Method and device for generating personalized content, and medium

Abstract

The present disclosure relates to a method and a device for generating personalized content, and a medium. The method for generating personalized content comprises: obtaining a plurality of sets of fine-tuning parameters of a plurality of personalized models, the plurality of personalized models being generated on the basis of a same base model, and each of the plurality of personalized models being used to generate personalized content satisfying at least one of a plurality of conditions; generating a new set of fine-tuning parameters on the basis of the plurality of sets of fine-tuning parameters; applying the new set of fine-tuning parameters to the base model, so as to generate a new personalized model; and by using the new personalized model, generating new personalized content satisfying the plurality of conditions.

Description

Method, device and medium for generating personalized content

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese patent application number 202410220270.7, filed on February 28, 2024, entitled “Methods, devices and media for generating personalized content,” the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to methods, devices, and media for generating personalized content. Background Art

[0004] Traditional content platforms require pre-production of content and use content recommendation systems to push it to users based on their preferences. In the AIGC era, however, AI models can be used to dynamically generate content that interests users. However, using current technology, when generating content that appeals to multiple users, models must be trained based on their individual interests, consuming significant time and resources. Training models for any combination of users would be prohibitively expensive. Summary of the Invention

[0005] The present disclosure provides a method, device, and medium for generating personalized content.

[0006] According to one aspect of the present disclosure, a method for generating personalized content is provided, comprising: obtaining multiple sets of fine-tuning parameters for multiple personalized models generated based on the same basic model, each of the multiple personalized models being used to generate personalized content that satisfies at least one of multiple conditions; generating a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters; applying the new set of fine-tuning parameters to the basic model to generate a new personalized model; and generating new personalized content that satisfies the multiple conditions using the new personalized model.

[0007] According to yet another aspect of the present disclosure, a device for generating personalized content is provided, including: a processor; and a memory storing program instructions, which, when executed by the processor, enable the processor to perform the method of the present disclosure.

[0008] According to yet another aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which program instructions are stored. When the program instructions are executed by a processor, the processor is caused to perform the method of the present disclosure.

[0009] According to yet another aspect of the present disclosure, a computer program product is provided, comprising program instructions, which, when executed by a processor, cause the processor to perform the method of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] A better understanding of the present disclosure may be obtained when the following detailed description of the embodiments is considered in conjunction with the accompanying drawings. The same or similar reference numerals are used in the various drawings to represent the same or similar parts. The accompanying drawings, together with the following detailed description, are incorporated into and form a part of this specification and are used to illustrate the embodiments of the present disclosure and to explain the principles and advantages of the present disclosure.

[0011] FIG1 shows a schematic diagram of applying fine-tuning parameters to a base model to obtain a personalized model and thus generate personalized content in some embodiments of the present disclosure.

[0012] FIG2 is a schematic diagram showing a process of generating a new personalized model based on two personalized models in some embodiments of the present disclosure.

[0013] FIG3 shows a flowchart of a method 300 for generating a new personalized model based on multiple (two or more) personalized models in some embodiments of the present disclosure.

[0014] FIG4 shows a schematic diagram of a method 400 for training a parameter generator in some embodiments of the present disclosure.

[0015] FIG5 shows a block diagram of an apparatus for generating personalized content in some embodiments of the present disclosure.

[0016] FIG6 shows a block diagram of an electronic device according to some embodiments of the present disclosure.

[0017] FIG7 is a block diagram illustrating an example structure of a computer system that may be employed in some embodiments of the present disclosure. DETAILED DESCRIPTION

[0018] Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Note that, in this specification and the drawings, structural elements that have substantially the same function and structure are denoted with the same reference numerals, and repeated explanation of these structural elements is omitted.

[0019] A personalized model for a single user can generate personalized content that is of interest to that user. However, when generating personalized content that is of interest to multiple users, related technologies require retraining the model based on the content that these users are interested in to obtain a new personalized model. This consumes considerable time and resources. Training the model for any combination of users would be extremely time-consuming and resource-intensive.

