Object recommendation method and apparatus, electronic device, and readable storage medium

Abstract

Provided are an object recommendation method and apparatus, an electronic device, and a readable storage medium. The method comprises: by means of a large language model, processing user-side data and attribute data corresponding to an object to be recommended, so as to obtain user preference data and extended data corresponding to said object; performing text encoding processing on the user preference data and the extended data to obtain a user preference vector and an extended vector corresponding to said object; on the basis of a gating mechanism, processing the user preference vector and the extended vector to obtain a gating vector corresponding to the user preference vector and a gating vector corresponding to the extended vector; performing spatial alignment processing on the two gating vectors to obtain a target user preference vector and a target extended vector; on the basis of the data and the vectors, determining a recommendation score corresponding to said object; and on the basis of a preset recommendation score threshold value and the recommendation score corresponding to said object, determining target object data and sending the target object data.

Description

Recommended methods, devices, electronic devices, and readable storage media

[0001] Related applications

[0002] This application claims priority to Chinese invention patent with patent application number 202510182931.6, application date February 19, 2025, entitled "Object Recommendation Method, Apparatus, Electronic Device and Readable Storage Medium". Technical Field

[0003] This application relates to, but is not limited to, the technical field of object recommendation, and in particular to an object recommendation method, apparatus, electronic device, and readable storage medium. Background Technology

[0004] Object recommendation methods are widely used in e-commerce platforms, social media, content delivery networks, and other fields. Traditional object recommendation methods mainly rely on users' historical behavior data, the attribute features of the objects to be recommended, and simple matching algorithms to predict users' preferences for specific objects and make recommendations accordingly. However, traditional recommendation methods only perform shallow feature extraction when processing user-side data and object-side data, resulting in insufficient accuracy and personalization of recommendation results. When matching user preferences and object-side features, conventional methods use fixed weight allocation or simple similarity calculations, lacking dynamic adjustment and adaptive mechanisms. To address these issues, existing technologies typically introduce more complex feature engineering to extract deeper features from user-side data and object-side data; and utilize machine learning algorithms to improve the accuracy and generalization ability of recommendation models. However, complex feature engineering and machine learning algorithms increase computational complexity and model training costs, and lack effective dynamic adjustment and adaptive mechanisms when processing user preferences and object-side features.

[0005] It is evident that existing technologies suffer from insufficient processing of user-side data and data on the objects to be recommended, and lack dynamic adjustment and adaptive mechanisms, resulting in low accuracy and personalization of recommendation results. Summary of the Invention

[0006] In view of this, embodiments of the present disclosure provide an object recommendation method, apparatus, electronic device, and readable storage medium to solve the problem in the prior art that the accuracy and personalization of recommendation results are low due to insufficient processing of user-side data and data of objects to be recommended, and the lack of dynamic adjustment and adaptive mechanisms.

[0007] A first aspect of this disclosure provides an object recommendation method, comprising: processing user-side data and attribute data corresponding to the object to be recommended using a large language model to obtain user preference data and extended data corresponding to the object to be recommended; performing text encoding processing on the user preference data and the extended data corresponding to the object to be recommended to obtain a user preference vector and an extended vector corresponding to the object to be recommended; processing the user preference vector and the extended vector corresponding to the object to be recommended based on a gating mechanism to obtain a gating vector corresponding to the user preference vector and a gating vector corresponding to the extended vector; performing spatial alignment processing on the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector to obtain a target user preference vector and a target extended vector corresponding to the object to be recommended; determining a recommendation score corresponding to the object to be recommended based on the user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended; determining target object data based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and sending the target object data to a target terminal device so that the target object data is displayed in the current graphical user interface of the target terminal device.

[0008] A second aspect of this disclosure provides an object recommendation apparatus, comprising: a first processing module, configured to process user-side data and attribute data corresponding to the object to be recommended respectively using a large language model to obtain user preference data and extended data corresponding to the object to be recommended; a second processing module, configured to perform text encoding processing on the user preference data and the extended data corresponding to the object to be recommended respectively to obtain a user preference vector and an extended vector corresponding to the object to be recommended; a third processing module, configured to process the user preference vector and the extended vector corresponding to the object to be recommended respectively based on a gating mechanism to obtain a gating vector corresponding to the user preference vector and a gating vector corresponding to the extended vector; and a fourth processing module, configured to process user-side data and attribute data corresponding to the object to be recommended respectively using a large language model to obtain user preference data and extended data corresponding to the object to be recommended respectively; and a fourth processing module, configured to process user-side data and attribute data corresponding to the object to be recommended respectively using a gating mechanism to obtain a gating vector corresponding to the user preference vector and a gating vector corresponding to the extended vector; and a fourth processing module, configured to process user-side data and attribute data corresponding to the object to be recommended respectively using a gating mechanism to obtain user preference data and extended data corresponding to the extended vector. The gating vectors corresponding to the user preference vector and the gating vectors corresponding to the extension vector are spatially aligned to obtain the target user preference vector and the target extension vector corresponding to the object to be recommended. The determination module is used to determine the recommendation score corresponding to the object to be recommended based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extension vector corresponding to the object to be recommended. The fifth processing module is used to determine the target object data based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and send the target object data to the target terminal device so that the target object data is displayed in the current graphical user interface of the target terminal device.

[0009] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0010] A fourth aspect of this disclosure provides a readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0011] The beneficial effects of this disclosure compared to the prior art are as follows: By performing text generation processing on user-side data and attribute data corresponding to the object to be recommended using a large language model, user preference data containing user preference information can be obtained by processing the user-side data, and extended data corresponding to the object to be recommended containing relevant information can be obtained by processing the attribute data corresponding to the object to be recommended. Furthermore, text encoding processing can be performed on the user preference data and the extended data corresponding to the object to be recommended to obtain a user preference vector and an extended vector corresponding to the object to be recommended. Gating processing can be performed on the user preference vector and the extended vector corresponding to the object to be recommended to obtain a gated vector corresponding to the user preference vector and a gated vector corresponding to the extended vector. Spatial alignment processing can be performed on the two gated vectors to align the gated vector corresponding to the user preference vector and the extended vector. The gating vector corresponding to the quantity is mapped to the object recommendation space for data alignment, resulting in the target user preference vector and the target extended vector corresponding to the object to be recommended. Based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended, the recommendation score corresponding to the object to be recommended is determined. Then, based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, the target object data can be determined and sent to the current graphical user interface of the target terminal device for display. This improves the matching accuracy between user preference information and attribute information of the object to be recommended, enhances the flexibility and adaptability of data processing, and improves the correlation and accuracy of vector representation through gating mechanism and space alignment processing, thereby enhancing the personalization and accuracy of recommendation results. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 is a schematic diagram of an application scenario of this disclosure embodiment;

[0014] Figure 2 is a flowchart illustrating an object recommendation method provided in an embodiment of this disclosure;

[0015] Figure 3 is a schematic diagram of the structure of an object recommendation device provided in an embodiment of this disclosure;

[0016] Figure 4 is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0017] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, so as to provide a thorough understanding of the embodiments of this disclosure. However, those skilled in the art will understand that this disclosure may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this disclosure with unnecessary detail.

