Recommendation method and computing device

Abstract

Provided in the embodiments of the present application are a recommendation method and a computing device. The method comprises: determining a recommendation object that matches at least one first user feature of a user; on the basis of object description information of the recommendation object, generating first recommendation information for the recommendation object; on the basis of the at least one first user feature, generating second recommendation information; generating target recommendation information for the recommendation object in combination with the first recommendation information and the second recommendation information; and sending the target recommendation information to a user end. By means of the technical solution provided in the embodiments of the present application, the accuracy of recommendation information is increased, the recommendation effect is improved, and an improvement in the system utilization rate is facilitated.

Description

Recommended methods and computing equipment

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on August 9, 2023, with application number 202311002963.0 and application name “Recommendation Method and Computing Device”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of the present application relate to the field of electronic technology, and in particular to a recommendation method and a computing device. Background Art

[0003] With the development of Internet technology and electronic technology, there are more and more online systems that provide online activities such as product exchange. People can use resources through online systems to exchange goods, etc.

[0004] In order to improve user experience and promote product exchange, recommendations have become an important part of these online systems. Currently, some pre-set copywriting is usually used to recommend products or other things to users. The recommendation information is not accurate enough, which will reduce users' trust in the recommended objects, resulting in a low recommendation conversion rate, which will also affect system utilization.

[0005] Summary of the Invention

[0006] The embodiments of the present application provide a recommendation method and a computing device to solve the technical problem of inaccurate recommendation information in the prior art.

[0007] A first aspect of the embodiments of the present application provides a recommendation method, comprising:

[0008] Determining a recommendation object that matches at least one first user characteristic of the user;

[0009] generating first recommendation information of the recommended object according to the object description information of the recommended object;

[0010] generating second recommendation information according to the at least one first user feature;

[0011] fusing the first recommendation information and the second recommendation information to generate target recommendation information;

[0012] The target recommendation information is sent to the user terminal.

[0013] Optionally, generating the second recommendation information according to the at least one preference feature includes:

[0014] generating second recommendation information using a first generation model according to the at least one first user feature;

[0015] The first generation model is a large language model; the first generation model is trained using at least one first user feature corresponding to a sample object and second recommendation information corresponding to the sample object.

[0016] Optionally, the method further includes:

[0017] Determining the at least one first user characteristic by combining the historical behavior data and / or the real-time behavior data of the user;

[0018] or,

[0019] Providing a conversation interaction interface to the user terminal; determining at least one conversation text matching the user, and sending the at least one conversation text to the user terminal so that the user terminal outputs the at least one conversation text on the conversation interaction interface; obtaining at least one user input text provided by the user for the at least one conversation text; and determining the at least one first user feature based on the at least one user input text.

[0020] Optionally, the method further includes:

[0021] Acquire at least one first attribute information of the recommended object;

[0022] obtaining at least one second user characteristic of the user;

[0023] Based on the at least one first attribute information and the at least one second user feature, a text material corresponding to the recommended object is generated using a text generation model; the object description information includes the text material;

[0024] The text generation model is a large language model; the text generation model is obtained by training using at least one first attribute information and at least one user feature corresponding to a sample object, as well as a sample text corresponding to the sample object.

[0025] Optionally, the method further includes:

[0026] Acquiring at least one second attribute information of the recommended object;

[0027] obtaining at least one third user characteristic of the user;

[0028] Based on the at least one second attribute information and the at least one third user feature, using a first image generation model to generate an image material corresponding to the recommended object, wherein the object description information includes the image material;

[0029] The first image generation model is obtained by training using at least one second attribute information and at least one third user feature corresponding to a sample object, as well as a sample image corresponding to the sample object.

[0030] Optionally, generating the first recommendation information of the recommended object according to the object description information of the recommended object includes:

[0031] Extracting key information from the object description information of the recommended object as first material information;

[0032] generating an object image of the recommended object using a second image generation model based on the object description information and at least one third user feature of the user, wherein the object image serves as second material information;

[0033] generating first recommendation information of the recommended object based on the first material information and the second material information;

[0034] The second image generation model is obtained by training using object description information corresponding to a sample object and at least one third user feature, as well as a sample image corresponding to the sample object.

[0035] Optionally, the recommendation object is recommended content;

[0036] Determining the recommended object that matches at least one first user feature of the user includes:

[0037] determining a target topic that matches at least one first user characteristic of the user;

[0038] The content material that matches the target theme is used as the recommended content corresponding to the user.

[0039] Optionally, the method further includes:

[0040] For any topic, determining at least one product associated with the topic;

[0041] According to the product-related information corresponding to the at least one product, corresponding content materials are generated using a content generation model.

[0042] A second aspect of the embodiments of the present application provides a recommendation method, including:

[0043] Obtain target recommendation information; the target recommendation information includes first recommendation information and second recommendation information; the first recommendation information is generated based on object description information of the recommended object; the second recommendation information is generated based on at least one first user feature; the recommended object matches the at least one first user feature;

[0044] Displaying the first recommendation information on the recommendation page in a first display style;

[0045] The second recommendation information is displayed on the recommendation page in a second display style.

[0046] Optionally, the first display style or the second display style includes one or more attributes of display position, display size, display color, font style, display background, and display border;

[0047] At least one attribute of the first display style is different from that of the second display style.

[0048] A third aspect of the embodiments of the present application provides a recommendation method, comprising:

[0049] Provide a conversational interaction interface to the user end;

[0050] Determining at least one conversation text matching the user, and sending the at least one conversation text to the user terminal, so that the user terminal outputs the at least one conversation text on the conversation interaction interface;

[0051] Obtaining at least one user input text provided by a user for the at least one conversation text;

[0052] Determining a recommended object that matches the at least one user input text;

[0053] generating target recommendation information of the recommended object;

[0054] The target recommendation information is sent to the user terminal.

[0055] Optionally, determining at least one conversation text matching the user and sending the at least one conversation text to the user terminal includes:

[0056] Obtaining text input by a first user in a current round of conversation; determining at least one text input by a second user from historical conversation information of the current round of conversation; combining the first user input text and the at least one second user input text, and using a first conversation recommendation model to determine a conversation text in the current round of conversation that matches the first user input text; and sending the conversation text to the user terminal;

[0057] Optionally, determining at least one conversation text matching the user and sending the at least one conversation text to the user terminal includes:

[0058] Determine at least one preference feature of the user; and based on the at least one preference feature, determine at least one conversation text using a second conversation recommendation model; and send the at least one conversation text to the user terminal using a multi-round conversation method.

[0059] Optionally, generating the target recommendation information of the recommended object includes:

[0060] generating third recommendation information of the recommended object according to the object description information of the recommended object;

[0061] generating fourth recommendation information using a second generation model according to the at least one user input text;

[0062] fusing the third recommendation information and the fourth recommendation information to generate target recommendation information of the recommendation object;

[0063] The target recommendation information is sent to the user terminal.

[0064] A fourth aspect of the embodiments of the present application provides a recommendation method, including:

[0065] Provide a conversational interaction interface;

[0066] Displaying at least one conversation text matching the user on the conversation interaction interface;

[0067] Obtaining at least one user input text provided by the user for the at least one conversation text;

[0068] Sending the at least one user input text to the server; the at least one user input text is used to determine matching recommendation objects;

[0069] Obtain target recommendation information sent by the server;

[0070] Output the target recommendation information.

[0071] According to a fifth aspect of the embodiments of the present application, a computing device is provided, including a processing component and a storage component;

[0072] The storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the recommendation method described in the first aspect above, or the recommendation method described in the second aspect above, or the recommendation method described in the third aspect above, or the recommendation method described in the fourth aspect above.

