is, the movie types or directors that the user may be interested in, through the user's historical behavior data. Then, the platform sends a request to the multimodal vector storage system to obtain the multimodal vectors of these movies and newly released movies (including vectors corresponding to text descriptions, poster images, trailer videos, etc.). Finally, the platform uses these multimodal vectors to calculate the similarity between the user and the new movie, and recommends movies with higher similarity to the user to improve the user experience.

[0115] In this way, by introducing multimodal vectors, the recommendation system can more comprehensively understand the characteristics and content of resources, thereby more accurately matching user interests and needs. For newly released resources, even if there is a lack of user interaction data, they can be initially displayed and promoted in the recommendation system through their multimodal vectors, accelerating their cold start process. The introduction of the multimodal vector storage system enables the recommendation system to easily handle resources of different types and formats, enhancing the scalability and flexibility of the system.

[0116] In some embodiments, Figure 2 As shown, the resource cold start method based on the large model also includes:

[0117] S201: Calculating similarity between the multimodal vector of the new resource and the multimodal vector of the published resource to identify a second target resource set having a similarity with the new resource higher than a second threshold;

[0118] S202: Generate recommendation information for new resources using historical data of published resources in the second target resource set;

[0119] S203: Push new resources to the target object in the form of recommendation information.

[0120] In the embodiments of the present disclosure, similarity calculation refers to the process of measuring the similarity between two or more objects. In a recommendation system, similarity calculation is often used to discover the intrinsic connection between user interests or resources, or the intrinsic connection between resources.

[0121] In the embodiments of the present disclosure, recommendation information refers to information used to recommend relevant resources or services to users. Recommendation information generally includes a description of the resource, highlights, reasons for recommendation, and other content.

[0122] In some implementations, the multimodal vector library of published resources is traversed to calculate the similarity between the new resource and each published resource one by one. The similarity calculation can adopt various metrics such as cosine similarity and Euclidean distance, and the specific selection depends on the data characteristics and business requirements.

[0123] In some implementations, a second threshold is set to determine the similarity level, and only published resources with a similarity higher than the threshold are included in the second target resource set.

[0124] In some implementations, for each published resource in the second target resource set, historical data is extracted, such as user ratings, comments, page views, conversion rates, etc. These historical data are analyzed to understand user preferences and reactions to similar resources.

[0125] In some implementations, based on the analysis results, recommendation information for the new resource is generated. The recommendation information may include highlights of the new resource, similarities and differences with similar resources, reasons why the user may be interested, etc.

[0126] In some implementations, the generated recommendation information is pushed to the target object in an appropriate form (such as a pop-up window, a recommendation list, an email notification, etc.) The push effect is monitored, and the feedback data of the target object on the new resource is collected for subsequent optimization and improvement.

[0127] For example, suppose an e-commerce platform has newly launched a smartwatch with multiple functions and innovations. The platform first extracts the multimodal vectors of the watch, including product descriptions, pictures, videos, etc. Then, the platform calculates the similarity between these vectors and the multimodal vectors of the released smartwatches, and finds smartwatches with a similarity higher than the set threshold as references. Next, the platform analyzes the historical sales data, user comments and other information of these reference smartwatches, and finds that users are particularly interested in health monitoring, sports tracking and other functions of smartwatches. Based on this information, the platform generates recommendation information for new smartwatches, emphasizing its health management and sports assistance functions, and pushes it to users who may be interested.

[0128] In this way, by utilizing the historical data of similar resources and user feedback, recommendation information is quickly generated for new resources and pushed to target users, effectively shortening the cold start cycle of new resources. Similarity calculation based on multimodal vectors can more accurately capture the intrinsic connection between resources, thereby generating recommendation information that is more in line with user interests and needs, and improving recommendation accuracy. Users can access new resources related to their interests in a timely manner, improving user experience.

[0129] In some embodiments, the resource cold start method based on the large model further includes:

[0130] Determine a target group based on the objects corresponding to the second target resource set, where the target group is the active push object of the new resource;

[0131] Push new resources to the terminals corresponding to the target group;

[0132] Gather feedback from target groups on new resources;

[0133] Based on feedback information, the multimodal large model is dynamically adjusted and optimized.