[0018] It should be noted that the user information (including but not limited to terminal device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties, and the technical solution involved in this disclosure is automated processing through an object recommendation model that has been trained.

[0019] A method and apparatus for recommending objects according to embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings.

[0020] Figure 1 is a schematic diagram of an application scenario according to an embodiment of this disclosure. The application scenario may include terminal devices 1, 2 and 3, server 4, and network 5.

[0021] Terminal devices 1, 2, and 3 can be hardware or software. When terminal devices 1, 2, and 3 are hardware, they can be various electronic devices with displays and supporting communication with server 4, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 1, 2, and 3 are software, they can be installed in the aforementioned electronic devices. Terminal devices 1, 2, and 3 can be implemented as multiple software programs or software modules, or as a single software program or software module; this disclosure does not limit this. Furthermore, various applications can be installed on terminal devices 1, 2, and 3, such as data processing applications, instant messaging tools, social platform software, search applications, shopping applications, etc.

[0022] Server 4 can be a server that provides various services, such as a backend server that receives requests sent by terminal devices with which it has established communication connections. This backend server can receive and analyze the requests sent by the terminal devices and generate processing results. Server 4 can be a single server, a server cluster consisting of several servers, or a cloud computing service center. This disclosure embodiment does not limit this.

[0023] It should be noted that server 4 can be either hardware or software. When server 4 is hardware, it can be various electronic devices that provide various services to terminal devices 1, 2, and 3. When server 4 is software, it can be multiple software programs or software modules that provide various services to terminal devices 1, 2, and 3, or it can be a single software program or software module that provides various services to terminal devices 1, 2, and 3. This disclosure does not limit the scope of the embodiments.

[0024] Network 5 can be a wired network using coaxial cable, twisted pair, and fiber optic connection, or it can be a wireless network that enables interconnection of various communication devices without wiring, such as Bluetooth, Near Field Communication (NFC), and Infrared. This disclosure does not limit the scope of the embodiments.

[0025] Users can establish a communication connection with server 4 via network 5 through terminal devices 1, 2, and 3 to receive or send information. Specifically, server 4 can obtain user-side data, attribute data corresponding to the object to be recommended, and historical interaction context data corresponding to the object to be recommended through terminal devices 1, 2, and 3. Then, it can perform text generation processing on the user-side data and the attribute data corresponding to the object to be recommended using a large language model. Processing the user-side data yields user preference data containing user preference information, and processing the attribute data corresponding to the object to be recommended yields extended data containing relevant information about the object to be recommended. Furthermore, text encoding processing can be performed on the user preference data and the extended data corresponding to the object to be recommended to obtain a user preference vector and an extended vector corresponding to the object to be recommended. Gating processing can then be applied to both the user preference vector and the extended vector corresponding to the object to be recommended. By obtaining the gate vectors corresponding to the user preference vector and the extended vector, spatial alignment can be performed on both types of gate vectors. The gate vectors corresponding to the user preference vector and the extended vector can be mapped to the object recommendation space for data alignment, resulting in the target user preference vector and the target extended vector corresponding to the object to be recommended. Based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended, the recommendation score corresponding to the object to be recommended can be determined. Then, based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, the target object data can be determined and sent to the current graphical user interface of the target terminal device for display.

[0026] It should be noted that the specific types, quantities, and combinations of terminal devices 1, 2, and 3, server 4, and network 5 can be adjusted according to the actual needs of the application scenario, and this disclosure embodiment does not impose any restrictions on this.

[0027] Figure 2 is a flowchart illustrating an object recommendation method provided in an embodiment of this disclosure. The object recommendation method in Figure 2 can be executed by the server in Figure 1. As shown in Figure 2, the object recommendation method includes:

[0028] Step 201: Process the user-side data and the attribute data corresponding to the object to be recommended using a large language model to obtain user preference data and extended data corresponding to the object to be recommended.

[0029] Specifically, a pre-trained large language model can be used to generate text from user-side data and attribute data corresponding to the objects to be recommended, resulting in user preference data and extended data corresponding to the objects to be recommended. User-side data can be user attribute data, including but not limited to user terminal device input information and / or historical behavior information. Attribute data corresponding to the objects to be recommended can be the description, price, category, and / or user reviews of the objects to be recommended, without limitation. User preference data can include information related to the user's interests and needs, while extended data corresponding to the objects to be recommended can enrich and expand the attribute information of the objects to be recommended, including but not limited to the object's historical interaction information, award information, and / or sales information. This enhances the personalization of the object recommendation model, improves the accuracy of the recommendation results, and enhances the precision of matching user needs with object attributes.

[0030] The large language model can be a deep learning model trained on a large-scale corpus, capable of processing and understanding complex textual information. User-side data can be a dataset related to user behavior and attributes, representing information entered by the user's terminal device and / or historical behavior, etc., without limitation. The object to be recommended can be an object prepared to be recommended to the user, including but not limited to products, text, videos, or images. The attribute data corresponding to the object to be recommended can refer to data representing the characteristics and attributes of the object, such as product specifications, price, and / or reviews. User preference data can be obtained by processing user-side data through the large language model, representing the user's personalized needs and interests. The extended data corresponding to the object to be recommended can be feature information obtained by processing the attribute data of the object to be recommended through the large language model, including but not limited to the object's historical interaction information, award information, and / or release information.

[0031] For example, on an e-commerce platform, the recommended item could be athletic shoes. The large language model can then process user-side data, such as information entered from the user's device, search history, browsing history, and purchasing behavior, to obtain the user's preference data for athletic shoes, including brand preference, price range, and style preference. Simultaneously, the large language model can process the athletic shoe's attribute data, such as material, weight, breathability, and user reviews, to generate corresponding extended data for the athletic shoe, including potential characteristics such as applicable scenarios and wearing comfort.

[0032] Step 202: Perform text encoding processing on the user preference data and the extended data corresponding to the object to be recommended, respectively, to obtain the user preference vector and the extended vector corresponding to the object to be recommended.