[0073] In a sixth aspect of an embodiment of the present application, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a computer, it implements the recommendation method described in the first aspect above, the recommendation method described in the second aspect above, the recommendation method described in the third aspect above, or the recommendation method described in the fourth aspect above.

[0074] In the embodiment of the present application, in combination with at least one first user feature, a recommended object matching the user is determined, the accuracy of the recommended object is ensured, and recommendation information is generated in combination with the recommendation basis, so that the recommendation reason is added to the recommendation information, and the interpretability of the recommendation information is improved, which helps to improve the user's trust in the recommendation results. Therefore, the technical solution of the embodiment of the present application can improve the accuracy of the recommendation information, improve the recommendation effect, and improve the recommendation conversion rate, so that users are more inclined to use the online system to purchase goods, thereby improving the system utilization rate.

[0075] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0077] FIG1 shows a system architecture diagram in which the technical solution of an embodiment of the present application is applied;

[0078] FIG2 shows a flow chart of an embodiment of a recommendation method provided by the present application;

[0079] FIG3 shows a flow chart of another embodiment of a recommendation method provided by the present application;

[0080] FIG4 shows a flow chart of another embodiment of a recommendation method provided by the present application;

[0081] FIG5 shows a flow chart of another embodiment of a recommendation method provided by the present application;

[0082] FIG6 is a schematic diagram showing a scene interaction of the technical solution of an embodiment of the present application in an actual application;

[0083] FIG7 is a schematic diagram showing an interface display of the technical solution of an embodiment of the present application in an actual application;

[0084] FIG8 shows a schematic structural diagram of an embodiment of a recommendation device provided by the present application;

[0085] FIG9 shows a schematic structural diagram of another embodiment of a recommendation device provided by the present application;

[0086] FIG10 shows a schematic structural diagram of another embodiment of a recommendation device provided by the present application;

[0087] FIG11 shows a schematic structural diagram of another embodiment of a recommendation device provided by the present application;

[0088] FIG12 shows a schematic structural diagram of an embodiment of a computing device provided by the present application. DETAILED DESCRIPTION

[0089] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0090] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0091] The technical solutions of the embodiments of the present application can be applied to online transaction scenarios, and of course, are also applicable to other recommendation scenarios involving objects.

[0092] The objects described in this article may refer to the products provided by the online system. Of course, with the development of the information age, content marketing has become the mainstream method of major online trading platforms. Users spend a lot of time on content information, thereby also irrigating a large amount of traffic into the content, and the impact of content on users gradually increases. The objects described in this article may also refer to the content in the online system, which may involve one or more products, etc. It can be an information carrier such as images, text and / or videos to achieve the purpose of guiding users to shop.

[0093] In order to promote commodity exchange and improve user experience, recommending suitable objects to users in order to increase users' willingness to purchase commodities based on the recommended objects has become a major way to improve system efficiency. In the process of realizing this application, the inventors found that whether the recommended objects meet user needs and whether the recommendation information used for the recommended objects is accurate will affect the recommendation effect. The current recommendation information is often inaccurate, which will reduce users' trust in the recommended objects and online systems. Users may choose not to use the online system or not purchase the commodities involved in the recommended objects. As a result, the recommendation conversion rate is low, which also affects the system utilization rate.

[0094] In order to improve the recommendation effect, increase the recommendation conversion rate, and improve the system utilization rate, the inventor has proposed the technical solution of the present application after a series of studies. The embodiment of the present application determines the recommendation objects that meet the user's needs or user preferences from the user's perspective, ensures the accuracy of the recommendation objects, and generates recommendation information based on the recommendation basis, so that the recommendation reasons are added to the recommendation information, and the interpretability of the recommendation results is improved, which helps to enhance the user's trust in the recommendation results. Therefore, the technical solution of the embodiment of the present application can improve the recommendation effect, increase the recommendation conversion rate, and users are more inclined to use the online system to purchase goods, thereby improving the system utilization rate.

[0095] The following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of this application, not all of them. All other embodiments derived by those skilled in the art based on the embodiments of this application without inventive effort shall fall within the scope of protection of this application.

[0096] FIG1 shows a system architecture diagram in which the technical solution of an embodiment of the present application can be applied. The system architecture may include a user terminal 101 and a server terminal 102 .

[0097] The client 101 and the server 102 are connected via a network. The network provides a medium for the communication link between the client 101 and the server 102. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. Alternatively, the client can establish a communication connection with the server via Bluetooth, WiFi (Wireless Fidelity), infrared, or other methods.

[0098] The client 101 can interact with the server 102 via the network to receive or send messages.

[0099] The user terminal 101 can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5) application, a light application (also known as a mini-program, a lightweight application), or a cloud application. The user terminal 101 can be deployed in an electronic device and needs to rely on the device to run or certain apps in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, tablet computer, personal computer, desktop computer, smart speaker, smart watch, etc. For ease of understanding, Figure 1 mainly uses the image of the device to represent the user terminal. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc. The electronic device can refer to a device used by the user and has the computing, Internet access, communication, etc. functions required by the user, such as a mobile phone, tablet computer, personal computer, wearable device, etc. The electronic device can generally include at least one processing component and at least one storage component. The electronic device may also include basic configurations such as a network card chip, an IO (input / output) bus, and audio and video components, which are not limited in this application. Optionally, depending on the implementation form of the electronic device, it may also include some peripheral devices, such as a keyboard, a mouse, an input pen, a printer, etc., which are not limited in this application.

[0100] The server 102 may include servers that provide various services, such as a server for background training that provides support for the model used on the user terminal 101, or a server that processes interactive information sent by the user terminal.

[0101] It should be noted that the server 102 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0102] It should be understood that the number of user terminals and server terminals in FIG1 is merely illustrative and any number of user terminals and server terminals may be provided according to implementation requirements.

[0103] It should be noted that the embodiments of the present application may involve the use of user data. In actual applications, user-specific personal data can be used in the scheme described herein within the scope permitted by applicable laws and regulations, subject to the requirements of applicable laws and regulations of the country where the user is located (for example, with the user's explicit consent, effective notification to the user, etc.).

[0104] It should be noted that the technical solution of the embodiment of the present application is applicable to a network virtual environment. The users described generally refer to "virtual users". Real users can register user accounts on the server through registration to obtain user identities in the network environment.

[0105] The implementation details of the technical solution of the embodiment of the present application are described in detail below.

[0106] FIG2 is a flowchart of an embodiment of a recommendation method provided by an embodiment of the present application. The technical solution of this embodiment can be executed by the server. The method may include the following steps:

[0107] 201: Determine a recommended object that matches at least one first user feature of a user.

[0108] The recommended object is determined based on at least one first user characteristic of the user and meets the user's needs, thereby ensuring the accuracy of the recommended object.

[0109] Among them, the at least one first user feature may include, for example, user attribute features and / or user preference features, etc. User attribute features may include, for example, population labels, age, occupation, etc. User preference features may include preferred object features. For example, when the object is a commodity, the preferred object features may include, for example, commodity brand, commodity category, price range, color, etc.

[0110] The at least one user feature can be identified in combination with the user's historical behavior data and / or real-time behavior data. For example, it can be obtained by performing data statistics based on historical browsing records, historical purchase records, real-time browsing records, etc. Of course, it can also be implemented in other ways, which will be described in detail below.

[0111] 202: Generate first recommendation information of the recommended object according to the object description information of the recommended object.

[0112] In online systems that provide product exchange, in order to facilitate users to understand the products and then exchange them, the system often provides a corresponding product details page for each product to introduce relevant information about the product and provide some product processing controls for users to trigger corresponding operations.

[0113] In one possible implementation, when the recommended object is a product, the object description information may include the product details on the product details page. When the recommended object is content, the object description information also refers to the information carried by the content or content features, such as the content title and content type.

