Content generation method and system

Abstract

The content generation method and system provided in this specification obtain the target features corresponding to the target user, as well as the target associated users and their corresponding target associated features. When generating content based on the content prompt data input by the target user, the target features are enriched by the target associated features to obtain more accurate user personalized features. This guides the generated content to develop in a direction that meets the user's personalized characteristics, thereby obtaining personalized target content and improving the matching degree between the target content and the personalized characteristics of the target user.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence, and in particular to a content generation method and system. Background Technology

[0002] Artificial Intelligence Generated Content (AIGC) is a content production method that uses artificial intelligence to find patterns in existing data and automatically generate content. It has been widely used in numerous applications. To make AIGC-generated content more tailored to user needs, users often manually adjust the content prompts or use reference images to refine the content. However, the content generated by these methods still has a low degree of personalization and may not meet application requirements.

[0003] In summary, there is a need to provide a new content generation method and system that can improve the matching degree between AIGC content and user personalization.

[0004] The information in the background section is merely information known only to the inventor and does not imply that such information had entered the public domain before the date of this application, nor does it imply that it can be considered prior art in this disclosure. Summary of the Invention

[0005] This specification provides a method and system for generating content that is more closely matched to user personalization.

[0006] In a first aspect, this specification provides a content generation method, comprising: acquiring content prompt data input by a target user and its corresponding target associated user, wherein the target associated user is an associated user with similar content preferences to the target user; acquiring target features of the target user and target association features of the target associated user; determining personalized target content corresponding to the target user based on the content prompt data, the target features, and the target association features, wherein the personalized target content matches the content prompt data; and outputting the personalized target content.

[0007] In some embodiments, the target feature includes at least one of the target user's profile features and historical content features, the target association feature includes at least one of the target associated user's profile features and historical content features, and the historical content features include historical content prompt data features and their corresponding historical adopted content features.

[0008] In some embodiments, obtaining the content prompt data input by the target user and its corresponding target associated user includes: obtaining M candidate associated users corresponding to the target user and their corresponding M candidate associated features, where M is a positive integer; determining M correlation degrees between the target user and the M candidate associated users based on the correlation relationship between the target user and the M candidate associated users, and comparing the target feature with the M candidate associated features; and filtering out the target associated user corresponding to the target user from the M candidate associated users based on the M correlation degrees.

[0009] In some embodiments, obtaining the M candidate associated users and their corresponding M candidate associated features corresponding to the target user includes: determining multiple similarities between the target feature and the user features of multiple users in the user database; sorting the multiple similarities in descending order to obtain a similarity ranking result; and selecting the users corresponding to the top M similarities in the similarity ranking result as the M candidate associated users.

[0010] In some embodiments, the association relationship includes at least one of shared device relationship, shared network relationship, and information transmission relationship, and the profile features of the target user and the profile features of the M candidate associated users include relationship features, which include at least one of device features, network features, and information transmission features.

[0011] In some embodiments, the profile features are features obtained based on profile data, which includes at least one of personal information data, recently viewed text, recently viewed images, and recently viewed videos; the profile data also includes at least one of device data, network data, and information transmission data.

[0012] In some embodiments, determining the M correlation degrees between the target user and the M candidate associated users based on the association relationship between the target user and the M candidate associated users, and the comparison between the target profile features and the M candidate associated profile features, includes: inputting the target features and the M candidate associated features into an associated user prediction model to obtain the M correlation degrees between the target user and the M candidate associated users; the training objective of the associated user prediction model includes constraining the difference between the predicted correlation degree and the actual correlation degree between sample user pairs to be less than a first preset difference threshold.

[0013] In some embodiments, the step of selecting the target associated user corresponding to the target user from the M candidate associated users based on the M correlation degrees includes: sorting the M correlation degrees in descending order; and selecting the L candidate associated users whose correlation degree is greater than a preset correlation degree threshold or whose ranking is higher from the sorting results as the target associated user, where L is a positive integer less than or equal to M.

[0014] In some embodiments, determining the personalized target content corresponding to the target user based on the content prompt data, the target feature, and the target association feature includes: fusing the target feature and the target association feature to obtain the personalized feature of the target user; and using the personalized feature as the diffusion direction of content generation, and generating content based on the content prompt data to obtain the personalized target content.

[0015] In some embodiments, the personalized target content is generated using a content generation model. The training data of the content generation model includes sample content prompt data corresponding to multiple sample users and sample personalized features. The sample personalized features are obtained based on the sample profile features of the corresponding sample user and associated sample users. The training objective includes constraining the difference between the predicted generated content output based on the sample content prompt data and the sample personalized features and the actual generated content to be less than a second preset difference threshold.

[0016] In some embodiments, the training objective further includes: constraining the difference between the predicted generated content features and the sample historical content features of the corresponding sample users to be less than a third preset difference threshold, wherein the predicted generated content features are obtained based on the predicted generated content.

[0017] In some embodiments, the sample historical content features include at least one of the average historical content prompt data features of the corresponding sample user and the associated sample related user and the corresponding average historical adopted content features.

[0018] Secondly, this specification also provides a content generation system, comprising: at least one storage medium storing at least one instruction set for content generation; and at least one processor communicatively connected to the at least one storage medium, wherein, when the content generation system is running, the at least one processor reads the at least one instruction set and executes the content generation method described in the first aspect of this specification according to the instructions of the at least one instruction set.

[0019] As can be seen from the above technical solutions, the content generation method and system provided in this specification, after obtaining the content prompt data input by the target user, further obtain the target associated user corresponding to the target user, as well as the target features corresponding to the target user and the target associated features corresponding to the target associated user. When generating target content for the target user based on the content prompt data, the target features are supplemented by the target associated features, thereby obtaining more accurate user personalized features to guide the content generation process towards the direction of user personalized features, generate target content that conforms to user personalized characteristics, and improve the matching degree between target content and user personalized characteristics.