[0033] Specifically, text encoding models in Natural Language Processing (NLP), including but not limited to Word2Vec or Bidirectional Encoder Representations from Transformers (BERT), can be used to process user preference data and the extended data corresponding to the recommended object. This transforms text information into numerical vectors in a high-dimensional vector space, resulting in user preference vectors and extended vectors corresponding to the recommended object. The user preference vector, obtained through text encoding, can represent the user's interests, preferences, or needs. The extended vector corresponding to the recommended object can be a numerical vector extracted and encoded from the extended data of the recommended object. This enhances the matching accuracy between user preferences and the recommended object, improving the personalization of the recommendation model. By converting unstructured text data into structured numerical vectors through text encoding, the recommendation algorithm can analyze the data more effectively, improving the accuracy of the recommendation results.

[0034] For example, in e-commerce platforms, user preference data can include information entered by the user's terminal device, browsing history, purchase records, and search keywords. Extended data for the recommended object can include the product's title, description, and / or user reviews. Through text encoding, the aforementioned text data can be converted into corresponding numerical vectors: the user's preference vector and the extended vector corresponding to the recommended object.

[0035] Step 203: Based on the gating mechanism, process the user preference vector and the extended vector corresponding to the object to be recommended respectively to obtain the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector.

[0036] Specifically, a gating mechanism can be used to perform a nonlinear transformation on the user preference vector and the extended vector corresponding to the object to be recommended, thereby obtaining the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector, respectively. This enhances the dynamic adjustment capability of the object recommendation model, allowing it to adjust its focus according to different users and objects to be recommended. This improves the accuracy and personalization of the object recommendation model, as well as its robustness and adaptability.

[0037] Among them, the gating mechanism can be a network structure used to control the flow of information. The gating mechanism can determine the information to be retained, discarded, or modified by learning the weights.

[0038] The gating vector corresponding to the user preference vector can be a vector obtained by processing the user preference vector through a gating mechanism. The gating vector corresponding to the user preference vector can be used to characterize the recognition and emphasis of information in user preferences by the recommendation model of the object.

[0039] The gating vector corresponding to the expansion vector can be a vector obtained by processing the expansion vector of the object to be recommended through a gating mechanism. The gating vector corresponding to the expansion vector can be used to characterize the object recommendation model's recognition and emphasis on information in the object features.

[0040] For example, in product recommendations on e-commerce platforms, gating mechanisms can be used to determine the gating vectors corresponding to user preference vectors or product feature extension vectors that are important to the recommendation results, thus obtaining the gating vectors corresponding to the user preference vectors and the extension vectors.

[0041] Step 204: Perform spatial alignment processing on the gating vector corresponding to the user preference vector and the gating vector corresponding to the expansion vector respectively to obtain the target user preference vector and the target expansion vector corresponding to the object to be recommended.

[0042] Specifically, attention mechanisms or matrix multiplication, among other methods (not limited here), can be used to align the spatial dimensions of the gating vectors corresponding to the user preference vector and the gating vectors corresponding to the extension vectors. This yields the target user preference vector and the target extension vector corresponding to the object to be recommended. This enhances the correlation between user preference information and the feature information of the object to be recommended, improves the personalization of the object recommendation model, and increases the accuracy of the object recommendation results.

[0043] The target user preference vector can be a vector representation obtained by spatial alignment of the gating vector corresponding to the user preference vector. The target user preference vector can be used to represent the user's personal interests and preference information.

[0044] The target extension vector corresponding to the object to be recommended can be a vector representation obtained by processing the gating vector corresponding to the extension vector through spatial alignment, so that the target extension vector corresponding to the object to be recommended and the target user preference vector are aligned in the same spatial dimension.

[0045] For example, in a recommendation system of an e-commerce platform, a user preference vector can contain information about a user's preferences for product type, price, and brand, while the corresponding extended vector for a product can contain attributes such as the product's description, user reviews, and sales volume. Through spatial alignment, we obtain the vector representations of the user's overall preference information and the product's extended information, i.e., the target user preference vector and the target extended vector corresponding to the recommended object.

[0046] Step 205: Determine the recommendation score for the object to be recommended based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, target user preference vector, and target extended vector corresponding to the object to be recommended.

[0047] Specifically, a Multilayer Perceptron (MLP) can be used to calculate and determine the recommendation score of the recommended object for the target user based on user-side data, attribute data of the object to be recommended, historical interaction context data of the object to be recommended, target user preference vector, and target extended vector of the object to be recommended. The historical interaction context data of the object to be recommended includes, but is not limited to, information such as the time, location, and environment of the user's historical interactions with the object to be recommended. This enhances the personalization of the object recommendation model and improves the accuracy and relevance of the recommendation results. By comprehensively considering multiple data sources and features, the recommendation results are made more in line with the user's actual needs.

[0048] The recommendation score for the object to be recommended can be a quantitative indicator, representing its attractiveness and recommendation value to users. This score can be obtained through multilayer perceptron processing based on user-side data, object attribute data, historical interaction context data, target user preference vector, and target extended vector of the object to be recommended. The recommendation score can be used for ranking, filtering recommended objects, and evaluating recommendation effectiveness, but is not limited here.

[0049] For example, on an e-commerce platform, when a user browses a product, the platform can calculate the product's recommendation score for that user based on user-side data such as information entered on the user's terminal device, browsing history, purchase records, and search keywords, as well as attribute data such as the product's category, price, brand, and sales volume, and historical interaction context data such as the time, location, and environment in which the user purchased the product. This data can be combined with user preference vectors and product extended vectors.

[0050] Step 206: Based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, determine the target object data and send the target object data to the target terminal device so that the target object data is displayed in the current graphical user interface of the target terminal device.

[0051] Specifically, a recommendation score can be obtained by evaluating the object to be recommended based on a pre-set recommendation score threshold (this threshold can be a critical value obtained by comprehensively considering factors such as business logic, user preferences, and data quality, and can be used to distinguish whether the object to be recommended is worth recommending). When the recommendation score of the object to be recommended is greater than or equal to the pre-set recommendation score threshold, the object data is determined to be the target object data. When the recommendation score of the object to be recommended is less than the pre-set recommendation score threshold, the object data is not the target object data, and the target object data can then be sent to the target terminal device. The target terminal device can be used to display the target object data on the currently displayed graphical user interface. The target terminal device can be the user's terminal device, including but not limited to smartphones, tablets, or computers. The current graphical user interface can be an interface that the user can see and interact with on the screen. This enhances the accuracy of the recommendation results, improves the efficiency of users obtaining useful information, enhances the user experience, and reduces the time cost for users to filter information by setting a recommendation score threshold, thereby increasing user satisfaction with the recommended content.