[0134] In the resource cold start method, it further includes determining target groups based on objects corresponding to the second target resource set, pushing new resources to these target groups, and collecting feedback information to dynamically adjust and optimize the multimodal large model.

[0135] In some implementations, determining the target group includes: analyzing the user group characteristics of each published resource in the second target resource set, including region, interest preference, etc. Using these characteristic information, through cluster analysis, user profiling and other technical means, determine the user group with high similarity to the new resource as the target group. The determination of the target group should comprehensively consider the user's historical behavior data, real-time behavior data and potential needs to improve the accuracy of the push.

[0136] In some implementations, new resources are pushed to the corresponding terminals (such as mobile phones, computers, tablets, etc.) according to the determined target groups. The push methods may include message push, email notification, pop-up window in the application, etc. When pushing, attention should be paid to the personalization and customization of information, and differentiated recommended content should be provided according to the interests, preferences and needs of different users.

[0137] In some implementations, feedback information from the target group on the new resource is collected through user behavior data (such as clicks, browsing, purchases, comments, etc.) and active feedback (such as questionnaires, user reviews, etc.). The feedback information is sorted and analyzed to extract useful information points, such as user satisfaction with the new resource, improvement suggestions, etc.

[0138] In some embodiments, dynamically adjusting and optimizing the multimodal big model includes: dynamically adjusting and optimizing the multimodal big model based on the collected feedback information. This includes adjusting the parameters of the model, updating the training data of the model, improving the algorithm of the model, etc. Through continuous learning and optimization, the multimodal big model's ability to understand and recommend new resources is improved, making it more in line with the actual needs and interests of users.

[0139] For example, suppose a music platform has just launched an original song. The platform first uses multimodal vector technology to calculate the similarity between the song and the released songs, and finds the user group corresponding to the songs with high similarity as the target group. Then, the platform pushes the new song to these target users through pop-up windows in the app. Within a period of time after the push, the platform collects feedback information such as user clicks, plays, shares, and comments on the song, and finds that users have a high evaluation of the melody and lyrics of the song. Based on this feedback information, the platform optimizes and adjusts the multimodal large model to improve its ability to recommend similar new songs.

[0140] In this way, by accurately determining the target group and pushing new resources in a personalized manner, the accuracy of recommendations and user satisfaction are improved. By quickly collecting user feedback and optimizing the model, the cold start process of new resources is accelerated, allowing them to be accepted and recognized by users more quickly. By continuously optimizing the recommendation algorithm and model, users are provided with recommended content that better suits their interests and needs, enhancing the user experience.

[0141] Figure 3 The schematic diagram of the architecture of resource cold start based on the large model is shown as Figure 3 The resource cold start based on the large model is divided into two parts: offline and online. The offline part is mainly the fine-tuning of the multimodal large model, and the online part includes the application of two recall methods: User to Item (U2I) and Item to Item (I2I).

[0142] Among them, the offline process includes: extracting the modal data corresponding to each resource from the resource image and text dataset, the resource text dataset, and the resource audio dataset, and using the multimodal fusion model to obtain the multimodal fusion feature vector of each resource. In order to improve the accuracy of personalized recommendations, the interactive behavior data between resources with historical distribution records and users are mined from offline logs, and the CB2CF model is used for fine-tuning so that the model can better adapt to the data distribution and feature representation of specific recommendation scenarios. Use the fine-tuned multimodal model to calculate the multimodal vectors of newly released resources. The multimodal vectors of all resources are stored in the key-value storage system (Redis), and the new resource vectors are used to build a multimodal vector index library for real-time calls by online requests.

[0143] Among them, the online process includes U2I recall process and I2I recall process.

[0144] Among them, the U2I recall process includes:

[0145] Parse user-related feature data from online real-time requests;

[0146] Use the CF model used when fine-tuning the CB2CF model to estimate the user CF vector corresponding to the current request;

[0147] Use the user CF vector to index the multimodal vector index library built in the offline part, and take the top K resources with similarity as the U2I recall result (i.e., the first recall result).