[0114] Furthermore, since product details are provided by the product provider, they may not meet user needs and may be less appealing to users. Therefore, in another possible implementation, the method may further include: obtaining at least one first attribute information of the recommended object; obtaining at least one second user characteristic of the user; and generating text material corresponding to the recommended object using a text generation model based on the at least one first attribute information and the at least one second user characteristic; the object description information may include the text material generated by the text generation model.

[0115] The text generation model can be implemented using a large language model (LLM). This text generation model can be trained using at least one first attribute information and at least one user feature corresponding to a sample object, as well as sample text corresponding to the sample object. The at least one first attribute information and at least one user feature corresponding to the sample object can serve as input data, and the sample text corresponding to the sample object can serve as training labels, thereby enabling training of the large language model.

[0116] A large language model refers to a deep learning model trained with a large amount of data, which can generate natural language text or understand the meaning of language text. It can be a pre-trained language model. The pre-trained language model is used in combination with the above-mentioned training data, that is, at least one first attribute information and at least one user feature corresponding to the sample object, and the sample text corresponding to the sample object, to perform fine-tuning training to obtain the large language model of the embodiment of the present application. The large language model can be implemented, for example, using ChatGPT (Chat Generative Pre-trained Transformer). Natural language text can be generated using the large language model, thereby generating high-quality text materials, and a more vivid expression of the recommended object can be displayed in natural language.

[0117] The large language model can be a multimodal model. The at least one first attribute information can include, for example, one or more of a product identifier, product image, product category, product price, product model, product title, product color, and product details. The product image can be provided by the product provider in the product details. The product identifier can be, for example, the product name, and the product category can refer to the product's category, such as clothing, electronics, or home furnishings.

[0118] Of course, the text generation model can also be implemented using other AIGC (AI generated content) models, and this application does not limit this.

[0119] The at least one second user feature mentioned above can be obtained by performing big data statistics on all users in the online system to obtain a user portrait. Of course, it can also be a user preference feature obtained by performing big data statistics on the user's historical behavior data and / or real-time behavior data, so that the text material can conform to the user's preferences, etc.

[0120] In a practical application, text materials can include a variety of information types. For example, they can include basic attribute information such as a product's appearance, size, color, material, and function. They can also include information describing the product's applicable scenarios, usage instructions, and precautions. Specifically, training and setting can be based on actual conditions, and text materials can enrich object descriptions. Furthermore, text materials can be set in different languages ​​based on actual conditions. When training a text generation model, training data in a specific language can be used. For example, if the sample text corresponding to a sample object is in a specific language, text materials in that specific language can be generated.

[0121] In another possible implementation, the object description information may include product details information and the above-mentioned text material, etc.

[0122] In addition, in another possible implementation, the method may also include: obtaining at least one second attribute information of the recommended object; obtaining at least one third user feature of the user; and generating image material corresponding to the recommended object using the first image generation model based on the at least one second attribute information and the at least one third user feature.

[0123] The object description information may include the image material.

[0124] The at least one third user feature may include, for example, a user preference feature, such as a user preferred image style.

[0125] The at least one second attribute information may include one or more of the following: product identification, product category, original image, product price, product model, product title, product color, and product details. The original image may refer to the product image set by the product provider on the product details page. In addition, since product review data is an important source of information for users, the at least one second attribute information may also include core information extracted from the product review data. Optionally, the at least one second attribute information may include one or more of the following: product identification, product category, original image, product price, product model, product title, product color, product details, and core information extracted from the product review data.

[0126] The first image generation model can be trained using at least one second attribute information and at least one second user feature corresponding to a sample object, as well as a sample image corresponding to the sample object. The at least one second attribute information and at least one second user feature corresponding to the sample object can serve as input data, while the sample image corresponding to the sample object can serve as a training label, thereby training the first image generation model.

[0127] The first image generation model can use a GAN (Generative Adversarial Networks) model. Using GAN technology, a variety of image styles can be generated, such as cartoon style, sketch style, oil painting style, etc.

[0128] The object comment data may include original comment data and target comment data generated based on the original comment data. The original comment data is comment data provided by users. Since the original comment data may be pure text data or pure image data, in order to enrich the object comment data, a data processing model may be used to generate image comment data based on the original comment data, or to generate corresponding text comment data based on the image comment data. The data processing model may be generated using a GAN model, for example, but the present application is not limited thereto.

[0129] Optionally, the object description information may include one or more of the above-mentioned product details information, text materials, and image materials.

[0130] Among them, key information can be extracted from the object description information as the first recommendation information, or key information can be extracted from the object description information using an information extraction model. Of course, other methods can also be used to generate the first recommendation information, which will be introduced in detail below.

[0131] 203: Generate second recommendation information based on at least one first user feature.

[0132] In an embodiment of the present application, in addition to generating first recommendation information in combination with object description information to facilitate introducing the recommended object to the user so that the user can quickly understand the recommended object, second recommendation information can also be generated based on at least one first user feature.

[0133] Since the at least one user characteristic is the basis for determining the recommended item, the second recommendation information can explain the recommendation reason to the user, making the user more receptive to the recommended item, and helps the user understand the recommendation algorithm, increase the user's trust in the recommendation algorithm, and help improve the recommendation conversion rate.

[0134] In order to improve the accuracy and recommendation effect of the second recommendation information, the second recommendation information may optionally be generated using a first generation model based on at least one first user feature.

[0135] The first generative model may be trained using at least one sample user feature and sample recommendation information corresponding to the at least one sample user feature. The at least one sample user feature may serve as input data, and the sample recommendation information may serve as training labels, thereby enabling training of the first generative model.

[0136] Optionally, the first generation model can be implemented using a large language model, such as ChatGPT; the large language model can be used to generate natural language text, thereby generating high-quality second recommendation information, and the reason for the recommendation can be displayed in natural language.

[0137] 204: The first recommendation information and the second recommendation information are integrated to generate target recommendation information of the recommendation object.

[0138] 205: Send the target recommendation information to the user end.

[0139] The user end is used to output target recommendation information, and the target recommendation information is used to recommend the recommended object.

[0140] As an optional method, the target recommendation information can be generated by fusing the first recommendation information and the second recommendation information. That is, the target recommendation information is composed of the first recommendation information and the second recommendation information.

[0141] In addition, in order to further improve the accuracy of the recommendation information and enhance the recommendation effect, as another optional method, a third image generation model can be used based on the first recommendation information and the second recommendation information to generate a first recommendation image, and the first recommendation image can be sent to the user end as the target recommendation information. The visualization effect is more intuitive in the form of an image. The third image generation model can be implemented, for example, using a GAN model, which can be pre-trained using the first sample recommendation information and the second sample recommendation information, as well as the first sample recommendation image corresponding to the first sample recommendation information and the second sample recommendation information.

[0142] The first recommendation effect is sent to the user terminal, and the user terminal can display the target recommendation information on the recommendation page. The recommendation page can be, for example, the current page of the user terminal, or of course, a specific target page. In the case of displaying the target page, the target recommendation information can be displayed on the target page.

[0143] In this embodiment, in combination with at least one first user feature, a recommended object matching the user is determined, the accuracy of the recommended object is ensured, and recommendation information is generated in combination with the recommendation basis, so that the recommendation reason is added to the recommendation information, and the interpretability of the recommendation result is improved, which helps to enhance the user's trust in the recommendation result. Therefore, the technical solution of the embodiment of the present application can improve the recommendation effect, improve the recommendation conversion rate, make users more inclined to use the online system to purchase goods, and thus improve the system utilization rate.