[0052] The preset recommendation score threshold serves as the score limit for determining whether an object to be recommended is the target object. When the recommendation score is greater than or equal to the preset recommendation score threshold, the object to be recommended can be determined as the target object data.

[0053] The target object data can be object data obtained through filtering, and the recommendation score of the target object data is greater than or equal to a preset recommendation score threshold. The target object data can be product information, text content, audio content, or video content, etc., and there are no restrictions here.

[0054] The target terminal device can be a device used to receive and display data of the target object. The target terminal device can be a smart device that the user is using or will use, including mobile phones, tablets, computers, etc., without limitation here.

[0055] The current graphical user interface (GUI) is the interface displayed on the screen of a target terminal device that allows the user to interact with it. The GUI can contain various information elements, including but not limited to text, images, videos, and / or operation buttons, and can be used to facilitate information exchange and operational control between the user and the target terminal device.

[0056] For example, in an e-commerce platform, if a product to be recommended has a recommendation score of 90, it can be determined as a target product based on a preset recommendation score threshold, such as 80. Since 90 is greater than 80, the product to be recommended can be sent to the user's mobile phone and displayed on the user's current graphical user interface.

[0057] According to the technical solution provided in this disclosure, text generation processing is performed on user-side data and attribute data corresponding to the object to be recommended using a large language model. User-side data containing user preference information can be obtained by processing the user-side data. Extended data containing relevant information about the object to be recommended can be obtained by processing the attribute data. Then, text encoding processing is performed on the user preference data and the extended data to be recommended to obtain a user preference vector and an extended vector corresponding to the object to be recommended. Gating processing is then performed on the user preference vector and the extended vector to obtain a gated vector corresponding to the user preference vector and a gated vector corresponding to the extended vector. Spatial alignment processing is then performed on the two gated vectors to align the gated vector corresponding to the user preference vector and the extended vector. The gating vector is mapped to the object recommendation space for data alignment, resulting in the target user preference vector and the target extended vector corresponding to the object to be recommended. Based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended, the recommendation score corresponding to the object to be recommended is determined. Then, based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, the target object data can be determined and sent to the current graphical user interface of the target terminal device for display. This improves the matching accuracy between user preference information and attribute information of the object to be recommended, enhances the flexibility and adaptability of data processing, and improves the correlation and accuracy of vector representation through gating mechanism and space alignment processing, thereby enhancing the personalization and accuracy of recommendation results.

[0058] In some embodiments, a large language model is used to process user-side data and attribute data corresponding to the object to be recommended, respectively, to obtain user preference data and extended data corresponding to the object to be recommended. This includes: processing user-side data based on a preset preference generation template to obtain user-side templated data; processing attribute data corresponding to the object to be recommended based on a preset extended generation template to obtain templated attribute data corresponding to the object to be recommended; and processing user-side templated data and templated attribute data corresponding to the object to be recommended separately using a large language model to obtain user preference data and extended data corresponding to the object to be recommended.

[0059] Specifically, user-side data can be structured using preset preference generation templates, configuring the user-side data into the corresponding positions of the preset preference generation templates to obtain user-side templated data; the attribute data of the object to be recommended can be structured using preset extended generation templates to obtain templated attribute data of the object to be recommended. Then, the user-side templated data and the templated attribute data corresponding to the object to be recommended can be input into the large language model for text generation processing to obtain user preference data and extended data corresponding to the object to be recommended.

[0060] The preset preference generation template serves as a prompting engineering framework for transforming user-side data into structured template data. This template can be used to define rules for extracting target information from raw user data and to organize the extracted target information into a format suitable for large language models.

[0061] User-side templated data can be structured data obtained by processing user-side data through templates generated by applying preset preferences.

[0062] The pre-defined extended generation templates serve as a framework for prompting engineering, transforming attribute data corresponding to objects to be recommended into structured template data. These templates can also be used to define rules for extracting target information from raw attribute data and organizing the extracted information into a format suitable for large language models.

[0063] The templated attribute data corresponding to the object to be recommended can be structured data obtained by processing the attribute data corresponding to the object to be recommended through the application of a preset extended generation template.

[0064] For example, in e-commerce platforms, user-side data includes the user's terminal device input information, browsing history, purchase history, and search keywords. The recommended object can be a product on the platform. User-side data can be transformed into templated data through preset preference generation templates, such as converting the user's browsing history into the form "The user browsed {product category}". The attribute data corresponding to the recommended product can be transformed into templated data through preset extended generation templates, such as converting the description information of the recommended product into the form "{product name} is {product category}, and has {features}". The above templated data can then be input into a large language model for text generation processing to obtain the user's preference data and the extended data corresponding to the recommended product.

[0065] According to the technical solution provided in this disclosure, user-side data is structured using a preset preference generation template, and the user-side data is configured to the corresponding position of the preset preference generation template to obtain user-side templated data. The attribute data of the object to be recommended can be structured using a preset extended generation template to obtain templated attribute data of the object to be recommended. Then, the user-side templated data and the templated attribute data corresponding to the object to be recommended can be input into the large language model for text generation processing to obtain user preference data and extended data corresponding to the object to be recommended. This enhances the structuring of the data, improves the efficiency and accuracy of data processing, and enhances the personalized recommendation capability of the object recommendation model.

[0066] In some embodiments, determining the recommendation score for a proposed object based on user-side data, attribute data corresponding to the proposed object, historical interaction context data corresponding to the proposed object, target user preference vector, and target extended vector corresponding to the proposed object includes: concatenating the target user preference vector and the target extended vector corresponding to the proposed object to obtain a recommendation enhancement vector; performing feature extraction processing on the user-side data, attribute data corresponding to the proposed object, and historical interaction context data corresponding to the proposed object to obtain a user-side vector, an attribute vector corresponding to the proposed object, and a historical interaction context vector corresponding to the proposed object; performing vector merging processing on the recommendation enhancement vector, user-side vector, attribute vector corresponding to the proposed object, and historical interaction context vector corresponding to the proposed object to obtain a target vector corresponding to the proposed object; and determining the recommendation score for the proposed object based on the target vector corresponding to the proposed object.

[0067] Specifically, the target user preference vector can be concatenated with the target extended vector corresponding to the object to be recommended to obtain the recommendation enhancement vector. Concatenation can be achieved through direct concatenation, dimension matching concatenation, or linear transformation concatenation, etc., without limitation here. Feature extraction can be performed on user-side data, attribute data corresponding to the object to be recommended, and historical interaction context data corresponding to the object to be recommended to obtain user-side vector, attribute vector corresponding to the object to be recommended, and historical interaction context vector, respectively. Feature extraction can be achieved through convolutional neural networks, word embedding, or recurrent neural networks, etc., without limitation here. Then, the above vectors can be processed by weighted merging or non-linear merging to obtain the target vector corresponding to the object to be recommended, without limitation here. The recommendation score corresponding to the object to be recommended can be calculated based on the target vector corresponding to the object to be recommended, and can be calculated by dot product, cosine similarity, or Euclidean distance, etc.