[0148] The I2I recall process includes:

[0149] Based on the user data obtained through analysis and combined with offline logs, the resource list consumed by the current requesting user is analyzed, and the user-consumed resources are filtered using a posteriori indicators such as click-through rate to obtain a resource set that satisfies the user (i.e., the first target resource set);

[0150] Use the resource collection that satisfies the user to request Redis in the offline part to obtain the multimodal vector corresponding to the target resource;

[0151] Use the multimodal vector of the target resource to index the multimodal vector index library built in the offline part, and take the top M resources with similarity as the I2I recall result (i.e., the second recall result).

[0152] The U2I recall result list and the I2I recall result list are deduplicated and returned to the downstream sorting model.

[0153] Compared with the common recommendation system resource cold start method, the resource cold start method disclosed in this disclosure has the following two significant advantages:

[0154] First, multimodal fusion can significantly improve the model's ability to learn and understand resource attributes, and more comprehensively and accurately characterize the resource's own attribute characteristics, which is especially important in the cold start stage when resource and user interaction behaviors are sparse.

[0155] Second, the multimodal large model used in the present disclosure has a large number of parameters and can learn more complex data patterns and knowledge representations. Its ability to understand graphics or text modalities is also stronger than that of single-modal models. The knowledge learned by the large model during training is more universal and can maintain high accuracy and stability when processing data from different fields.

[0156] The solution proposed in this disclosure can effectively improve the cold start efficiency of new resources in the recommendation system, and the improvement is particularly obvious for long-tail new resources.

[0157] The present disclosure provides a resource cold start device based on a large model, such as Figure 4 As shown, the device may include: a parsing module 401, used to parse the feature data of the target object from the online request; a first determination module 402, used to determine the vector of the target object based on the feature data of the target object, and obtain a first recall result from the multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores the multimodal vectors of the new resources, and the multimodal vectors of the new resources are extracted from the fine-tuned multimodal large model; a second determination module 403, used to determine the first target resource set based on the feature data of the target object, and obtain a second recall result from the multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set; an integration and sending module 404, used to perform integration processing based on the first recall result and the second recall result, and send the integration result to the recommendation system, so that the recommendation system determines the new resource to be recommended based on the integration result.

[0158] In some embodiments, the device also includes: a first extraction module, used to extract multiple modal data from multiple data sets corresponding to each published resource; an acquisition module, used to input the multiple modal data corresponding to each published resource into a multimodal large model to obtain a multimodal vector for each published resource; a fine-tuning module, used to use the interaction data between the published resources and the objects mined from the first offline log to fine-tune the multimodal large model to obtain a fine-tuned multimodal large model.

[0159] In some embodiments, the device also includes: a second extraction module, used to extract multimodal vectors of new resources using the fine-tuned multimodal large model; and a construction module, used to construct a multimodal vector index library based on the multimodal vectors of the new resources.

[0160] In some embodiments, the fine-tuning module is used to: utilize the interaction data between published resources and objects mined from the first offline log to fine-tune the multimodal large model based on the CB2CF model.

[0161] In some embodiments, the first determination module 402 includes: a first determination submodule, configured to determine the vector of the target object based on the CF model used when fine-tuning the multimodal large model based on the CB2CF model.

[0162] In some embodiments, the second determination module 403 includes: an acquisition submodule, used to obtain a list of published resources consumed by the target object from the second offline log based on the characteristic data of the target object; a screening submodule, used to screen out published resources whose posterior indicators are greater than a first threshold from the list of published resources to form a first target resource set.

[0163] In some embodiments, the second determination module 403 also includes: a second determination submodule, used to request the multimodal vector storage system based on the first target resource set to obtain the multimodal vector corresponding to the target resource in the first target resource set, and the multimodal vector storage system stores the multimodal vectors of the published resources and the multimodal vectors of the new resources.

[0164] In some embodiments, the device also includes: a third determination module, used to calculate the similarity between the multimodal vector of the new resource and the multimodal vector of the published resource to identify a second target resource set whose similarity with the new resource is higher than a second threshold; a generation module, used to generate recommendation information for the new resource using the historical data of the published resources in the second target resource set; and a first push module, used to push the new resource to the target object in the form of recommendation information.