[0144] In some embodiments, the method may further include:

[0145] Providing a conversation interaction interface to a user terminal; determining at least one conversation text matching the user; sending the at least one conversation text to the user terminal so that the user terminal outputs the at least one conversation text on the conversation interaction interface; obtaining at least one user input text provided by the user for the at least one conversation text; and determining at least one first user feature based on the at least one user input text.

[0146] In the embodiment of the present application, in addition to being obtained through statistical analysis based on user registration information, user historical behavior data, and user real-time behavior data, at least one user feature can be determined as another optional method using a conversation guidance method based on user input text.

[0147] Among them, a feature extraction model can be used to identify at least one first user feature from at least one user input text. The feature extraction model can be implemented using a machine learning model, which is not limited in this application.

[0148] As an optional implementation, the determining of at least one conversation text matching the user and sending the at least one conversation text to the user terminal may include:

[0149] Obtaining text input by a first user in a current round of conversation; determining at least one text input by a second user from historical conversation information of the current round of conversation; combining the first user input text and the at least one second user input text, using a first conversation recommendation model to determine a conversation text in the current round of conversation that matches the current user input text; and sending the conversation text to a user terminal.

[0150] That is, a multi-round conversation method can be used to guide the user to provide user input text. The conversation text in each round of conversation can be determined in combination with historical conversation information. Optionally, if the current round of conversation is the first round of conversation, the conversation text of the first round of conversation can be pre-set.

[0151] The first conversation recommendation model can be trained using a first training text, at least one second training text associated with the first training text, and a conversation training text corresponding to the first training text. For example, the first conversation recommendation model can be trained using the first training text, at least one second training text, and the conversation training text, with the matching probability between the first training text and the conversation training text as the training result.

[0152] The first conversational recommendation model can be implemented using a large language model, such as ChatGPT. This allows for the use of natural language processing technology to process user input text and generate corresponding conversational text as response content, thereby improving the user experience. Through multiple rounds of conversation, the goal of gradually receiving, processing, and responding to user needs is achieved in a more natural conversational manner. For example, if the user enters the text "My niece is starting to learn makeup in college. What brands are suitable for this beautiful girl?", the corresponding conversational text may be "We have found the following makeup brands for you, please select." The makeup brand selected by the user is then used as the user input text, and the corresponding conversational text may be "What color do you like?"

[0153] In addition, as can be seen from the above description, the recommended objects can be products or content. In addition, as another optional method, the recommended objects can include both products and content, etc., for user screening, etc. For example, at least one user input text includes "I am going to Thailand for vacation with my husband and children next week. Please help me pick out a set of parent-child outfits. The budget is 3,000 yuan." The at least one first user feature extracted from it can include, for example, "Thailand", "vacation", "parent-child", and "3,000 yuan". Then the recommended objects can include, for example, content including "Thailand tourism, such as cultural customs, travel suggestions, itinerary planning, etc.", as well as products belonging to the "parent-child outfit" category and priced at "3,000 yuan" or products with the "Thailand" and "travel" tags.

[0154] In addition, as another optional implementation manner, determining at least one conversation text matching the user and sending the at least one conversation text to the user terminal includes:

[0155] Determine at least one preference feature of the user; determine at least one conversation text based on the at least one preference feature using a second conversation recommendation model; and send the at least one conversation text to the user terminal using a multi-round conversation method.

[0156] Specifically, at least one conversation text matching at least one user preference characteristic can be determined in advance, and then the at least one conversation text can be output sequentially by the user terminal in an output order. Each conversation text can be output after obtaining the user input text corresponding to the previous conversation text. Thus, user input text corresponding to each of the at least one conversation texts can be obtained.

[0157] The at least one preference feature can be determined in combination with historical user behavior data and / or real-time behavior data, and may include, for example, the user's preferred product category, product name, or product brand. Thus, based on the at least one preference feature, at least one corresponding conversation text can be determined. For example, if the user's preferred product category is lipstick, the second conversation recommendation model can be used to determine at least one conversation text corresponding to lipstick, such as "What brand do you like?", "What is your preferred price range?", "What is your skin color?", etc. The at least one conversation text is a question text, and the user can provide their own response text for each conversation text as needed. These response texts serve as user input text for identifying the at least one first user feature.

[0158] The second conversation recommendation model can be trained using at least one sample feature and a text list that matches at least one sample feature.

[0159] The first conversation recommendation model can be implemented using a large language model such as ChatGPT, so that natural language processing technology can be used to generate conversation text to improve user experience.

[0160] In some embodiments, generating the first recommendation information of the recommended object based on the object description information of the recommended object includes:

[0161] Extract key information from the object description information of the recommended object as the first material information; based on the object description information and at least one third user feature of the user, use the second image generation model to generate an object image of the recommended object, and use the object image as the second material information; based on the first material information and the second material information, generate the first recommendation information of the recommended object.

[0162] That is, the key information extracted from the object description information can be used as the first material information. In order to enrich the first recommendation information, improve the accuracy of the first recommendation information, and improve the recommendation effect, the object description information and at least one third user feature can be combined to use the second image generation model to generate an object image of the recommended object. The object image can be used as the second material information.

[0163] The second image generation model may be obtained by training using object description information corresponding to the sample object, at least one third user feature corresponding to the sample object, and a sample image corresponding to the sample object. Optionally, the second image may be implemented using a GAN model, for example.

[0164] The first material information and the second material information may constitute the first recommendation information. Of course, based on the first and second material information, a fourth image generation model may be used to generate a second recommendation image, which may serve as the first recommendation information to enhance the recommendation effect in an image format.

[0165] The fourth image generation model is implemented using a GAN model, which can be pre-trained using the first sample material information and the second sample material information, as well as the second sample recommended image corresponding to the first sample material information and the second sample material information.

[0166] In some embodiments, the above-mentioned recommendation object may refer to recommended content, where the content may refer to information carriers such as images, texts and / or videos that carry content information, which are used to achieve the purpose of guiding users in shopping. The content may involve one or more commodities, etc.

[0167] In the case where the recommendation object is recommended content, the above-mentioned determination of the recommendation pair that matches at least one first user feature of the user may include: determining a target theme that matches at least one first user feature of the user; and using content materials that match the target theme as the recommended content corresponding to the user.

[0168] Among them, recommended content corresponding to different topics can be pre-configured, so that for a user, the target topic that matches the user can be determined based on the at least one first user feature corresponding to the user, and the content material corresponding to the target topic can be used as recommended content.

[0169] The above different topics can be abstractly generated based on user shopping needs, such as "How to choose colored contact lenses for Valentine's Day", "What to buy for camping", or "Gifts for my wife on March 8th".

[0170] In some embodiments, the method may further include: for any topic, determining at least one product associated with the topic; and generating corresponding content material using a content generation model based on product-related information corresponding to the at least one product.

[0171] That is, the content materials corresponding to each topic can be pre-generated using the content generation model.

[0172] The content generation model can be pre-trained using sample content and product-related information of at least one sample product associated with the sample content. The content generation model can be implemented using a large language model, such as ChatGPT, so as to generate natural language, high-quality content materials.

[0173] The product-related information may include, for example, product description information; the product description information may include, for example, product details information, text materials generated using the above-mentioned text generation model, and one or more of the image materials generated using the above-mentioned first image generation model, etc. Of course, the product-related information may also refer to product attribute information such as product name, product price, product image, product model, product title, product color, and product details information.

[0174] In some embodiments, the method may further include:

[0175] Obtain user feedback information regarding the target recommendation information; and retrain the first generation model based on the feedback information using the second recommendation information corresponding to the recommended object and at least one first user feature as training data.

[0176] Among them, the feedback information can indicate whether the user accepts the target recommendation information. The target recommendation information may include a feedback control. The feedback information may be generated by the user end in response to a trigger operation on the feedback control. For example, the feedback control may include a first control representing acceptance of the target recommendation information and a second control representing rejection of the target recommendation information.