[0068] The recommendation enhancement vector can be a vector obtained by concatenating the target user preference vector and the target extension vector corresponding to the object to be recommended. The recommendation enhancement vector can be used to represent the user's preference information and the extension information of the object to be recommended. The concatenation can be achieved by direct concatenation, dimension matching concatenation, or linear transformation concatenation, etc., and there is no limitation here.

[0069] The historical interaction context data corresponding to the object to be recommended can be the historical context information of the interaction between the user and the object to be recommended, including but not limited to time, location, weather at the time, and / or user behavior. The historical interaction context data corresponding to the object to be recommended can be obtained by recording and analyzing the user's historical behavior information. This historical interaction context data can be used to assess the user's potential interest in the object to be recommended.

[0070] User side vectors can be vectors obtained by performing feature extraction on user side data. User side vectors can be used to represent the user's feature information, including but not limited to the user's age, gender, and / or interests and preferences. User side vectors can be used to understand user needs and preferences.

[0071] The attribute vector corresponding to the object to be recommended can be a vector obtained by feature extraction processing of the attribute data corresponding to the object to be recommended. The attribute vector corresponding to the object to be recommended can be used to represent the inherent attribute information of the object to be recommended, including but not limited to the price, brand, and / or category of the object to be recommended.

[0072] The historical interaction context vector corresponding to the object to be recommended can be a vector obtained by feature extraction processing of the historical interaction context data corresponding to the object to be recommended. The historical interaction context vector corresponding to the object to be recommended can be used to characterize the contextual features of the user's historical interaction with the object to be recommended, including but not limited to the user's behavior patterns at different times and locations. This historical interaction context vector can be used to analyze the changes in the user's preference for the object to be recommended in different contexts.

[0073] For example, on an e-commerce platform, when a user browses products, a recommendation enhancement vector can be obtained based on user-side data such as information entered by the user's terminal device, historical purchase records, browsing history, and search history; product attribute data such as price, brand, and category; historical interaction context data between the product and the user, such as the time, location, and weather when the user clicked or purchased the product; user preference vector; and product extended vectors such as sales volume and reviews. These vectors can then be merged to obtain the product's target vector, and the recommendation score of the product to be recommended can be calculated based on the target vector.

[0074] According to the technical solution provided in this disclosure, a recommendation enhancement vector is obtained by concatenating the target user preference vector with the target extended vector corresponding to the object to be recommended. Feature extraction can be performed on user-side data, attribute data corresponding to the object to be recommended, and historical interaction context data corresponding to the object to be recommended, respectively, to obtain user-side vector, attribute vector corresponding to the object to be recommended, and historical interaction context vector. Then, the above vectors can be processed by weighted merging or non-linear merging to obtain the target vector corresponding to the object to be recommended. The recommendation score corresponding to the object to be recommended can be calculated based on the target vector corresponding to the object to be recommended, thereby enhancing the personalization of the object recommendation model, improving the accuracy of the recommendation results, and improving the user experience.

[0075] In some embodiments, determining the recommendation score of the object to be recommended based on the target vector of the object to be recommended includes: flattening the target vector of the object to be recommended to obtain a low-dimensional target vector of the object to be recommended; mapping the low-dimensional target vector of the object to be recommended using a multilayer perceptron to obtain a spatial mapping vector of the object to be recommended; and calculating the recommendation score of the spatial mapping vector of the object to be recommended to obtain the recommendation score of the object to be recommended.

[0076] Specifically, the target vector corresponding to the object to be recommended can be flattened to reduce its dimension, resulting in a low-dimensional target vector. This low-dimensional target vector can then be nonlinearly mapped using a multilayer perceptron to obtain a spatial mapping vector. Finally, a recommendation score can be calculated from this spatial mapping vector to obtain the recommendation score for the object to be recommended.

[0077] The target low-dimensional vector corresponding to the object to be recommended can be a low-dimensional representation obtained by flattening the target vector corresponding to the high-dimensional object to be recommended. The flattening process can be achieved by dimensionality reduction methods such as Principal Component Analysis (PCA) or Singular Value Decomposition (SVD), and there is no limitation here.

[0078] A multilayer perceptron can be a type of feedforward artificial neural network model. A multilayer perceptron can consist of an input layer, several hidden layers, and an output layer. The neurons in each layer are fully connected to the neurons in the next layer. The input to output is nonlinearly mapped through nonlinear activation functions, including but not limited to ReLU or sigmoid.

[0079] The spatial mapping vector corresponding to the object to be recommended can be a vector representation obtained by performing a nonlinear transformation on the target low-dimensional vector through a multilayer perceptron. This spatial mapping vector representation can be obtained by processing in the hidden layer or output layer of the multilayer perceptron.

[0080] For example, in e-commerce platforms, a low-dimensional target vector corresponding to the product to be recommended can be obtained through flattening. This low-dimensional target vector can then be mapped using a multilayer perceptron to obtain a spatial mapping vector corresponding to the product. The recommendation score for the product to be recommended can then be calculated based on the spatial mapping vector.

[0081] According to the technical solution provided in this disclosure, by flattening the target vector corresponding to the object to be recommended, the dimensionality of the target vector corresponding to the object to be recommended is reduced, resulting in a low-dimensional target vector corresponding to the object to be recommended. This low-dimensional target vector can be nonlinearly mapped using a multilayer perceptron to obtain a spatial mapping vector corresponding to the object to be recommended. The recommendation score of the object to be recommended can then be obtained by calculating the recommendation score using this spatial mapping vector. This enhances the data processing capability of the object recommendation model, improves the accuracy and precision of the recommendation results, and enhances user experience and satisfaction.

[0082] In some embodiments, the user preference vector and the extended vector corresponding to the object to be recommended are processed based on a gating mechanism to obtain the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector, including: performing a nonlinear transformation on the user preference vector to obtain the retained vector corresponding to the user preference vector; performing a reset transformation on the retained vector corresponding to the user preference vector to obtain the gating vector corresponding to the user preference vector; performing a nonlinear transformation on the extended vector corresponding to the object to be recommended to obtain the retained vector corresponding to the extended vector; and performing a reset transformation on the retained vector corresponding to the extended vector to obtain the gating vector corresponding to the extended vector.