[0020] The disclosed embodiments generate a new set of fine-tuning parameters based on the fine-tuning parameters of personalized models for multiple users. This new set of fine-tuning parameters can be directly applied to the base model to obtain a new personalized model, thereby generating personalized content that is of interest to multiple users. This avoids retraining the model for each user combination, saving time and resources.

[0021] Figure 1 illustrates a schematic diagram of applying fine-tuning parameters to a base model to obtain a personalized model and thereby generate personalized content, in some embodiments of the present disclosure. Fine-tuning parameters can be obtained by fine-tuning the base model using a fine-tuning method. These fine-tuning parameters are then applied to the base model to obtain a personalized model. Personalized content is then generated using this personalized model.

[0022] The base model is a large model that has been pre-trained using a large amount of data. This process requires enormous computing resources and a large amount of data. Fine-tuning the base model by applying a small number of fine-tuning parameters can maximize the performance of the large model on new tasks while significantly reducing model training time and computing costs. The base model can be any generative artificial intelligence (AIGC) base model.

[0023] In some embodiments of the present disclosure, one or more fine-tuning methods are used to fine-tune the base model. In some examples, a low-rank adapter (LoRA) is used to fine-tune the base model, with the low-rank matrix serving as a fine-tuning parameter. In some examples, textual inversion (TI) is used to fine-tune the base model, with embedding serving as a fine-tuning parameter. In some examples, prompt engineering (PE) is used to fine-tune the base model, with prompts serving as fine-tuning parameters.

[0024] However, fine-tuning the base model also requires training with a small amount of data, which would still require considerable training time and computational cost if retrained for each new task.

[0025] Figure 2 illustrates a schematic diagram of a process 200 for generating a new personalized model based on two personalized models, in some embodiments of the present disclosure. Personalized model A 202 is generated by applying a set of fine-tuning parameters A 204 to a base model 201, and is used to generate personalized content that satisfies condition A. Personalized model B 203 is generated by applying a set of fine-tuning parameters B 205 to the base model 201, and is used to generate personalized content that satisfies condition B. Fine-tuning parameters A 204 and B 205 are input to a parameter generator 206 to generate a new set of fine-tuning parameters 207. The new fine-tuning parameters 207 are applied to the base model 201 to generate a new personalized model 208. The new personalized model 208 is used to generate new personalized content 209 that satisfies both conditions A and B.

[0026] In some embodiments of the present disclosure, fine-tuning parameters A 204 and B 205 are obtained by fine-tuning base model 201 using the same fine-tuning method and different training data. One or more fine-tuning methods are used to fine-tune base model 201. When a low-rank adapter is used to fine-tune base model 201, a low-rank matrix serves as fine-tuning parameters A 204 and B 205. When text reversal is used to fine-tune base model 201, embedding serves as fine-tuning parameters A 204 and B 205. When prompt word engineering is used to fine-tune base model 201, prompt words serve as fine-tuning parameters A 204 and B 205.

[0027] Parameter generator 206 receives input fine-tuning parameters A 204 and fine-tuning parameters B 205, and outputs new fine-tuning parameters 207 generated based on fine-tuning parameters A 204 and fine-tuning parameters B 205. New fine-tuning parameters 207 are of the same type as fine-tuning parameters A 204 and fine-tuning parameters B 205. In some embodiments of the present disclosure, parameter generator 206 is a pre-trained model, including at least one of a linear model, a random forest, an xgboost, and a neural network.

[0028] In some embodiments of the present disclosure, personalization model A 202 is used to generate personalized content of interest to user A, and personalization model B 203 is used to generate personalized content of interest to user B. Therefore, the personalized content 209 generated by process 200 is personalized content of interest to both user A and user B.

[0029] In some embodiments of the present disclosure, personalization model A 202 is used to generate personalized content with feature A, and personalization model B 203 is used to generate personalized content with feature B. Therefore, personalized content 209 generated by process 200 is personalized content with both feature A and feature B.

[0030] Next, two examples are used to illustrate the personalized content generation method according to the embodiment of the present disclosure.