[0177] If the feedback information is to accept the target recommendation information, it indicates that the user likes the target recommendation information. At this time, the relevant information corresponding to the recommended object can be used as a positive sample to retrain the first generation model. If the feedback information is to reject the target recommendation information, it indicates that the user does not like the target recommendation information. At this time, the relevant information corresponding to the recommended object can be used as a negative sample to retrain the first generation model.

[0178] Similarly, based on this feedback information, the first and second image generation models mentioned above can also be retrained in combination with relevant information corresponding to the recommended objects. By adding a feedback system, users can provide feedback and evaluate recommendation information, helping to improve the quality and transparency of the recommendation algorithm.

[0179] In some embodiments, determining a recommended object that matches at least one first user characteristic of the user may include:

[0180] In response to the object recommendation event, a recommended object matching at least one first user characteristic of the user is determined.

[0181] In an optional implementation, the object recommendation event may be, for example, an order request for a target product.

[0182] The target recommendation information can be displayed on the order result prompt page. The order result prompt page is used to display the order success prompt information. The order success prompt information is used to prompt the user that the target product has been purchased successfully.

[0183] In another optional implementation, the object recommendation event may be, for example, a target page display event, and the target page may be, for example, a home page, a shopping cart page, or an object promotion page for promoting multiple objects.

[0184] Of course, the object recommendation event may also be a periodically generated event, for example, an object recommendation event may be generated at fixed intervals, and the fixed period may be, for example, 1 day, 1 month, etc., so that objects can be regularly recommended to users.

[0185] FIG3 is a flowchart of another embodiment of a recommendation method provided by an embodiment of the present application. The technical solution of this embodiment is executed by a user terminal. The method may include the following steps:

[0186] 301: Get target recommendation information.

[0187] Among them, the target recommendation information can be generated by fusing the first recommendation information and the second recommendation information; the first recommendation information is generated based on the object description information of the recommended object; the second recommendation information is generated based on at least one first user feature; and the recommended object matches the at least one first user feature.

[0188] The target recommendation information can be generated by the server and sent to the user. The specific method of generating the target recommendation information can be found in the above corresponding embodiments, and will not be repeated here.

[0189] 302: Output target recommendation information.

[0190] The user terminal may display the target recommendation information on a recommendation page for the user to view.

[0191] In some embodiments, the target recommendation information may be composed of the first recommendation information and the second recommendation information. To further improve the recommendation effect, the output target recommendation information may include:

[0192] Displaying the first recommendation information on the recommendation page in a first display style;

[0193] The second recommendation information is displayed on the recommendation page in a second display style.

[0194] The form of display, etc. For example, dividing into two areas or how to display.

[0195] The first display style and the second display style may be different, so that the user can distinguish the first recommendation information from the second recommendation information to respectively obtain the recommendation reason and the object-related information.

[0196] Optionally, the first display style or the second display style includes one or more attributes of display position, display size, display color, font style, background style, and display border;

[0197] When the first recommendation information and the second recommendation information include text materials, the font style may include the font size, model, and / or color of the text information.

[0198] The background style may refer to, for example, the background color or pattern of the display area corresponding to the display location of the first recommendation information or the second recommendation information. The display frame may refer to a frame structure used to define the first recommendation information or the second recommendation information, and may have attributes such as shape and color.

[0199] Optionally, the first display style and the second display style may be different in at least one corresponding attribute.

[0200] In some embodiments, the method may further include:

[0201] Providing a conversation interaction interface; displaying at least one conversation text matching the user on the conversation interaction interface; obtaining at least one user input text provided by the user for the at least one conversation text; and sending the at least one user input text to a server.

[0202] The server may determine at least one first user feature based on at least one user input text.

[0203] FIG4 is a flowchart of another embodiment of a recommendation method provided by an embodiment of the present application. The technical solution of this embodiment is executed by the server. The method may include the following steps:

[0204] 401: Provide a conversational interaction interface to the user.

[0205] 402: Determine at least one conversation text matching the user, and send the at least one conversation text to the user terminal, so that the user terminal outputs the at least one conversation text on the conversation interaction interface.

[0206] 403: Obtain at least one user input text provided by the user for at least one conversation text.

[0207] 404: Determine recommended objects that match at least one user input text.

[0208] 405: Generate target recommendation information for the recommended object.

[0209] 406: Send the target recommendation information to the user end.

[0210] The user terminal may output the target recommendation information, and optionally, may display the target recommendation information on a recommendation page for the user to view.

[0211] In this embodiment, a conversation guidance method is adopted to obtain at least one user input text. The at least one user input text can represent the user's real needs. Therefore, a more accurate recommendation object can be determined based on the at least one user input text, thereby ensuring the accuracy of the recommendation information, helping to improve the recommendation conversion rate, making users more inclined to use the online system to purchase products, and thus improving system utilization.

[0212] As an optional implementation, the determining of at least one conversation text matching the user and sending the at least one conversation text to the user terminal may include:

[0213] Obtaining text input by a first user in a current round of conversation; determining at least one text input by a second user from historical conversation information of the current round of conversation; combining the first user input text and the at least one second user input text, using a first conversation recommendation model to determine conversation text in the current round of conversation that matches the text input by the first user; and sending the conversation text that matches the text input by the first user to a user terminal.

[0214] That is, a multi-round conversation method can be used to guide the user to provide user input text. The conversation text in each round of conversation can be determined in combination with historical conversation information. Optionally, if the current round of conversation is the first round of conversation, the conversation text of the first round of conversation can be pre-set.

[0215] The first conversation recommendation model can be trained using a first training text, at least one second training text associated with the first training text, and a conversation training text corresponding to the first training text. For example, the first conversation recommendation model can be trained using the first training text, at least one second training text, and the conversation training text, with the matching probability between the first training text and the conversation training text as the training result.

[0216] The first conversational recommendation model can be implemented using a large language model, such as ChatGPT. This allows for the use of natural language processing technology to process user input text and generate corresponding conversational text as response content, thereby improving the user experience. Through multiple rounds of conversation, the goal of gradually receiving, processing, and responding to user needs is achieved in a more natural conversational manner. For example, if the user enters the text "My niece is starting to learn makeup in college. What brands are suitable for this beautiful girl?", the corresponding conversational text may be "We have found the following makeup brands for you, please select." The makeup brand selected by the user is then used as the user input text, and the corresponding conversational text may be "What color do you like?"

[0217] Furthermore, as can be seen from the above description, recommendation objects can be products or content. Furthermore, as another optional method, recommendation objects can include both products and content, etc., for user screening. For example, if at least one user input text includes "I'm going on vacation to Thailand with my husband and children next week. Please help me pick out a family outfit. My budget is 3,000 yuan." At least one first user feature extracted from the text may include, for example, "Thailand," "vacation," "family," and "3,000 yuan." Recommended objects may include, for example, content containing "Thailand travel, such as cultural customs, travel advice, and itinerary planning," as well as products in the "family outfit" category priced at "3,000 yuan," or products tagged with "Thailand" or "travel."

[0218] As another optional implementation, determining at least one conversation text matching the user and sending the at least one conversation text to the user terminal includes:

[0219] Determine at least one preference feature of the user; determine at least one conversation text based on the at least one preference feature using a second conversation recommendation model; and send the at least one conversation text to the user terminal using a multi-round conversation method.

[0220] Specifically, at least one conversation text matching at least one user preference characteristic can be determined in advance, and then the at least one conversation text can be output sequentially by the user terminal in an output order. Each conversation text can be output after obtaining the user input text corresponding to the previous conversation text. Thus, user input text corresponding to each of the at least one conversation texts can be obtained.