[0020] Other functions of the content generation methods and systems provided in this specification will be partially listed in the following description. The figures and examples described below will be apparent to those skilled in the art. The inventive aspects of the content generation methods and systems provided in this specification can be fully understood through practice or use of the methods, apparatus, and combinations described in the detailed examples below. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A schematic diagram illustrating an application scenario of a content generation system provided according to an embodiment of this specification is shown.

[0023] Figure 2 A hardware structure diagram of a computing device provided according to an embodiment of this specification is shown;

[0024] Figure 3 A flowchart of a content generation method provided according to an embodiment of this specification is shown;

[0025] Figure 4 A schematic diagram illustrating the data flow of a training process for a preset associated user prediction network provided according to embodiments of this specification is shown; and

[0026] Figure 5 A schematic diagram illustrating the data flow during the training process of a preset cold start content generation network provided according to an embodiment of this specification is shown. Detailed Implementation

[0027] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.

[0028] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.

[0029] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0030] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0031] For ease of description, the terms that will appear in the following descriptions will be explained as follows:

[0032] AIGC refers to methods of generating content using artificial intelligence (AI) technology. This content can include text, audio, images, and video, among other things.

[0033] Related users: refers to users who have the same or similar content preferences as the current user.

[0034] Personalized content generation: This refers to the generation of different personalized content when different users input the same prompt.

[0035] Before describing the specific embodiments in this specification, the application scenarios of this specification will be introduced as follows:

[0036] The content generation methods provided in this manual can be applied to any scenario requiring content generation. For example, in text generation scenarios, the AIGC model can be used to quickly generate initial text content, such as articles, news reports, advertising copy, and creative writing, thus providing writers and editors with an efficient way to create content, quickly generate first drafts, and improve text creation efficiency. In image generation scenarios, initial image content can be automatically generated based on user-input keywords or real images, such as generating red envelope covers using the AIGC model. In video generation scenarios, the AIGC model can be used to automatically generate initial video content, such as short videos, animations, and movie trailers, thereby accelerating the video creation process through automation. In addition, it can automatically generate subtitles and video descriptions, as well as automated video editing and rendering, thereby improving the efficiency and quality of video production. In audio generation scenarios, the AIGC model can also be used to generate various initial audio content, such as mobile phone navigation voice, text reading, and virtual human voices, thus providing an efficient way to produce audio.

[0037] After generating initial content, if the user is not satisfied with it, they can adjust the content hint data themselves, such as expanding the hint data, adjusting the style, replacing synonyms, etc. However, this method has a high barrier to entry for users and relies heavily on human experience. Users can also provide a reference image, allowing the AIGC model to generate new AIGC content based on specific cues from the reference image (such as contours, depth, or segmentation maps). However, this method requires manual specification of the cue type and manual selection of the reference image by the user, also relying heavily on human experience. Therefore, this specification provides a new content generation method that introduces personalized data of the user associated with the current user during the content generation process to create a profile of the current user, thereby enriching the user's profile features and making the content generated based on the user's profile features more closely matched to the user's personalization. The content generation method of this specification can also be applied to any content generation scenario, which will not be elaborated upon here. For ease of description, this application will use the application of the content generation method in an image generation scenario to optimize the initial image generated by the AIGC model as an example.

[0038] Those skilled in the art should understand that the content generation methods and systems described in this specification, when applied to other use cases, are also within the scope of protection of this specification.

[0039] Figure 1This diagram illustrates an application scenario of a content generation system 001 provided according to an embodiment of this specification. The content generation system 001 (hereinafter referred to as System 001) can be applied to content generation in any scenario, such as content generation in scenarios involving text content, image content, audio content, and video content. Figure 1 As shown, system 001 may include terminal device 200. In some embodiments, system 001 may include server 300. The application scenarios of system 001 may include target user 100, system 001, and network 400.

[0040] The target user 100 can be a user who needs to generate content or a user who is currently generating content.

[0041] Terminal device 200 can be a device for generating content for target user 100. In some embodiments, the content generation method can be executed on terminal device 200. In this case, terminal device 200 may store data or instructions for executing the content generation method described herein, and may execute or be used to execute the data or instructions. In some embodiments, terminal device 200 may include hardware devices with data processing capabilities and necessary programs for driving the hardware devices. In some embodiments, terminal device 200 may include mobile devices, tablets, laptops, built-in devices in motor vehicles, or similar content, or any combination thereof. In some embodiments, the mobile device may include smart home devices, smart mobile devices, virtual reality devices, augmented reality devices, or similar devices, or any combination thereof. In some embodiments, the smart home device may include smart TVs, desktop computers, etc., or any combination thereof. In some embodiments, the smart mobile device may include smartphones, personal digital assistants, gaming devices, navigation devices, etc., or any combination thereof. In some embodiments, the virtual reality device or augmented reality device may include virtual reality headsets, virtual reality glasses, virtual reality patches, augmented reality headsets, augmented reality glasses, augmented reality patches, or similar content, or any combination thereof. For example, the virtual reality device or the augmented reality device may include Google Glass, a head-mounted display, VR, etc. In some embodiments, the built-in device in the motor vehicle may include an in-vehicle computer, an in-vehicle television, etc. In some embodiments, the terminal device 200 may be a device with positioning technology for locating the position of the terminal device 200.

[0042] In some embodiments, the terminal device 200 may have one or more applications (APPs) installed. The APP provides the target user 100 with the ability and interface to interact with the outside world via the network 400. The APPs include, but are not limited to: web browser APPs, search APPs, chat APPs, shopping APPs, video APPs, financial management APPs, instant messaging tools, email terminals, social media platform software, etc. In some embodiments, a target APP may be installed on the terminal device 200. The target APP can collect content prompt data input by the target user for the terminal device 200. In some embodiments, the target user 100 can also trigger a content generation request through the target APP. The target APP can respond to the content generation request by executing the content generation method described in this specification. The content generation method will be described in detail later.