[0083] Specifically, the user preference vector can be processed through a nonlinear transformation, which can be implemented using the Sigmoid function or the ReLU function (the specific method is not limited here). This yields a retained vector corresponding to the user preference vector, which preserves the target information of the user preference. The retained vector can be further processed through a reset transformation, which can be implemented using a reset gate mechanism to obtain a gated vector corresponding to the user preference vector. A nonlinear transformation can be performed on the extended vector corresponding to the object to be recommended to obtain a retained vector corresponding to the extended vector. This retained vector can be used to characterize the important features of the object to be recommended. Finally, the retained vector corresponding to the extended vector can be processed through a reset gate to obtain a gated vector corresponding to the extended vector.

[0084] The retained vector corresponding to the user preference vector can be a vector obtained by processing the user preference vector through nonlinear transformation. The retained vector corresponding to the user preference vector can be used to represent the target information of user preference. This target information can be extracted by nonlinear activation functions in neural networks, including but not limited to Sigmoid or ReLU.

[0085] The retained vector corresponding to the extended vector can be a vector obtained by processing the extended vector corresponding to the object to be recommended through nonlinear transformation. The retained vector corresponding to the extended vector can be used to characterize the target feature of the object to be recommended. This target feature can be extracted through the nonlinear activation function in the neural network. This target feature can be used to evaluate the degree of matching between the object to be recommended and the user's preferences.

[0086] For example, in an e-commerce platform, a user preference vector can contain the user's encoding of information such as product category, price range, and brand preference, while the extended vector of the recommended object can contain the encoding of information such as detailed product descriptions, user reviews, and sales volume. Through nonlinear transformation, the retained vectors corresponding to the user preference vector and the product features can be extracted separately. A reset transformation is then performed using a reset gate to generate the gated vectors corresponding to the user preference vector and the gated vectors corresponding to the product features.

[0087] According to the technical solution provided in this disclosure, the user preference vector is processed by nonlinear transformation to obtain the retained vector corresponding to the user preference vector. The retained vector retains the target information of the user preference. The retained vector can be further processed by reset transformation to obtain the gate vector corresponding to the user preference vector. The extended vector corresponding to the object to be recommended can be nonlinearly transformed to obtain the retained vector corresponding to the extended vector. Then, the retained vector corresponding to the extended vector can be processed by reset gate to obtain the gate vector corresponding to the extended vector. This enhances the ability of the object recommendation model to express user preferences and the features of the object to be recommended, and improves the accuracy and flexibility of the object recommendation model.

[0088] In some embodiments, spatial alignment is performed on the gate vector corresponding to the user preference vector and the gate vector corresponding to the extension vector to obtain the target user preference vector and the target extension vector corresponding to the object to be recommended. This includes: performing dimensional alignment on the gate vector corresponding to the user preference vector and the gate vector corresponding to the extension vector to obtain the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extension vector; and performing spatial transformation on the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extension vector to obtain the target user preference vector and the target extension vector corresponding to the object to be recommended.

[0089] Specifically, the gating vectors corresponding to the user preference vector and the extended vector can be dimension-aligned to ensure they have the same dimension, resulting in vectors of the same dimension for both the user preference vector and the extended vector. These vectors can then undergo spatial transformations, such as matrix multiplication or nonlinear transformations, which are not limited here. The vectors are mapped to a specified feature space to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended.

[0090] The same-dimensional vector corresponding to the user preference vector can be a vector obtained through dimension alignment. The same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector have the same dimension. The same-dimensional vector corresponding to the user preference vector can be obtained by adjusting the dimension of the gate vector corresponding to the user preference vector. The adjustment process can be achieved by padding, truncation, or mapping, etc., which is not limited here.

[0091] The vector of the same dimension corresponding to the extended vector can be a vector obtained through dimension alignment. This vector has the same dimension as the vector of the same dimension corresponding to the user preference vector. Alternatively, the vector of the same dimension corresponding to the extended vector can be obtained by adjusting the dimension of the gate vector corresponding to the extended vector.

[0092] For example, in online shopping platforms, a user preference vector can represent a user's preference features for products, such as price, brand, and / or color, while an extended vector can represent detailed attributes of the product to be recommended, such as size, material, and / or reviews. By performing dimensional alignment and spatial transformation on the gating vectors corresponding to the user preference vector and the extended vector, a more accurate target user preference vector and target extended vector corresponding to the recommended object can be obtained.

[0093] According to the technical solution provided in this disclosure, by performing dimension alignment processing on the gate vector corresponding to the user preference vector and the gate vector corresponding to the extended vector, the dimensions of the gate vector corresponding to the user preference vector and the gate vector corresponding to the extended vector are made consistent, resulting in the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector. Spatial transformation processing can be performed on the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector respectively, mapping the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector to the specified feature space, so as to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended. This enhances the comparability between the user preference vector and the extended vector corresponding to the object to be recommended, improves the accuracy of the object recommendation model, and enhances the user experience.

[0094] In some embodiments, spatial transformation processing is performed on the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended. This includes: performing spatial rotation processing on the same-dimensional vector corresponding to the user preference vector to obtain the spatial matching vector corresponding to the user preference vector; performing weight matching processing on the same-dimensional vector corresponding to the extended vector to obtain the weighted vector corresponding to the extended vector; and aligning the spatial matching vector corresponding to the user preference vector and the weighted vector corresponding to the extended vector to obtain the target extended vector corresponding to the target user preference vector and the object to be recommended.

[0095] Specifically, spatial rotation can be performed on the same-dimensional vector corresponding to the user preference vector to obtain a spatial matching vector that is more compatible with the multilayer perceptron. Weight matching can be performed on the same-dimensional vector corresponding to the extended vector, that is, different weights are assigned according to the importance of each dimension to obtain the weighted vector corresponding to the extended vector. Then, the spatial matching vector corresponding to the user preference vector and the weighted vector corresponding to the extended vector can be aligned to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended.

[0096] The spatial matching vector corresponding to the user preference vector can be a vector obtained by spatially rotating the same-dimensional vector corresponding to the user preference vector.

[0097] The weighted vector corresponding to the extended vector can be a vector obtained by weight matching of the vectors of the same dimension corresponding to the extended vector. The weights of each dimension can be used to characterize the importance of that dimension in the recommendation.

[0098] For example, in online shopping platforms, a user preference vector can represent a user's preferences for dimensions such as price, brand, color, and / or material of a product, while an extended vector can represent additional information such as sales volume, reviews, and / or new product identification of the product to be recommended. Through spatial rotation, the same-dimensional vector corresponding to the user preference vector can be converted into a spatial matching vector. Furthermore, through weighted matching, the same-dimensional vector corresponding to the extended vector can be weighted. This allows for the alignment of the spatial matching vector corresponding to the user preference vector and the weighted vector corresponding to the extended vector, resulting in the target user preference vector and the target extended vector corresponding to the product to be recommended.