[0221] The at least one preference feature can be determined in combination with historical user behavior data and / or real-time behavior data, and may include, for example, the user's preferred product category, product name, or product brand. Thus, based on the at least one preference feature, at least one corresponding conversation text can be determined. For example, if the user's preferred product category is lipstick, the second conversation recommendation model can be used to determine at least one conversation text corresponding to lipstick, such as "What brand do you like?", "What is your preferred price range?", "What is your skin color?", etc. The at least one conversation text is a question text, and the user can provide their own response text for each conversation text as needed. These response texts serve as user input text for identifying the at least one first user feature.

[0222] The second conversation recommendation model can be trained using at least one sample feature and a text list that matches at least one sample feature.

[0223] The first conversation recommendation model can be implemented using a large language model such as ChatGPT, so that natural language processing technology can be used to generate conversation text to improve user experience.

[0224] In some embodiments, the target recommendation information for generating the recommended object may include:

[0225] Generate third recommendation information of the recommended object based on the object description information of the recommended object; generate fourth recommendation information using the second generation model based on at least one user input text; and fuse the third recommendation information and the fourth recommendation information to generate target recommendation information of the recommended object.

[0226] The third recommendation information can be the same as the second recommendation information mentioned in the above embodiment. The specific generation method can be described in detail in the above corresponding embodiment, and will not be repeated here. Thus, combined with the third recommendation information, the user can quickly understand the recommended object.

[0227] Furthermore, the second generation model can be used to generate fourth recommendation information based on at least one user-entered text. The at least one user-entered text serves as the basis for determining the recommended item. This fourth recommendation information can be used to explain the recommendation rationale to the user, making them more receptive to the recommended item. This information also helps them understand the recommendation algorithm, increasing their trust in it and improving the recommendation conversion rate.

[0228] The second generation model can be trained using at least one sample text and sample recommendation information corresponding to the at least one sample text. The at least one sample text can serve as input data, and the sample recommendation information can serve as training labels, thereby enabling the training of the second generation model.

[0229] The second generation model can be implemented using a large language model, such as ChatGPT; the large language model can be used to generate natural language text, thereby generating high-quality second recommendation information and displaying the reasons for the recommendation in natural language.

[0230] As an optional method, the third recommendation information and the fourth recommendation information are combined to generate the target recommendation information of the recommendation object. The third recommendation information and the fourth recommendation information can be directly used as the target recommendation information, that is, the target recommendation information is obtained by combining them.

[0231] In addition, in order to further improve the accuracy of the recommendation information and enhance the recommendation effect, as another optional method, the third recommendation information and the fourth recommendation information are fused to generate the target recommendation information of the recommendation object. Based on the third recommendation information and the fourth recommendation information, a fourth image generation model can be used to generate a third recommendation image, which can be sent to the user end as the target recommendation information. The visualization effect is more intuitive in the form of an image. The fourth image generation model can be implemented, for example, using a GAN model, which can be pre-trained using the third sample recommendation information and the fourth sample recommendation information, as well as the third sample recommendation image corresponding to the third sample recommendation information and the fourth sample recommendation information.

[0232] In some embodiments, determining a recommended object that matches at least one user input text includes: identifying at least one first user feature based on the at least one user input text; and determining a recommended object that matches the at least one first user feature.

[0233] The specific implementation method of determining the recommended objects matching at least one first user feature can be found in the embodiment shown in FIG. 2 , and will not be repeated here.

[0234] Among them, a feature extraction model can be used to identify at least one first user feature from at least one user input text. The feature extraction model can be implemented using a machine learning model, which is not limited in this application.

[0235] FIG5 is a flowchart of another embodiment of a recommendation method provided by an embodiment of the present application. The technical solution of this embodiment is executed by a user terminal. The method may include the following steps:

[0236] 501: Provides a conversational interaction interface.

[0237] 502: Display at least one conversation text matching the user on the conversation interaction interface.

[0238] 503: Obtain at least one user input text provided by the user for at least one conversation text.

[0239] 504: Send at least one user input text to the server; the at least one user input text is used to determine matching recommendation objects.

[0240] 505: Get the target recommendation information sent by the server.

[0241] 506: Output target recommendation information.

[0242] The user terminal may display the target recommendation information on a recommendation page for the user to view. The recommendation page may refer to the current page of the user terminal, or may refer to a specific target page.

[0243] Optionally, when the user terminal detects a display operation on the target page, the target recommendation information is displayed on the target page.

[0244] The target page may be, for example, an order success prompt page, an order page, a shopping cart page, or other pages with shopping guide requirements, so as to realize product recommendations before or after shopping.

[0245] In some embodiments, displaying at least one conversation text matching the user on the conversation interaction interface may include:

[0246] Get the first user input text in the current round of conversation;

[0247] Sending the first user input text to the server, so that the server can determine at least one second user input text from historical session information of the current session; and combining the first user input text and the at least one second user input text, using the first session recommendation model to determine a session text in the current session that matches the first user input text;

[0248] The conversation text that matches the first user input text in the current round of conversation is displayed in the conversation interaction interface.

[0249] In some embodiments, displaying at least one conversation text matching the user on the conversation interaction interface may include:

[0250] Obtain the first user input text in the current round of conversation; send the first user input text to the server, so that the server can determine the next conversation text based on the conversation text corresponding to the first user input text; and display the next conversation text in the conversation interaction interface.

[0251] Among them, the at least one conversation text can be determined by the server based on at least one preference feature using the second conversation recommendation model. The at least one conversation text has an output order and can be output in sequence in the conversation interaction interface using multiple rounds of dialogue in the output order.

[0252] The technical solutions of the embodiments of this application can improve the accuracy of recommendation information and can include recommendation basis in the recommendation information, thereby enhancing users' trust in the recommendation algorithm and ultimately improving the efficiency of recommendation conversion. The recommendation effect can also be determined through multiple rounds of conversations, improving users' understanding of the recommendation results. Furthermore, by combining large language models and GAN models, it can not only generate high-quality, realistic information, but also present the recommendation information to users in a more concrete way in natural language, thereby improving the recommendation effect and making users more inclined to use the online system to purchase products, which helps to improve system utilization.

[0253] In the scene interaction diagram of an actual application of the technical solution of the embodiment of the present application shown in Figure 6, the server 102 can provide conversation text to the user terminal 101 in a multi-round conversation manner. The user terminal 101 can provide a conversation interaction interface and display the conversation text, and can perceive the user input operation and send the user input text to the server 102.

[0254] The server 102 can determine at least one user feature from at least one user input text and determine a recommended object that matches the at least one user feature. In addition, it can also generate first recommendation information in combination with the object description information of the recommended object, and generate second recommendation information in combination with at least one user feature, and fuse the first recommendation information and the second recommendation information to generate target recommendation information. After that, the target recommendation information can be sent to the user terminal 101.

[0255] The user terminal 101 may display the target recommendation information on the recommendation page.

[0256] Among them, the target recommendation information may include first recommendation information and second recommendation information. The user terminal 101 may display the target recommendation information in the order result prompt page of any product. Of course, the target recommendation information may also be displayed in other types of target pages. As shown in Figure 7, a schematic diagram of a display interface of a user terminal in an actual application, target recommendation information can be displayed in the display interface, and the target recommendation information may include first recommendation information 701 and second recommendation information 702. The first display style corresponding to the first recommendation information 701 may be different from the second display style corresponding to the second recommendation information 702.

[0257] The first recommendation information or the second recommendation information may include materials such as images and / or texts to enhance the recommendation effect.

[0258] The target recommendation information may also include feedback controls such as the first control 703 and the second control 704 to facilitate the perception of feedback information and send the feedback information to the server 102, which will optimize the recommendation algorithm based on the information. For detailed implementation, please refer to the description in the corresponding embodiment above and will not be repeated here.