[0043] like Figure 1 As shown, terminal device 200 can communicate with server 300. In some embodiments, server 300 can communicate with multiple terminal devices 200 and receive data sent by terminal devices 200. In some embodiments, terminal device 200 can interact with server 300 through network 400 to receive or send messages, etc. Server 300 can be a server that provides various services, such as a backend server that supports content generation methods deployed on multiple terminal devices 200. In some embodiments, the content generation method can be executed on server 300. In this case, server 300 can store data or instructions for executing the content generation method described in this specification, and can execute or be used to execute the data or instructions. In some embodiments, server 300 may include hardware devices with data information processing capabilities and the necessary programs required to drive the hardware devices.

[0044] Network 400 serves as a medium to provide a communication connection between terminal device 200 and server 300. Network 400 facilitates the exchange of information or data. For example... Figure 1 As shown, terminal device 200 and server 300 can connect to network 400 and transmit information or data to each other through network 400. In some embodiments, network 400 can be any type of wired or wireless network, or a combination thereof. For example, network 400 may include cable network, wired network, fiber optic network, telecommunications network, intranet, Internet, local area network (LAN), wide area network (WAN), wireless local area network (WLAN), metropolitan area network (MAN), public switched telephone network (PSTN), Bluetooth network. TM ZigBee TMA network, a near-field communication (NFC) network, or a similar network. In some embodiments, network 400 may include one or more network access points. For example, network 400 may include wired or wireless network access points, such as base stations or internet exchange points, through which one or more components of terminal device 200 and server 300 can connect to network 400 to exchange data or information.

[0045] It should be understood that Figure 1 The number of terminal devices 200, servers 300, and networks 400 shown is merely illustrative. Depending on implementation needs, any number of terminal devices 200, servers 300, and networks 400 can be included.

[0046] It should be noted that the content generation method can be executed entirely on terminal device 200, entirely on server 300, or partially on terminal device 200 and partially on server 300. The following description will use the execution of the content generation method on both terminal device 200 and server 300 as an example.

[0047] Figure 2 A hardware structure diagram of a computing device 600 according to an embodiment of this specification is shown. The computing device 600 can execute the content generation method described in this specification. The content generation method is described in other parts of this specification. When the content generation method is executed on a terminal device 200, the computing device 600 can be the terminal device 200. When the content generation method is executed on a server 300, the computing device 600 can be the server 300. When the content generation method is executed partly on the terminal device 200 and partly on the server 300, the computing device 600 can be either the terminal device 200 or the server 300.

[0048] like Figure 2 As shown, the computing device 600 may include at least one storage medium 630 and at least one processor 620. In some embodiments, the computing device 600 may also include a communication port 650 and an internal communication bus 610. Additionally, the computing device 600 may include I / O components 660.

[0049] The internal communication bus 610 can connect different system components, including storage medium 630, processor 620 and communication port 650.

[0050] I / O component 660 supports input / output between computing device 600 and other components.

[0051] Communication port 650 is used for data communication between computing device 600 and external sources. For example, communication port 650 can be used for data communication between computing device 600 and network 400. Communication port 650 can be a wired communication port or a wireless communication port.

[0052] Storage medium 630 may include a data storage device. The data storage device may be a non-transitory storage medium or a temporary storage medium. For example, the data storage device may include one or more of a disk 632, a read-only storage medium (ROM) 634, or a random access storage medium (RAM) 636. Storage medium 630 also includes at least one instruction set stored in the data storage device. The instructions are computer program code, which may include programs, routines, objects, components, data structures, procedures, modules, etc., that execute the content generation methods provided in this specification.

[0053] At least one processor 620 can be communicatively connected to at least one storage medium 630 and a communication port 650 via an internal communication bus 610. The at least one processor 620 is used to execute the at least one instruction set described above. When the computing device 600 is running, the at least one processor 620 reads the at least one instruction set and, according to the instructions of the at least one instruction set, executes the content generation method provided in this specification. The processor 620 can execute all the steps included in the content generation method. The processor 620 can be in the form of one or more processors. In some embodiments, the processor 620 may include one or more hardware processors, such as a microcontroller, microprocessor, reduced instruction set computer (RISC), application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), central processing unit (CPU), graphics processing unit (GPU), physical processing unit (PPU), microcontroller unit, digital signal processor (DSP), field-programmable gate array (FPGA), advanced RISC machine (ARM), programmable logic device (PLD), any circuit or processor capable of performing one or more functions, or any combination thereof. For illustrative purposes only, only one processor 620 is described in this specification for the computing device 600. However, it should be noted that the computing device 600 in this specification may also include multiple processors. Therefore, the operation and / or method steps disclosed in this specification may be executed by one processor as described in this specification, or they may be executed jointly by multiple processors. For example, if the processor 620 of the computing device 600 in this specification executes steps A and B, it should be understood that steps A and B may also be executed jointly or separately by two different processors 620 (e.g., the first processor executes step A, the second processor executes step B, or the first and second processors jointly execute steps A and B).

[0054] Figure 3 A flowchart of a content generation method P100 according to an embodiment of this specification is shown. As previously described, computing device 600 can execute the content generation method P100 of this specification. Specifically, processor 620 can read an instruction set stored in its local storage medium and then execute the content generation method P100 of this specification according to the instructions in the instruction set. Figure 3 As shown, method P100 may include:

[0055] S120: Obtain the content prompt data input by the target user and its corresponding target associated user.