[0099] According to the technical solution provided in this disclosure, by performing spatial rotation processing on the same-dimensional vector corresponding to the user preference vector, a spatial matching vector that is more compatible with the multilayer perceptron can be obtained. Weight matching processing can be performed on the same-dimensional vector corresponding to the extended vector to obtain the weighted vector corresponding to the extended vector. Then, the spatial matching vector corresponding to the user preference vector and the weighted vector corresponding to the extended vector can be aligned to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended. This enhances the correlation between user preferences and the object to be recommended, improves the accuracy of the object recommendation model, and enhances the user experience.

[0100] All of the above-mentioned optional technical solutions can be combined in any way to form optional embodiments of this disclosure, and will not be described in detail here.

[0101] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0102] Figure 3 is a schematic diagram of an object recommendation device provided in an embodiment of this disclosure. As shown in Figure 3, the object recommendation device includes:

[0103] The first processing module 301 is used to process the user-side data and the attribute data corresponding to the object to be recommended through a large language model to obtain user preference data and extended data corresponding to the object to be recommended.

[0104] The second processing module 302 is used to perform text encoding processing on the user preference data and the extended data corresponding to the object to be recommended, respectively, to obtain the user preference vector and the extended vector corresponding to the object to be recommended.

[0105] The third processing module 303 is used to process the user preference vector and the extended vector corresponding to the object to be recommended based on the gating mechanism, so as to obtain the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector.

[0106] The fourth processing module 304 is used to perform spatial alignment processing on the gate vector corresponding to the user preference vector and the gate vector corresponding to the expansion vector, respectively, to obtain the target user preference vector and the target expansion vector corresponding to the object to be recommended;

[0107] The determination module 305 is used to determine the recommendation score of the object to be recommended based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, target user preference vector, and target extended vector corresponding to the object to be recommended.

[0108] The fifth processing module 306 is used to determine the target object data based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and send the target object data to the target terminal device so that the target object data is displayed in the current graphical user interface of the target terminal device.

[0109] According to the technical solution provided in this disclosure, text generation processing is performed on user-side data and attribute data corresponding to the object to be recommended using a large language model. User-side data containing user preference information can be obtained by processing the user-side data. Extended data containing relevant information about the object to be recommended can be obtained by processing the attribute data. Then, text encoding processing is performed on the user preference data and the extended data to be recommended to obtain a user preference vector and an extended vector corresponding to the object to be recommended. Gating processing is then performed on the user preference vector and the extended vector to obtain a gated vector corresponding to the user preference vector and a gated vector corresponding to the extended vector. Spatial alignment processing is then performed on the two gated vectors to align the gated vector corresponding to the user preference vector and the extended vector. The gating vector is mapped to the object recommendation space for data alignment, resulting in the target user preference vector and the target extended vector corresponding to the object to be recommended. Based on user-side data, attribute data corresponding to the object to be recommended, historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended, the recommendation score corresponding to the object to be recommended is determined. Then, based on the preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, the target object data can be determined and sent to the current graphical user interface of the target terminal device for display. This improves the matching accuracy between user preference information and attribute information of the object to be recommended, enhances the flexibility and adaptability of data processing, and improves the correlation and accuracy of vector representation through gating mechanism and space alignment processing, thereby enhancing the personalization and accuracy of recommendation results.

[0110] In some embodiments, the first processing module 301 is specifically used to: process user-side data based on a preset preference generation template to obtain user-side templated data; process attribute data corresponding to the object to be recommended based on a preset extended generation template to obtain templated attribute data corresponding to the object to be recommended; and process the user-side templated data and the templated attribute data corresponding to the object to be recommended respectively through a large language model to obtain user preference data and extended data corresponding to the object to be recommended.

[0111] In some embodiments, the determining module 305 is specifically used to: concatenate the target user preference vector and the target extended vector corresponding to the object to be recommended to obtain a recommendation enhancement vector; perform feature extraction processing on the user-side data, the attribute data corresponding to the object to be recommended, and the historical interaction context data corresponding to the object to be recommended, respectively, to obtain the user-side vector, the attribute vector corresponding to the object to be recommended, and the historical interaction context vector corresponding to the object to be recommended; perform vector merging processing on the recommendation enhancement vector, the user-side vector, the attribute vector corresponding to the object to be recommended, and the historical interaction context vector corresponding to the object to be recommended to obtain the target vector corresponding to the object to be recommended; and determine the recommendation score corresponding to the object to be recommended based on the target vector corresponding to the object to be recommended.

[0112] In some embodiments, determining the recommendation score of the object to be recommended based on the target vector of the object to be recommended specifically involves: flattening the target vector of the object to be recommended to obtain a low-dimensional target vector of the object to be recommended; mapping the low-dimensional target vector of the object to be recommended using a multilayer perceptron to obtain a spatial mapping vector of the object to be recommended; and calculating the recommendation score of the spatial mapping vector of the object to be recommended to obtain the recommendation score of the object to be recommended.

[0113] In some embodiments, the third processing module 303 is specifically used to: perform nonlinear transformation processing on the user preference vector to obtain the retained vector corresponding to the user preference vector; perform reset transformation processing on the retained vector corresponding to the user preference vector to obtain the gate vector corresponding to the user preference vector; perform nonlinear transformation processing on the extended vector corresponding to the object to be recommended to obtain the retained vector corresponding to the extended vector; and perform reset transformation processing on the retained vector corresponding to the extended vector to obtain the gate vector corresponding to the extended vector.

[0114] In some embodiments, the fourth processing module 304 is specifically used to perform dimension alignment processing on the gate vector corresponding to the user preference vector and the gate vector corresponding to the extension vector respectively to obtain the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extension vector; and to perform spatial transformation processing on the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extension vector respectively to obtain the target user preference vector and the target extension vector corresponding to the object to be recommended.

[0115] In some embodiments, spatial transformation processing is performed on the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended. Specifically, spatial rotation processing is performed on the same-dimensional vector corresponding to the user preference vector to obtain the spatial matching vector corresponding to the user preference vector; weight matching processing is performed on the same-dimensional vector corresponding to the extended vector to obtain the weighted vector corresponding to the extended vector; and alignment processing is performed on the spatial matching vector corresponding to the user preference vector and the weighted vector corresponding to the extended vector to obtain the target extended vector corresponding to the target user preference vector and the object to be recommended.