[0259] FIG8 is a schematic diagram of the structure of an embodiment of a recommendation device provided in an embodiment of the present application, which may include:

[0260] A first determining module 801 is configured to determine a recommended object that matches at least one first user feature of a user;

[0261] A first generating module 802 is configured to generate first recommendation information of a recommended object based on the object description information of the recommended object;

[0262] A second generating module 803 is configured to generate second recommendation information based on at least one first user feature;

[0263] The third generating module 804 is configured to fuse the first recommendation information and the second recommendation information to generate target recommendation information of the recommendation object;

[0264] The first recommendation module 805 is configured to send target recommendation information to the user terminal.

[0265] In some embodiments, the second generation module can specifically generate second recommendation information based on at least one first user feature using a first generation model; wherein the first generation model is a large language model; the first generation model is trained using at least one first user feature corresponding to the sample object and the second recommendation information corresponding to the sample object.

[0266] In some embodiments, the apparatus may further include:

[0267] A first feature determination module, configured to determine at least one first user feature by combining historical behavior data and / or real-time behavior data of the user;

[0268] In some embodiments, the apparatus may further include:

[0269] The second feature determination module is used to provide a conversation interaction interface to the user terminal; determine at least one conversation text matching the user, and send the at least one conversation text to the user terminal so that the user terminal outputs the at least one conversation text on the conversation interaction interface; obtain at least one user input text provided by the user for the at least one conversation text; and determine at least one first user feature based on the at least one user input text.

[0270] In some embodiments, the apparatus may further include:

[0271] The first material module is configured to obtain at least one first attribute information of a recommended object; obtain at least one second user feature of a user; and generate text material corresponding to the recommended object using a text generation model based on the at least one first attribute information and the at least one second user feature; the object description information includes the text material;

[0272] The text generation model is a large language model; the text generation model is obtained by training using at least one first attribute information and at least one user feature corresponding to the sample object, as well as the sample text corresponding to the sample object.

[0273] In some embodiments, the apparatus may further include:

[0274] The second material module is configured to obtain at least one second attribute information of the recommended object; obtain at least one third user characteristic of the user; and generate an image material corresponding to the recommended object using the first image generation model based on the at least one second attribute information and the at least one third user characteristic, wherein the object description information includes the image material;

[0275] The first image generation model is obtained by training using at least one second attribute information and at least one third user feature corresponding to the sample object, as well as a sample image corresponding to the sample object.

[0276] In some embodiments, the first generating module may specifically extract key information from the object description information of the recommended object as the first material information; based on the object description information and at least one third user feature of the user, generate an object image of the recommended object using a second image generation model, with the object image serving as the second material information; and generate first recommendation information of the recommended object based on the first material information and the second material information;

[0277] The second image generation model is obtained by training using object description information corresponding to the sample object and at least one third user feature, as well as a sample image corresponding to the sample object.

[0278] In some embodiments, the recommendation object may be recommended content. The first determination module may specifically determine a target theme that matches at least one first user feature of the user, and use content materials that match the target theme as the recommended content corresponding to the user.

[0279] In some embodiments, the apparatus may further include:

[0280] The content generation model is used to determine at least one product associated with any topic; and to generate corresponding content materials using the content generation model based on the product-related information corresponding to the at least one product.

[0281] The recommendation device shown in FIG8 can implement the recommendation method described in the embodiment shown in FIG2 . Its implementation principles and technical effects are not further described. The specific manner in which the various modules and units in the recommendation device of the aforementioned embodiment perform their operations has been described in detail in the embodiment of the method and will not be further elaborated here.

[0282] FIG9 is a schematic structural diagram of another embodiment of a recommendation device provided in an embodiment of the present application, which may include:

[0283] The first acquisition module 901 is configured to acquire target recommendation information sent by the server; the target recommendation information is generated by combining the first recommendation information and the second recommendation information; the first recommendation information is generated based on the object description information of the recommended object; the second recommendation information at the Siping city level is generated based on at least one first user characteristic; and the recommended object matches the at least one first user characteristic.

[0284] The first output module 902 is configured to output target recommendation information.

[0285] In some embodiments, the first output module may specifically display the first recommendation information on the recommendation page in a first display style; and display the second recommendation information on the recommendation page in a second display style.

[0286] In some embodiments, the first display style or the second display style includes one or more attributes of display position, display size, display color, font style, display background, and display border;

[0287] At least one attribute of the first display style is different from that of the second display style.

[0288] The recommendation device shown in FIG9 can implement the recommendation method described in the embodiment shown in FIG3 , and its implementation principles and technical effects are not further described. The specific manner in which the various modules and units in the recommendation device of the above embodiment perform operations has been described in detail in the embodiment of the method, and will not be further elaborated here.

[0289] FIG10 is a schematic structural diagram of another embodiment of a recommendation device provided in an embodiment of the present application, which may include:

[0290] The first providing module 1001 is used to provide a conversation interaction interface to the user terminal;

[0291] The second determining module 1002 is configured to determine at least one conversation text matching the user, and send the at least one conversation text to the user terminal so that the user terminal outputs the at least one conversation text on the conversation interaction interface;

[0292] The second acquisition module 1003 is configured to acquire at least one user input text provided by the user for at least one conversation text;

[0293] The third determining module 1004 is configured to determine a recommended object that matches at least one user input text;

[0294] The fourth generating module 1005 is used to generate target recommendation information of the recommendation object;

[0295] The second recommendation module 1006 is configured to send the target recommendation information to the user terminal so that the user terminal can output the target recommendation information.

[0296] In some embodiments, the second determination module may specifically obtain the first user input text in the current round of conversation; determine at least one second user input text from the historical conversation information of the current round of conversation; combine the first user input text and the at least one second user input text, and use the first conversation recommendation model to determine the conversation text in the current round of conversation that matches the first user input text; and send the conversation text to the user end.

[0297] In some embodiments, the second determination module may specifically determine at least one preference feature of the user; based on the at least one preference feature, determine at least one conversation text using the second conversation recommendation model; and send the at least one conversation text to the user end using a multi-round conversation method.

[0298] In some embodiments, the fourth generation module can specifically generate third recommendation information of the recommended object based on the object description information of the recommended object; generate fourth recommendation information using the second generation model based on at least one user input text; fuse the third recommendation information and the fourth recommendation information to generate target recommendation information of the recommended object; and send the target recommendation information to the user end.

[0299] The recommendation device shown in FIG10 can implement the recommendation method described in the embodiment shown in FIG4 , and its implementation principles and technical effects are not further described. The specific manner in which the various modules and units of the recommendation device in the above embodiment perform operations has been described in detail in the embodiment of the method and will not be further elaborated here.

[0300] FIG11 is a schematic structural diagram of another embodiment of a recommendation device provided in an embodiment of the present application, which may include:

[0301] The second providing module 1101 is used to provide a conversation interaction interface;

[0302] A first display module 1102 is configured to display at least one conversation text matching the user on the conversation interaction interface;

[0303] The third acquisition module 1103 is configured to acquire at least one user input text provided by the user for at least one conversation text;

[0304] The sending module 1104 is configured to send at least one user input text to the server; the at least one user input text is used to determine a matching recommendation object;

[0305] The fourth acquisition module 1105 is used to acquire target recommendation information sent by the server;

[0306] The second output module 1106 is configured to output target recommendation information.

[0307] The recommendation device shown in FIG11 can implement the recommendation method described in the embodiment shown in FIG5 , and its implementation principles and technical effects are not further described. The specific manner in which the various modules and units in the recommendation device of the above embodiment perform operations has been described in detail in the embodiment of the method and will not be further elaborated here.