[0056] Content prompts are input guidance tools used to guide users to input appropriate keywords to accurately describe the target content generated based on their intent. This allows the machine to accurately understand the user's intent and generate high-quality content that matches it. Content can include text, audio, images, and video, among other things. By acquiring the target user's content prompts, the computing device 600 can help the AIGC model understand the target user's intent and generate matching content. Content prompts can be a piece of text, at least one question, at least one keyword, or at least one image. For example, content prompts could be keywords describing a girl, such as hairstyle, face shape, clothing style, clothing color, or facial expression. Alternatively, content prompts could be a real image input by the target user and a selected reference image. Different image types can be selected for the reference image. Image types can include image styles, such as characters from comics or movies.

[0057] In some embodiments, when the computing device 600 is a terminal device, the computing device 600 directly obtains the content prompt data input by the target user 100 on the terminal device. In some embodiments, when the computing device 600 includes a terminal device and a server, the terminal device sends the content prompt data input by the target user to the server after obtaining it.

[0058] Target related users refer to users who share similar content preferences with the target user. Similar content preferences include identical or highly similar content preferences. Content preference refers to a user's preference for a specific characteristic of content. Taking images as an example, content preference could be a user's preferred image style, or a preference for certain image parameters, etc. Image parameters could be image sharpness, brightness, etc.

[0059] The target associated user can be obtained based on a user database. The user database includes multiple users and their corresponding user characteristics. These multiple users can be users registered on the AIGC platform, users who have browsed content on the AIGC platform, or users who have generated historical content on the AIGC platform. User characteristics can be obtained in the following ways: In some embodiments, the computing device 600 can obtain profile data corresponding to multiple users, perform feature encoding on the profile data corresponding to multiple users to obtain intermediate user characteristics corresponding to multiple users, and perform feature compression on the intermediate user characteristics corresponding to multiple users to obtain user characteristics corresponding to multiple users. That is, the profile characteristics are features obtained based on profile data. The profile data can include at least one of the following: personal information data, text content recently browsed by the user on the AIGC platform, image content recently browsed by the user on the AIGC platform, and video content recently browsed by the user on the AIGC platform. Correspondingly, intermediate user characteristics can include multiple intermediate sub-user characteristics, which can also be called multiple sub-profile characteristics. Multiple sub-profile characteristics can include at least one of intermediate personal information profile characteristics, intermediate text profile characteristics, intermediate image profile characteristics, and intermediate video profile characteristics. Correspondingly, user characteristics can include multiple sub-user characteristics. Multiple sub-user features may include at least one of personal information profile features, text profile features, image profile features, and video profile features.

[0060] In some embodiments, the computing device 600 can also obtain profile data and historical content generation data corresponding to multiple users, perform feature encoding on the profile data and historical content generation data corresponding to multiple users to obtain intermediate user features corresponding to multiple users, and perform feature compression on the intermediate user features corresponding to multiple users to obtain user features corresponding to multiple users. The historical content generation data may include historical content prompt data and historical adopted content.

[0061] In other words, user characteristics can include at least one of profile characteristics and historical content characteristics.

[0062] Intermediate user features can be obtained in the following ways: For example, for personal information data, computing device 600 can encode it into a data feature vector based on rule-based encoding to obtain intermediate personal information profile features. For text content, computing device 600 can use a Large Language Model (LLM) to encode the text content that the user has recently browsed on the AIGC platform to obtain intermediate text profile features. For image content, computing device 600 can use a Contrastive Language-Image Pre-Training (CLIP) model to encode the image content that the user has recently browsed to obtain intermediate image profile features. For video content, computing device 600 can use a CLIP model to encode the video content that the user has recently browsed to obtain intermediate video profile features.

[0063] After obtaining intermediate personal information profile features, intermediate text profile features, intermediate image profile features, and intermediate video profile features, the computing device 600 can further compress these four intermediate profile features to a preset dimension, thereby obtaining the user features. The preset dimension can be 128 dimensions, 64 dimensions, etc. Feature compression can be achieved using an autoregressive model.

[0064] After obtaining user characteristics corresponding to multiple users, the computing device 600 can generate a mapping table based on the identifiers and corresponding user characteristics of multiple users and store it in the user database. This facilitates the retrieval of users with similar content preferences during subsequent content generation. The computing device 600 can store user characteristics in the following way: for example, for user A, its corresponding user characteristics can be denoted as fA; for user B, its corresponding user characteristics can be denoted as Fb, and so on. Multiple users in the user database can be labeled to distinguish different users and their corresponding initial user characteristics. The user identifier is data that represents the user's identity and can uniquely identify the user to distinguish different users. The user identifier can be the user's registered account on the AIGC platform, mobile phone number, etc.

[0065] After obtaining the aforementioned mapping table, the computing device 600 can filter out target associated users from multiple users based on the target user and the mapping table. The computing device 600 can perform coarse-grained filtering, fine-grained filtering, or both simultaneously. To ensure both filtering speed and accuracy, the computing device 600 can perform coarse-grained filtering to narrow the filtering range, and then perform fine-grained filtering within the narrowed range. The coarse-grained filtering and fine-grained filtering are referred to as the first stage and the second stage, respectively. In the first stage of filtering, the computing device 600 can determine the similarity between the target user's user features and the user features of multiple users, obtain the similarity scores of multiple user features, sort the similarity scores from high to low, obtain the similarity ranking results, and select the M users corresponding to the top M user features in the similarity ranking results as M candidate associated users. The similarity between the target user and the user features of multiple users can be calculated using cosine similarity, Euclidean distance, or Manhattan distance, etc. For ease of description, the user features of the M candidate associated users are denoted as the M candidate associated features. The user features corresponding to the target user are denoted as the target features. That is, the target features include at least one of the target user's profile features and historical content features.