[0031] Example 1: Generate Van Gogh-style paper-cutting images using a personalized model that generates personalized images in the paper-cutting style and a personalized model that generates personalized images in the Van Gogh style. In this example, user A and user B respectively use the basic model Stable Diffusion (whose model parameter is W) to fine-tune the personalized AIGC model in the paper-cutting style and Van Gogh style, respectively, using LoRA as the fine-tuning method. The fine-tuning parameters of LoRA are two low-rank matrices. The fine-tuning parameter of user A is L A ,U A , the fine-tuning parameter of user B is L B ,U B When using LoRA, update the model parameters according to the following formula: W′ A =W+L A U A ,W′ B =W+L B U B

[0032] Set the fine-tuning parameter of user A to L A ,U A and user B’s fine-tuning parameter is L B ,U B Input the parameter generator to generate new fine-tuning parameters. For example, the parameter generator generates new fine-tuning parameters according to the following formula: X = θ1L A U A +θ2L A U B +θ3L B U A +θ4L B U B

[0033] Among them, θ1, θ2, θ3, θ4 are the neural network based on L A ,U A ,L B ,U B Generate non-negative coefficients. Use X to update the model parameters according to the following formula to generate a new personalized model for generating Van Gogh-style paper-cut images: W′=W+X

[0034] Although Stable Diffusion is used as the base model and LoRA is used as the fine-tuning method in this example, those skilled in the art will appreciate, based on this disclosure, that other base models and fine-tuning methods can also be used in this example. Furthermore, the parameter generator in this example can also adopt other models.

[0035] Example 2: Generate text of philosopher jokes using a personalized model for generating comedian text and a personalized model for generating philosopher text. This example uses ChatGLM2 as the base model and uses text reversal as the fine-tuning method. The fine-tuning parameters for the personalized models for the comedian and philosopher are E and E respectively. A ,E B (Embedding, which is a vector of fixed dimension.) When using ChatGLM2 to generate personalized text, the embedding is input as the first token into the ChatGLM2 decoder, and decoded to generate the subsequent text.

[0036] Based on the fine-tuning parameter E A ,E B Generate new fine-tuning parameters according to the following formula: E X =f(E A ,E B )

[0037] Among them, E X With E A ,E B is a vector of the same dimension, and f is a Transformer model that can input any number (at least two) of embeddings and generate new embeddings. X As the first flag input to the ChatGLM2 decoder, it can decode and generate text of philosophers telling jokes.

[0038] Although ChatGLM2 is used as the base model and text reversal is used as the fine-tuning method in this example, those skilled in the art will appreciate that other base models and fine-tuning methods can also be used in this example based on the present disclosure. In addition, the parameter generator in this example can also adopt other models.

[0039] In addition to generating a new personalized model based on two personalized models, some embodiments of the present disclosure can generate a new personalized model based on more personalized models.

[0040] FIG3 shows a flowchart of a method 300 for generating a new personalized model based on multiple (two or more) personalized models in some embodiments of the present disclosure.

[0041] In step S302, multiple sets of fine-tuning parameters for multiple personalized models generated based on the same base model are obtained. Each of the multiple personalized models is used to generate personalized content that satisfies at least one of a plurality of conditions. In some embodiments of the present disclosure, each set of fine-tuning parameters in the multiple sets of fine-tuning parameters is obtained by fine-tuning the base model using the same fine-tuning method and different training data. In some embodiments of the present disclosure, the fine-tuning method includes at least one of a low-rank adapter, text reversal, and prompt word engineering. In some embodiments of the present disclosure, each set of fine-tuning parameters in the multiple sets of fine-tuning parameters includes at least one of a low-rank matrix, an embedding, and a prompt word.

[0042] In step S304, a new set of fine-tuning parameters is generated based on the multiple sets of fine-tuning parameters. In some embodiments of the present disclosure, the new set of fine-tuning parameters is of the same type as the multiple sets of fine-tuning parameters. In some embodiments of the present disclosure, generating a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters includes: inputting the multiple sets of fine-tuning parameters into a trained parameter generator to generate the new set of fine-tuning parameters. In some embodiments of the present disclosure, the parameter generator includes at least one of a linear model, a random forest, an xgboost, and a neural network. The model parameters of the parameter generator can be pre-trained and fixed during use.