[0165] In some embodiments, the device also includes: a fourth determination module, used to determine the target group based on the object corresponding to the second target resource set, the target group is the active push object of the new resource; a second push module, used to push the new resource to the terminal corresponding to the target group; a collection module, used to collect feedback information from the target group regarding the new resource; an adjustment and optimization module, used to dynamically adjust and optimize the multimodal large model based on the feedback information.

[0166] For the description of specific functions and examples of each module and submodule of the device in the embodiment of the present disclosure, reference can be made to the relevant description of the corresponding steps in the above method embodiment, which will not be repeated here.

[0167] The large model-based resource cold start device of the disclosed embodiment realizes personalized recommendations in the absence of historical data support by combining the feature data of the target object and the multimodal vector of the new resource. This not only improves the accuracy and diversity of the recommendations, but also accelerates the exposure and distribution of new resources, thereby promoting the effective use of new resources.

[0168] The embodiment of the present disclosure provides a scenario schematic diagram of a resource cold start method based on a large model, such as Figure 5 shown.

[0169] As mentioned above, the resource cold start method based on the large model provided by the embodiment of the present disclosure is applied to electronic devices. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices.

[0170] Specifically, the electronic device can perform the following operations:

[0171] Parse the characteristic data of the target object from the online request;

[0172] Determine a vector of the target object based on feature data of the target object, and obtain a first recall result from a multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores multimodal vectors of new resources, and the multimodal vectors of the new resources are extracted from the fine-tuned multimodal macro model;

[0173] Determine a first target resource set based on the feature data of the target object, and obtain a second recall result from a multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set;

[0174] An integration process is performed based on the first recall result and the second recall result, and the integration result is sent to the recommendation system, so that the recommendation system determines a new resource to be recommended based on the integration result.

[0175] It should be understood that Figure 5 The scene diagram shown is only illustrative and not restrictive. Those skilled in the art can Figure 5 Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0176] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0177] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0178] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, 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.

[0179] like Figure 6As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 to a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0180] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0181] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a CPU, a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a resource cold start method based on a large model. For example, in some embodiments, the resource cold start method based on a large model may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the resource cold start method based on the large model described above may be executed. Alternatively, in other embodiments, the computing unit 601 may be configured to execute the large model-based resource cold start method in any other appropriate manner (for example, by means of firmware).

[0182] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems on chips (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 can be executed and / or interpreted on a programmable system including 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.

[0183] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0184] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (EPROM), a 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.

[0185] 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 cathode ray tube (CRT) or a liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and a 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).

[0186] 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: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

[0187] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0188] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0189] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A resource cold start method based on a large model, comprising: Parse the characteristic data of the target object from the online request; Determine the vector of the target object based on the feature data of the target object, and obtain a first recall result from a multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores multimodal vectors of new resources, and the multimodal vectors of the new resources are extracted from the fine-tuned multimodal macro model; Determine a first target resource set based on the feature data of the target object, and obtain a second recall result from the multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set; An integration process is performed based on the first recall result and the second recall result, and the integration result is sent to a recommendation system, so that the recommendation system determines a new resource to be recommended based on the integration result.

2. The method according to claim 1, further comprising: Extract multi-modal data from multiple datasets corresponding to each published resource; Input the multi-modal data corresponding to each published resource into the multi-modal big model to obtain the multi-modal vector of each published resource; The multimodal large model is fine-tuned using the interaction data between the published resources and the objects mined from the first offline log to obtain a fine-tuned multimodal large model.

3. The method according to claim 1, wherein: Using the published resource-object interaction data mined from the first offline log, fine-tuning the multimodal large model includes: The content-based collaborative filtering (CB2CF) model is adopted to fine-tune the multimodal large model using the interaction data of published resources and objects mined from the first offline log.

4. The method according to claim 3, wherein: The determining the vector of the target object based on the feature data of the target object comprises: The collaborative filtering CF model of the CB2CF model is adopted to determine the vector of the target object based on the feature data of the target object.

5. The method according to claim 1, wherein: The determining the first target resource set based on the characteristic data of the target object includes: Based on the characteristic data of the target object, obtaining a list of published resources consumed by the target object from a second offline log; Published resources whose a posteriori indicators are greater than a first threshold are screened out from the published resource list to form the first target resource set.