[0066] In the first-stage screening process, the compression of intermediate user features reduces the feature dimensionality, thus improving the screening efficiency. This means the first-stage screening process can efficiently guarantee the recall rate of associated users. To further improve screening accuracy, and to exclude potentially noisy users with low relevance from the M candidate associated users selected in the first stage, the computing device 600 can perform further fine-grained screening. Therefore, the computing device 600 improves screening accuracy through a second-stage screening, aiming to select candidate associated users with high relevance (which can be understood as relevance greater than a preset relevance), i.e., target associated users. The second-stage screening can be implemented as follows: For example, the computing device 600 obtains the M candidate associated users corresponding to the target user and their corresponding M candidate associated features, where M is a positive integer; based on the association relationship between the target user and the M candidate associated users, and the comparison between the intermediate user features corresponding to the target user and the intermediate user features corresponding to the M candidate associated users, it determines the M relevance degrees between the target user and the M candidate associated users; and based on the M relevance degrees, it selects the target associated users corresponding to the target user from the M candidate associated users.

[0067] When computing device 600 determines the M degrees of association between the target user and the M candidate associated users based on the association relationship between the target user and the M candidate associated users, and by comparing the intermediate user features corresponding to the target user with the intermediate user features corresponding to the M candidate associated users, there are multiple implementation methods, as follows:

[0068] In some embodiments, the computing device 600 may sequentially designate each of the multiple sub-features of the intermediate user features as a target sub-feature, and compare the target sub-feature with the corresponding sub-feature among M candidate associated features to obtain M similarities between the target sub-feature and the corresponding sub-feature among the M candidate associated features. Then, the M similarities are weighted and summed to obtain M target sub-similarity between the target sub-feature and the corresponding M candidate associated features. Repeating the above process for each of the multiple sub-features yields M target sub-similarity for each of the multiple sub-features. Finally, the computing device 600 weighted and sums the M target sub-similarity for each of the multiple sub-features to obtain M associations between the target user and the M candidate associated users.

[0069] In some embodiments, the computing device 600 may also acquire M candidate associated users corresponding to the target user and their corresponding M candidate associated features; determine M degrees of association between the target user and the M candidate associated users based on the association relationship between the target user and the M candidate associated users, and the comparison between the target feature and the M candidate associated features; and filter out the target associated user corresponding to the target user from the M candidate associated users based on the M degrees of association.

[0070] When acquiring the M candidate associated users and their corresponding M candidate associated features corresponding to the target user, the computing device 600 may adopt the following method: For example, the computing device 600 determines multiple similarities between the target feature and the user features of multiple users in the user database; sorts the multiple similarities in descending order to obtain the similarity ranking result; and selects the users corresponding to the top M similarities in the similarity ranking result as the M candidate associated users.

[0071] The relationships can include at least one of shared device relationships, shared network relationships, and information transmission relationships. The profile features of the target user and the profile features of the M candidate associated users include relationship features. Relationship features include at least one of device features, network features, and information transmission features. The profile data may also include data corresponding to the relationship features, such as at least one of device data, network data, and information transmission data. Specifically, a shared device relationship refers to different users sharing a single device for content generation. A shared network relationship refers to different users sharing the same network for content generation. An information transmission relationship refers to information sharing, such as user A sharing a link with user B, users A and B frequently chatting, or there being transfers between users A and B, etc., involving data flow relationships between users A and B.

[0072] For example, based on the profile data described above, if user A and user B share a device for content generation, or share the same network for content generation, or have an information transmission relationship, or are colleagues, family members, share an address, or have a message exchange relationship, computing device 600 can determine that there is a relationship between user A and user B. This relationship can be obtained based on a network topology diagram of user relationships. The network topology diagram includes multiple nodes, each representing a user. When two users are related, they are connected by a straight line; when two users are not related, they are not connected.

[0073] When determining the M correlation degrees between the target user and the M candidate associated users based on the association relationship between the target user and the M candidate associated users, and by comparing the target feature with the M candidate associated features, the computing device 600 can use a neural network model. For example, the computing device 600 acquires a sample dataset and trains a pre-defined associated user prediction network based on the sample dataset to obtain a trained associated user prediction model. The sample dataset can include multiple sample user pairs. The two users in each sample user pair are denoted as the current sample user and its corresponding associated user. Each sample user pair has a corresponding true correlation degree. The true correlation degree is the label of the sample user pair, representing whether the associated user is a highly correlated user or a less correlated user of the current sample user. The true correlation degree can be obtained through manual annotation.

[0074] After obtaining the sample dataset, the computing device 600 can train a pre-defined associated user prediction network based on the sample dataset to obtain an associated user prediction model. Each sample user in the sample dataset can correspond to one highly associated user, multiple highly associated users, or noisy user samples (i.e., associated users with an association degree less than a pre-defined association degree threshold). Noisy users refer to non-highly associated user samples. The training process of the pre-defined associated user prediction network is described below with reference to the accompanying diagram:

[0075] Figure 4 A schematic diagram illustrating the data flow of a training process for a preset associated user prediction network provided according to an embodiment of this specification is shown. Figure 4 As shown, the preset associated user prediction network may include a preset target user feature encoding network, a preset associated user feature encoding network, and an association matching network.

[0076] To improve training efficiency, the computing device 600 can use the user features of the sample user pairs as training data. For ease of description, the user features corresponding to the current sample user are denoted as the first sample user features, and the user features of the sample-related users corresponding to the current sample user are denoted as the first sample-related user features.

[0077] The preset target user feature encoding network can be a Transform network. The computing device 600 inputs the first sample user features into the preset target user feature encoding network, so that the preset target user feature encoding network encodes the first sample user features to obtain the corresponding predicted second sample user features.

[0078] The preset associated user feature encoding network can be a Transform network. Similar to the principle of the preset target user feature encoding network, the computing device 600 can input the first sample associated user features into the preset associated user feature encoding network, so that the preset associated user feature encoding network encodes the first sample associated user features to obtain the predicted second sample associated user features.