[0116] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure.

[0117] Figure 4 is a schematic diagram of an electronic device 4 provided in an embodiment of this disclosure. As shown in Figure 4, the electronic device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the various method embodiments described above. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the various device embodiments described above.

[0118] Electronic device 4 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 4 may include, but is not limited to, processor 401 and memory 402. Those skilled in the art will understand that FIG4 is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4, and may include more or fewer components than shown, or different components.

[0119] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0120] The memory 402 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM of the electronic device 4. The memory 402 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 4. The memory 402 can also include both internal and external storage units of the electronic device 4. The memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0121] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0122] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0123] The above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.

Claims

An object recommendation method, wherein, include: The user-side data and the attribute data corresponding to the object to be recommended are processed by a large language model to obtain user preference data and extended data corresponding to the object to be recommended. The user preference data and the extended data corresponding to the object to be recommended are respectively processed by text encoding to obtain the user preference vector and the extended vector corresponding to the object to be recommended. Based on the gating mechanism, the user preference vector and the extended vector corresponding to the object to be recommended are processed respectively to obtain the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector; Spatial alignment is performed on the gate vector corresponding to the user preference vector and the gate vector corresponding to the expansion vector to obtain the target user preference vector and the target expansion vector corresponding to the object to be recommended. The recommendation score for the object to be recommended is determined based on the user-side data, the attribute data corresponding to the object to be recommended, the historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended. Based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, target object data is determined and sent to the target terminal device so that the target object data is displayed on the current graphical user interface of the target terminal device. According to the object recommendation method of claim 1, wherein, The process involves using a large language model to process both the user-side data and the attribute data corresponding to the object to be recommended, resulting in user preference data and extended data corresponding to the object to be recommended, including: The user-side data is processed based on a preset preference generation template to obtain user-side templated data; Based on a preset extended generation template, the attribute data corresponding to the object to be recommended is processed to obtain the templated attribute data corresponding to the object to be recommended. The user-side templated data and the templated attribute data corresponding to the object to be recommended are processed by the large language model to obtain the user preference data and the extended data corresponding to the object to be recommended. According to the object recommendation method of claim 1, wherein, The step of determining the recommendation score corresponding to the object to be recommended based on the user-side data, the attribute data corresponding to the object to be recommended, the historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extended vector corresponding to the object to be recommended includes: The target user preference vector and the target extended vector corresponding to the object to be recommended are concatenated to obtain the recommendation enhancement vector. Feature extraction processing is performed on the user-side data, the attribute data corresponding to the object to be recommended, and the historical interaction context data corresponding to the object to be recommended, respectively, to obtain the user-side vector, the attribute vector corresponding to the object to be recommended, and the historical interaction context vector corresponding to the object to be recommended; The recommendation enhancement vector, user-side vector, attribute vector corresponding to the object to be recommended, and historical interaction context vector corresponding to the object to be recommended are merged to obtain the target vector corresponding to the object to be recommended. The recommendation score for the object to be recommended is determined based on the target vector corresponding to the object to be recommended. According to claim 3, the object recommendation method, wherein, The step of determining the recommendation score corresponding to the object to be recommended based on the target vector corresponding to the object to be recommended includes: Flatten the target vector corresponding to the object to be recommended to obtain the target low-dimensional vector corresponding to the object to be recommended. The target low-dimensional vector corresponding to the object to be recommended is mapped by a multilayer perceptron to obtain the spatial mapping vector corresponding to the object to be recommended. The spatial mapping vector corresponding to the object to be recommended is processed to calculate the recommendation score, thereby obtaining the recommendation score corresponding to the object to be recommended. According to the object recommendation method of claim 1, wherein, The process of processing the user preference vector and the extended vector corresponding to the object to be recommended based on the gating mechanism to obtain the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector includes: The user preference vector is subjected to a nonlinear transformation to obtain the retained vector corresponding to the user preference vector. The reserved vector corresponding to the user preference vector is reset and transformed to obtain the gate vector corresponding to the user preference vector. A non-linear transformation is performed on the extended vector corresponding to the object to be recommended to obtain the retained vector corresponding to the extended vector. The retained vector corresponding to the extended vector is reset and transformed to obtain the gated vector corresponding to the extended vector. According to the object recommendation method of claim 1, wherein, The step of spatially aligning the gate vector corresponding to the user preference vector and the gate vector corresponding to the expansion vector to obtain the target user preference vector and the target expansion vector corresponding to the object to be recommended includes: The gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector are respectively subjected to dimension alignment processing to obtain the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector; Spatial transformation is performed on the same-dimensional vectors corresponding to the user preference vector and the same-dimensional vectors corresponding to the extended vector to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended. According to the object recommendation method of claim 6, wherein, The step of performing spatial transformation processing on the same-dimensional vector corresponding to the user preference vector and the same-dimensional vector corresponding to the extended vector to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended includes: Perform spatial rotation processing on the same-dimensional vector corresponding to the user preference vector to obtain the spatial matching vector corresponding to the user preference vector; Perform weight matching processing on the same-dimensional vectors corresponding to the extended vector to obtain the weighted vectors corresponding to the extended vectors; Align the spatial matching vector corresponding to the user preference vector and the weighted vector corresponding to the extended vector to obtain the target user preference vector and the target extended vector corresponding to the object to be recommended. An object recommendation device, wherein, include: The first processing module is configured to process the user-side data and the attribute data corresponding to the object to be recommended respectively through a large language model to obtain user preference data and extended data corresponding to the object to be recommended; The second processing module is configured to perform text encoding processing on the user preference data and the extended data corresponding to the object to be recommended, respectively, to obtain the user preference vector and the extended vector corresponding to the object to be recommended. The third processing module is configured to process the user preference vector and the extended vector corresponding to the object to be recommended based on a gating mechanism, so as to obtain the gating vector corresponding to the user preference vector and the gating vector corresponding to the extended vector. The fourth processing module is configured to perform spatial alignment processing on the gate vector corresponding to the user preference vector and the gate vector corresponding to the expansion vector, respectively, to obtain the target user preference vector and the target expansion vector corresponding to the object to be recommended; The determination module is configured to determine the recommendation score corresponding to the object to be recommended based on the user-side data, the attribute data corresponding to the object to be recommended, the historical interaction context data corresponding to the object to be recommended, the target user preference vector, and the target extension vector corresponding to the object to be recommended. The fifth processing module is configured to determine target object data based on a preset recommendation score threshold and the recommendation score corresponding to the object to be recommended, and send the target object data to the target terminal device so that the target object data is displayed on the current graphical user interface of the target terminal device. An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images