6. The method according to claim 1, further comprising: Based on the first target resource set, a multimodal vector storage system is requested to obtain multimodal vectors corresponding to target resources in the first target resource set, wherein the multimodal vector storage system stores multimodal vectors of published resources and multimodal vectors of new resources.

7. The method according to claim 1, further comprising: Calculating similarity between the multimodal vector of the new resource and the multimodal vector of the published resource to identify a second target resource set having a similarity with the new resource higher than a second threshold; Generate recommendation information for the new resource by using historical data of published resources in the second target resource set; The new resource is pushed to the target object in the form of the recommendation information.

8. The method according to claim 7, further comprising: Determine a target group based on the objects corresponding to the second target resource set, where the target group is an active push object of the new resource; Pushing the new resource to terminals corresponding to the target group; Collecting feedback information from the target group regarding the new resource; Based on the feedback information, the multimodal large model is dynamically adjusted and optimized.

9. A resource cold start device based on a large model, comprising: A parsing module is used to parse the characteristic data of the target object from the online request; A first determination module is used to determine the vector of the target object based on the feature data of the target object, and obtain a first recall result from a multimodal vector index library based on the vector of the target object, wherein the multimodal vector index library stores the multimodal vector of the new resource, and the multimodal vector of the new resource is extracted from the fine-tuned multimodal large model; A second determination module, configured to determine a first target resource set based on the feature data of the target object, and obtain a second recall result from the multimodal vector index library based on the multimodal vectors corresponding to the target resources in the first target resource set; The integration and sending module is used to perform integration processing based on the first recall result and the second recall result, and send the integration result to the recommendation system, so that the recommendation system determines the new resource to be recommended based on the integration result.

10. The apparatus according to claim 9, further comprising: A first extraction module is used to extract multiple modal data from multiple data sets corresponding to each published resource; An acquisition module is used to input the multi-modal data corresponding to each published resource into the multi-modal large model to obtain the multi-modal vector of each published resource; The fine-tuning module is used to fine-tune the multimodal large model by using the interaction data between the published resources and the objects mined from the first offline log to obtain the fine-tuned multimodal large model.

11. The device according to claim 10, wherein: The fine-tuning module is used to: The content-based collaborative filtering (CB2CF) model is adopted to fine-tune the multimodal large model using the interaction data of published resources and objects mined from the first offline log.

12. The device according to claim 11, wherein The first determining module includes: The first determination submodule is used to adopt the collaborative filtering CF model of the CB2CF model to determine the vector of the target object based on the feature data of the target object.

13. The device according to claim 9, wherein: The second determining module includes: an acquisition submodule, configured to acquire, from a second offline log, a list of published resources consumed by the target object based on the characteristic data of the target object; The screening submodule is used to screen out the published resources whose a posteriori indicators are greater than a first threshold from the published resource list to form the first target resource set.

14. The device according to claim 9, wherein: The second determining module further includes: The second determination submodule is used to request a multimodal vector storage system based on the first target resource set to obtain a multimodal vector corresponding to a target resource in the first target resource set, wherein the multimodal vector storage system stores multimodal vectors of published resources and multimodal vectors of new resources.

15. The apparatus according to claim 9, further comprising: A third determination module is used to calculate the similarity between the multimodal vector of the new resource and the multimodal vector of the published resource to identify a second target resource set having a similarity with the new resource higher than a second threshold; A generating module, configured to generate recommendation information for the new resource by using the historical data of the published resources in the second target resource set; The first push module is used to push the new resource to the target object in the form of the recommendation information.

16. The apparatus according to claim 15, further comprising: A fourth determination module, configured to determine a target group based on the objects corresponding to the second target resource set, wherein the target group is an active push object of the new resource; A second push module, used to push the new resource to the terminals corresponding to the target group; A collection module, used to collect feedback information from the target group regarding the new resource; An adjustment and optimization module is used to dynamically adjust and optimize the multimodal large model based on the feedback information.

17. An electronic device comprising: at least one processor; as well as a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 8.

18. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are for causing a computer to execute a method according to any one of claims 1-8.

19. A computer program product comprising a computer program stored on a storage medium, the computer program implementing the method according to any one of claims 1 to 8 when executed by a processor.

Images