[0079] The preset association matching network can be a multilayer perceptron (MLP) or a Transform network. The computing device 600 can input the predicted features of the second sample user and the predicted features of the second sample associated users into the preset association matching network, so that it can predict the association degree between the current sample user and its corresponding associated sample users, thereby obtaining the predicted association degree.

[0080] The predicted correlation degree represents the predicted correlation between the current sample user and its corresponding associated sample users. The predicted correlation degree can be represented by probability. The computing device 600 can also determine the predicted correlation classification result based on the predicted correlation degree. The predicted correlation classification result indicates whether the associated sample users in the sample user pair are highly correlated users of the current sample user.

[0081] After obtaining the predicted correlation degree, the computing device 600 can determine the correlation classification loss based on the difference between the predicted correlation classification result and the actual correlation classification result, and converge the preset correlation user prediction network based on the correlation classification loss to obtain the trained correlation user prediction model. The target feature of the target user and the M candidate correlation features of its corresponding M candidate correlation users are input into the correlation user prediction model to obtain the M correlation degrees between the target user and the M candidate correlation users.

[0082] The training objective of the associated classification loss is to make it less than the first preset difference threshold.

[0083] M correlation scores represent the degree of correlation between the target user and M candidate related users. After obtaining the M correlation scores, the computing device 600 can sort the M correlation scores in descending order, and select the L candidate related users with correlation scores greater than a preset correlation score threshold or the top-ranked correlation scores from the sorted results as the target related users, where L is a positive integer less than or equal to M.

[0084] Continue reading Figure 3 After step S120, the method P100 may further include the following step S140.

[0085] S140: Obtain the target characteristics of the target user and the target association characteristics of the target related user.

[0086] Target user features refer to the user characteristics corresponding to the target user. Target association features refer to the user characteristics corresponding to the target associated user. Target features may include at least one of the target user's profile features and historical content features. Target association features may include at least one of the target associated user's profile features and historical content features. Historical content features may include historical content prompt data features and their corresponding historical adopted content features. As mentioned above, the user database stores a mapping relationship table. The computing device 600 can obtain the target user's target features by looking up the mapping relationship table based on the target user's identifier. Similarly, the computing device 600 can also obtain the target association features of the target associated user by looking up the mapping relationship table based on the target associated user's identifier.

[0087] S160: Based on content prompt data, target features, and target association features, determine the personalized target content corresponding to the target user.

[0088] In this process, personalized target content is matched with content prompt data. Target features represent the target user's content preferences. Target association features represent the content preferences of related target users. During content generation, the computing device 600 can supplement the target user's user features using the user features of related target users, thereby obtaining personalized features that better match the target user's content preferences. This improves the accuracy of the target user profile, enabling different personalized content to be generated for different users even when using the same content prompt data.

[0089] When the computing device 600 determines the personalized target content corresponding to the target user based on content prompt data, target features, and target association features, it can perform feature fusion on the target features and target association features to obtain the personalized features of the target user, and use the personalized features as the diffusion direction of content generation. Based on the content prompt data, it generates content to obtain personalized target content.

[0090] The computing device 600 can use feature fusion to combine target features and target-related features into personalized features for the target user. Feature fusion can be achieved through methods such as feature concatenation, feature summation, and element-wise multiplication. Compared to target features, target-related features contain richer feature information, providing a more comprehensive and accurate personalized description of the target user.

[0091] After obtaining personalized features, the computing device 600 can generate content based on these features and content prompt data. Specifically, the computing device 600 can use the personalized features as the direction of content generation, guiding the generated content towards the personalized features of the target user during the content generation process, ensuring that the final generated content matches the user's personalized characteristics.

[0092] When generating personalized target content for target users based on content prompt data and personalized features, the computing device 600 can utilize a neural network model. This neural network model can be a content generation model, meaning the personalized target content is generated using this model. Based on the target user's user type, the computing device 600 can employ a neural network model corresponding to that user type for content generation. Target user types include new users and returning users. New users are those who are generating content for the first time on the AIGC platform, with no historical content generation data available. Returning users are those who have generated content multiple times on the AIGC platform, with historical content generation data available. The neural network model for new users can be called a cold-start content generation model. The neural network model for returning users can be called a non-cold-start content generation model.

[0093] The computing device 600 can determine the user type of a target user based on whether the target user has generated historical content on the AIGC platform. When generating content based on the target user's user type, if the target user is a new user, the computing device 600 can use a cold-start content generation model to generate content for the target user, thereby obtaining personalized target content. If the target user is an existing user, the computing device 600 can use a non-cold-start content generation model to generate content for the target user, thereby obtaining personalized target content.

[0094] The training data for the cold-start content generation model includes sample content prompt data and sample personalized features corresponding to multiple sample users. The sample personalized features are obtained based on the sample profile features of the corresponding sample user and associated sample users. The training objective includes constraining the difference between the predicted generated content output based on the sample content prompt data and sample personalized features and the actual generated content to be less than a second preset difference threshold. The training objective of the cold-start content generation model also includes constraining the difference between the predicted generated content features and the sample historical content features of the corresponding sample users to be less than a third preset difference threshold. The predicted generated content features are obtained based on the predicted generated content. The sample historical content features include at least one of the average historical content prompt data features of the corresponding sample user and associated sample users, and the corresponding average historical adopted content features. The training process of the cold-start content generation model is described below with reference to the accompanying figures:

[0095] Figure 5 A schematic diagram illustrating the data flow during the training process of a preset cold-start content generation network provided according to an embodiment of this specification is shown. Figure 5 As shown, the preset cold start content generation network includes a preset user feature encoding network, a preset diffusion network, and a CLIP model.