[0308] The embodiment of the present application further provides a computing device, as shown in FIG12 , which may include a storage component 1201 and a processing component 1202 ;

[0309] The storage component 1201 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 1202 to implement the recommendation method shown in FIG. 2 , the recommendation method shown in FIG. 3 , the recommendation method shown in FIG. 4 , or the recommendation method shown in FIG. 5 .

[0310] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0311] In the case where the processing component in the computing device is used to implement the recommendation method shown in FIG3 or the recommendation method shown in FIG5 , the computing device may further include a display component to perform corresponding display operations.

[0312] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc. The communication component is configured to facilitate wired or wireless communication between the computing device and other devices. The processing component can include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0313] The storage component is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0314] The display component may be an electroluminescent (EL) element, a liquid crystal display or a micro display having a similar structure, or a direct retinal display or a similar laser scanning display.

[0315] It should be noted that when the computing device implements the recommended method shown in FIG2 or the recommended method shown in FIG4 , it can be a physical device or an elastic computing host provided by a cloud computing platform. It can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device.

[0316] When the above-mentioned computing device implements the recommendation method shown in Figure 3 or the recommendation method shown in Figure 5, it can be specifically implemented as an electronic device. The electronic device can refer to a device used by the user and has the computing, Internet access, communication and other functions required by the user, such as a mobile phone, tablet computer, personal computer, wearable device, etc.

[0317] It should be noted that the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. It can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device.

[0318] The present application also provides a computer-readable storage medium storing a computer program. When executed by a computer, the computer program can implement the recommendation method shown in FIG2 , the recommendation method shown in FIG3 , the recommendation method shown in FIG4 , or the recommendation method shown in FIG5 . The computer-readable medium can be included in the electronic device described in the above embodiments, or it can exist independently and not be incorporated into the electronic device.

[0319] The present application also provides a computer program product, including a computer program carried on a computer-readable storage medium. When the computer program is executed by a computer, it can implement the recommendation method shown in Figure 2, the recommendation method shown in Figure 3, the recommendation method shown in Figure 4, or the recommendation method shown in Figure 5. In such an embodiment, the computer program can be downloaded and installed from a network and / or installed from a removable medium. When the computer program is executed by a processor, it performs the various functions defined in the system of the present application.

[0320] It should be noted that the embodiments of the present application may involve the use of user data. In actual applications, user-specific personal data can be used in the scheme described herein within the scope permitted by applicable laws and regulations, subject to the requirements of applicable laws and regulations of the country where the user is located (for example, with the user's explicit consent, effective notification to the user, etc.).

[0321] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0322] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0323] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0324] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A recommendation method, characterized in that: include: Determining a recommended object that matches at least one first user characteristic of the user; generating first recommendation information of the recommended object according to the object description information of the recommended object; generating second recommendation information according to the at least one first user feature; Fusion the first recommendation information and the second recommendation information to generate target recommendation information; The target recommendation information is sent to the user terminal.

2. The method according to claim 1, characterized in that: The generating the second recommendation information according to the at least one preference feature comprises: Generate second recommendation information using a first generation model according to the at least one first user feature; The first generation model is a large language model; the first generation model is trained by using at least one first user feature corresponding to a sample object and second recommendation information corresponding to the sample object.

3. The method according to claim 1, characterized in that Also includes: Determine the at least one first user feature in combination with the historical behavior data and / or the real-time behavior data of the user; or, Providing a conversation interaction interface to the user terminal; and determining at least one conversation text matching the user, and sending the at least one conversation text to the user terminal, so that the user terminal outputs the at least one conversation text on the conversation interaction interface; And obtaining at least one user input text provided by the user for the at least one conversation text; and determining the at least one first user feature according to the at least one user input text.

4. The method according to claim 1, characterized in that: Also includes: Acquire at least one first attribute information of the recommended object; Acquire at least one second user characteristic of the user; Based on the at least one first attribute information and the at least one second user feature, a text material corresponding to the recommended object is generated using a text generation model; the object description information includes the text material; The text generation model is a large language model; the text generation model is obtained by training using at least one first attribute information and at least one user feature corresponding to a sample object, and a sample text corresponding to the sample object.

5. The method according to claim 1, characterized in that Also includes: Acquire at least one second attribute information of the recommended object; acquiring at least one third user characteristic of the user; Based on the at least one second attribute information and the at least one third user feature, using the first image generation model to generate an image material corresponding to the recommended object, the object description information including the image material; The first image generation model is obtained by training using at least one second attribute information and at least one third user feature corresponding to a sample object, and a sample image corresponding to the sample object.

6. The method according to claim 1, characterized in that The generating first recommendation information of the recommended object according to the object description information of the recommended object comprises: Extracting key information from the object description information of the recommended object as first material information; Based on the object description information and at least one third user feature of the user, using a second image generation model to generate an object image of the recommended object, the object image serving as second material information; Based on the first material information and the second material information, generating first recommendation information of the recommended object; The second image generation model is obtained by training using object description information corresponding to the sample object and at least one third user feature, as well as a sample image corresponding to the sample object.

7. The method according to claim 1, characterized in that The recommended object is the recommended content; The determining of the recommended object matching at least one first user feature of the user comprises: determining a target topic that matches at least one first user characteristic of the user; The content material that matches the target theme is used as the recommended content corresponding to the user.

8. The method according to claim 7, characterized in that Also includes: For any topic, determining at least one commodity associated with the topic; According to the commodity-related information respectively corresponding to the at least one commodity, corresponding content material is generated by using a content generation model.

9. A recommendation method, characterized in that: include: Get target recommendation information; The target recommendation information includes first recommendation information and second recommendation information; The first recommendation information is generated according to the object description information of the recommendation object; the second recommendation information is generated according to at least one first user feature; The recommended object matches the at least one first user feature; Displaying the first recommendation information on the recommendation page in a first display style; The second recommendation information is displayed on the recommendation page in a second display style.

10. A recommendation method, characterized in that: include: Providing a conversational interaction interface to the user end; Determining at least one conversation text matching the user, and sending the at least one conversation text to the user terminal, so that the user terminal outputs the at least one conversation text on the conversation interaction interface; Acquire at least one user input text provided by a user for the at least one conversation text; Determining a recommended object matching the at least one user input text; Generating target recommendation information of the recommended object; The target recommendation information is sent to the user terminal.

11. The method according to claim 10, characterized in that The determining at least one conversation text matching the user and sending the at least one conversation text to the user terminal comprises: Acquire a first user input text in a current round of conversation; and determine at least one second user input text from historical conversation information of the current round of conversation; and determine a conversation text matching the first user input text in the current round of conversation using a first conversation recommendation model in combination with the first user input text and the at least one second user input text; and send the conversation text to the user terminal; or, Determine at least one preference feature of the user; and based on the at least one preference feature, determine at least one conversation text using a second conversation recommendation model; and send the at least one conversation text to the user terminal in a multi-round conversation manner.

12. A recommendation method, characterized in that: include: Provide a conversational interaction interface; Displaying at least one conversation text matching the user on the conversation interaction interface; Acquire at least one user input text provided by the user for the at least one conversation text; Sending the at least one user input text to the server; the at least one user input text is used to determine a matching recommended object; Obtaining target recommendation information sent by the server; The target recommendation information is output.

13. A computing device, characterized in that: including a processing component and a storage component; The storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the recommendation method as described in any one of claims 1 to 8, the recommendation method as described in claim 9, the recommendation method as described in any one of claims 10 to 11, or the recommendation method as described in any one of claim 12.

14. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, the recommendation method according to any one of claims 1 to 8, the recommendation method according to claim 9, the recommendation method according to any one of claims 10 to 11, or the recommendation method according to any one of claim 12 is implemented.

Images