[0096] For ease of description, we denote the user features corresponding to the current sample user as "sample user features" and the user features of highly correlated users corresponding to the current sample user as "highly correlated user features." The current sample user may correspond to one or more highly correlated users; here, the highly correlated user corresponding to the current sample user is referred to as at least one highly correlated user. The computing device 600 can also perform a weighted summation of the historical content generation features corresponding to the historical content generation data of the sample user and the historical content generation features corresponding to the historical content generation data of at least one highly correlated user to obtain historical content generation features. The historical content generation data may include historical content prompt data and historical adopted content. The historical content generation features may include historical content prompt data features and historical adopted content features. The computing device 600 can perform a weighted summation of the historical content prompt data features of the current sample user and the historical content prompt data features of at least one highly correlated user to obtain historical content prompt data features. Similarly, the computing device 600 can also perform a weighted summation of the historical adopted content features of the current sample user and the historical adopted content features of at least one highly correlated user to obtain historical adopted content features.

[0097] The preset user feature encoding network can be a Transform network or an MLP. The computing device 600 can input the sample user features and at least one corresponding highly correlated user feature into the preset user feature encoding network, so that the preset user feature encoding network encodes the sample user features and at least one corresponding highly correlated user feature to obtain the predicted personalized features of the sample user.

[0098] The preset diffusion network can be a Unet network. The computing device 600 can input the predicted sample user's personalized features and sample content prompt data into the preset diffusion network, so that the preset diffusion network diffuses in the direction of the predicted sample user's personalized features and generates content based on the sample content prompt data to obtain the predicted generated content.

[0099] The computing device 600 can also input the predicted content into the CLIP model for feature extraction to obtain the predicted content features.

[0100] After obtaining the personalized features of the predicted sample users, the predicted generated content, and the features of the predicted content, the computing device 600 can determine the comprehensive loss based on these three factors, and then converge the preset cold-start content generation network based on the comprehensive loss to obtain the cold-start content generation model. The comprehensive loss may include image prediction loss and personalization loss.

[0101] Image prediction loss can be determined based on the difference between the predicted generated content and the historical adopted content. Historical adopted content refers to the generated content that was ultimately adopted by the current sample user, corresponding to the content suggestion data.

[0102] Personalization loss can be determined based on the difference between predicted content features and historical content generation features. The personalization loss aims to constrain the Euclidean distance between predicted content features and historical content suggestion data features to be as close as possible, and also to constrain the Euclidean distance between predicted content features and historical adopted content features to be as close as possible. Personalization loss ensures that the target content generated by the cold-start content generation model conforms to the content preferences of the target user and related users, reflecting the personalized characteristics of the target user.

[0103] The training process for the non-cold-start content generation model is similar to that for the cold-start model. The difference lies in the personalization loss determination: the historical content suggestion data features are obtained by weighted summation of the historical content generation features corresponding to the historical content generation data of at least one highly relevant sample of users; and the historical adopted content features are obtained by weighted summation of the historical adopted content features corresponding to the historical adopted content of at least one highly relevant sample of users.

[0104] It is important to note that the CLIP model does not update its parameters during training.

[0105] After obtaining the cold start content generation model and the non-cold start content generation model, the computing device 600 can select the matching content generation model to generate content based on the determined user type. For example, when the target user is a new user, the computing device 600 can use the cold start content generation model to generate personalized target content for the target user. When the target user is an existing user, the computing device 600 can use the non-cold start content generation model to generate personalized target content for the target user.

[0106] It should be noted that when the target user is a new user, since the target user has no browsing history on the AIGC platform, the target features do not include text profile features, image profile features, and video profile features. In this case, the computing device 600 can generate personal information profile features based on the personal information registered by the target user on the AIGC platform. That is to say, when the target user is a new user, the target features include personal information profile features. However, when the target user is an existing user, the target features include not only personal information profile features but also at least one of text profile features, image profile features, and video profile features.

[0107] Continue reading Figure 3 After step S160, the method P100 may further include the following step S180.

[0108] S180: Output personalized target content.

[0109] There are several ways to output personalized target content. For example, the computing device 600 can directly return the personalized target content to the terminal device, or it can visualize the personalized target content. Visualizing the personalized target content can be done in various ways, such as displaying the personalized target content on a screen, or issuing prompts about the personalized target content through sound and light, etc.

[0110] In summary, the content generation method P100 and system 001 provided in this specification, for the target user to whom content is to be generated, filter out target related users with similar content preferences and high correlation, and use the target association features of the target related users, the target features of the target user, and the content prompt data input by the target user to jointly create a profile of the target user, generating personalized target content that matches the content prompt data and conforms to the target user's personalized characteristics. In this scheme, by using target association features to create a feature profile of the target user, the personalized characteristics of the target user can be enriched, thereby achieving the effect that the generated target content conforms to the personalized characteristics of the target user when generating target content based on content prompt data.

[0111] It should be noted that the profile data in this embodiment is not profile data for a specific user and does not reflect the personal information of a specific user.

[0112] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0113] This specification, in another aspect, provides a non-transitory storage medium storing at least one set of executable instructions for performing content generation. When the executable instructions are executed by a processor, they instruct the processor to implement the steps of the content generation method P100 described herein. In some possible embodiments, various aspects of this specification can also be implemented as a program product comprising program code. When the program product is run on a computing device 600, the program code causes the computing device 600 to perform the steps of the content generation method P100 described herein. The program product for implementing the above method may employ a portable compact disc read-only memory (CD-ROM) containing program code and may run on the computing device 600. However, the program product of this specification is not limited thereto. In this specification, a readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system. The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. Program code for performing the operations described herein can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on computing device 600, partially on computing device 600, as a standalone software package, partially on computing device 600 and partially on a remote computing device, or entirely on a remote computing device.

[0114] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0115] In summary, after reading this detailed disclosure, those skilled in the art will understand that the foregoing detailed disclosure is presented by way of example only and is not restrictive. Although not explicitly stated herein, those skilled in the art will understand that this specification requires various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be made by this specification and are within the spirit and scope of the exemplary embodiments described herein.

[0116] Furthermore, certain terms in this specification have been used to describe embodiments of this specification. For example, "an embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in connection with that embodiment may be included in at least one embodiment of this specification. Therefore, it is to be emphasized and understood that two or more references to "an embodiment" or "an embodiment" or "alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Moreover, specific features, structures, or characteristics may be suitably combined in one or more embodiments of this specification.

[0117] It should be understood that in the foregoing description of the embodiments in this specification, various features are combined in a single embodiment, drawing, or description for the purpose of simplifying the description and aiding in the understanding of a feature. However, this does not mean that the combination of these features is necessary, and those skilled in the art may readily identify some of the devices as separate embodiments when reading this specification. That is, the embodiments in this specification can also be understood as an integration of multiple secondary embodiments. It is also valid when each secondary embodiment contains fewer than all the features of a single foregoing disclosed embodiment.

[0118] Each patent, patent application, publication of the patent application, and other materials such as articles, books, specifications, publications, documents, articles, etc., cited herein may be incorporated by reference. All contents used for all purposes, except for any history of prosecution documents relating to it, that may be inconsistent with or conflict with this document, or any such history of prosecution documents that may have a limiting effect on the widest extent of the claims, are now or hereafter associated with this document. For example, in the event of any inconsistency or conflict between the description, definition, and / or use of terms associated with any of the included materials and the terms, description, definition, and / or used in connection with this document, the terms used herein shall prevail.

[0119] Finally, it should be understood that the embodiments disclosed herein are illustrative of the principles of the embodiments described in this specification. Other modified embodiments are also within the scope of this specification. Therefore, the embodiments disclosed in this specification are merely examples and not limitations. Those skilled in the art can implement the applications described in this specification using alternative configurations based on the embodiments in this specification. Therefore, the embodiments in this specification are not limited to the embodiments precisely described in the applications.

Claims

1. A content generation method, comprising: Obtain the content prompt data input by the target user and the target associated users corresponding to the target user, wherein the target associated users are associated users who have similar content preferences to the target user; Obtain the target features of the target user and the target association features of the target associated user; Based on the target features and the target association features, the personalized features of the target user are determined; The personalized features are used as the diffusion direction of content generation, and based on the content prompt data, the artificial intelligence content generation model generates content to obtain the personalized target content corresponding to the target user. as well as Output the personalized target content.

2. The method of claim 1, wherein, The target features include at least one of the target user's profile features and historical content features; the target association features include at least one of the target associated user's profile features and historical content features; and the historical content features include historical content prompt data features and their corresponding historical adopted content features.

3. The method according to claim 2, wherein, The acquisition of the target user's input prompt data and the target associated user corresponding to the target user includes: Obtain M candidate associated users corresponding to the target user and their corresponding M candidate associated features, where M is a positive integer; Based on the association relationship between the target user and the M candidate associated users, and the comparison between the target feature and the M candidate associated features, M association degrees between the target user and the M candidate associated users are determined; and Based on the M correlation degrees, the target associated user corresponding to the target user is selected from the M candidate associated users.

4. The method according to claim 3, wherein, The step of obtaining the M candidate associated users corresponding to the target user and their corresponding M candidate associated features includes: Determine multiple similarities between the target feature and user features of multiple users in the user database; The multiple similarities are sorted in descending order to obtain the similarity ranking result; and The top M users with the highest similarity scores in the similarity ranking results are selected as the M candidate associated users.

5. The method according to claim 3, wherein, The association relationship includes at least one of shared device relationship, shared network relationship, and information transmission relationship. The profile features of the target user and the profile features of the M candidate associated users include relationship features, which include at least one of device features, network features, and information transmission features.

6. The method according to claim 5, wherein, The profile features are features obtained based on profile data, which includes at least one of personal information data, recently viewed text, recently viewed images, and recently viewed videos. The profile data also includes at least one of device data, network data, and information transmission data.

7. The method according to claim 3, wherein, The determination of M correlation degrees between the target user and the M candidate associated users based on the association relationship between the target user and the M candidate associated users, and the comparison between the target profile features and the M candidate associated profile features, includes: The target feature and the M candidate association features are input into the associated user prediction model to obtain the M association degrees between the target user and the M candidate associated users; The training objective of the associated user prediction model includes constraining the difference between the predicted association degree and the actual association degree between sample user pairs to be less than a first preset difference threshold.

8. The method according to claim 3, wherein, The step of selecting the target associated user corresponding to the target user from the M candidate associated users based on the M association degrees includes: The M correlation degrees are sorted in descending order; and In the ranking results, the L candidate related users whose relevance is greater than the preset relevance threshold or whose ranking is the top are selected as the target related users, where L is a positive integer less than or equal to M.

9. The method according to claim 1, wherein, The training data for the AI ​​content generation model includes sample content prompt data and sample personalization features corresponding to multiple sample users. The sample personalization features are obtained based on the sample profile features of the corresponding sample user and associated sample users. The training objective includes constraining the difference between the predicted generated content output based on the sample content prompt data and the sample personalization features and the actual generated content to be less than a second preset difference threshold.

10. The method according to claim 9, wherein, The training objective further includes constraining the difference between the predicted generated content features and the corresponding sample user's sample historical content features to be less than a third preset difference threshold, wherein the predicted generated content features are obtained based on the predicted generated content.

11. The method according to claim 10, wherein, The sample historical content features include at least one of the following: the average historical content prompt data features of the corresponding sample user and the associated sample related users, and the corresponding average historical adopted content features.

12. A content generation system, comprising: At least one storage medium storing at least one instruction set for content generation; as well as At least one processor is communicatively connected to the at least one storage medium. When the content generation system is running, the at least one processor reads the at least one instruction set and executes the content generation method according to any one of claims 1-11 according to the instructions of the at least one instruction set.

Images