[0043] In step S306, the new set of fine-tuning parameters is applied to the base model to generate a new personalized model. In step S308, new personalized content that meets the multiple conditions is generated using the new personalized model.

[0044] The likelihood that the new personalized model will generate new personalized content that meets the multiple conditions is assessed by comparing the new personalized content generated by the new personalized model with the personalized content generated by each of the multiple personalized models and calculating similarity. Similarity can be measured using likelihood or KL divergence. A higher likelihood or a lower KL divergence indicates a higher similarity. A higher similarity indicates a higher likelihood that the new personalized model will generate new personalized content that meets the multiple conditions.

[0045] In some embodiments of the present disclosure, at least one of the multiple conditions includes being of interest to at least one of the multiple users. The new personalized content is personalized content that is of common interest to the multiple users. Therefore, based on the personalized models of the multiple users, content of common interest to the multiple users can be generated without retraining the models. Since retraining the models is unnecessary, content of common interest to multiple users can be generated conveniently.

[0046] In some embodiments of the present disclosure, at least one of the multiple conditions includes at least one personalized feature from a plurality of personalized features. The new personalized content has the plurality of personalized features. Therefore, content with the plurality of personalized features can be generated based on multiple personalized models without requiring retraining the models. Since retraining the models is not required, content with the plurality of personalized features can be generated efficiently and at a low cost.

[0047] FIG4 shows a schematic diagram of a method 400 for training a parameter generator in some embodiments of the present disclosure.

[0048] In step S402 , multiple groups of specific fine-tuning parameters of multiple specific personalized models generated based on the basic model are obtained, where each specific personalized model in the multiple specific personalized models is used to generate personalized content that meets at least one specific condition among multiple specific conditions.

[0049] In step S404, the plurality of sets of specific fine-tuning parameters are input into a parameter generator to generate a new set of specific fine-tuning parameters. In step S406, the new set of specific fine-tuning parameters is applied to the base model to generate a new specific personalized model.

[0050] In step S408, target personalized content for training that meets the multiple specific conditions is obtained. In some embodiments of the present disclosure, at least one of the multiple specific conditions includes being of interest to at least one of the multiple users. The target personalized content is personalized content that is of interest to all of the multiple users. In some embodiments of the present disclosure, at least one of the multiple specific conditions includes having at least one specific personalized feature from multiple specific personalized characteristics. The target personalized content has the multiple specific personalized characteristics.

[0051] In step S410 , the base model is fine-tuned using the target personalized content to generate a reference model.

[0052] In step S412, the personalized content generated by the new specific personalized model is compared with the personalized content generated by the reference model and similarity is calculated. Similarity can be measured using likelihood or KL divergence. A higher likelihood or lower KL divergence indicates higher similarity.

[0053] In step S414, it is determined whether the similarity meets the requirements. If the similarity does not meet the requirements, the process proceeds to step S416 to adjust the model parameters of the parameter generator, and then returns to step S404. If the similarity meets the requirements, the process proceeds to step S418 to end the training of the parameter generator.

[0054] FIG5 shows a block diagram of an apparatus for generating personalized content in some embodiments of the present disclosure.

[0055] The parameter extractor 501 is configured to receive multiple personalized models generated based on the same base model and extract multiple sets of fine-tuning parameters. Each of the multiple personalized models is used to generate personalized content that satisfies at least one of a plurality of conditions. In some embodiments of the present disclosure, each set of fine-tuning parameters in the multiple sets of fine-tuning parameters is obtained by fine-tuning the base model using the same fine-tuning method and different training data. In some embodiments of the present disclosure, the fine-tuning method includes at least one of a low-rank adapter, text reversal, and prompt word engineering. In some embodiments of the present disclosure, each set of fine-tuning parameters in the multiple sets of fine-tuning parameters includes at least one of a low-rank matrix, an embedding, and a prompt word.

[0056] Parameter generator 502 is configured to generate a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters. In some embodiments of the present disclosure, the new set of fine-tuning parameters is of the same type as the multiple sets of fine-tuning parameters. In some embodiments of the present disclosure, parameter generator 502 is pre-trained. In some embodiments of the present disclosure, parameter generator 502 includes at least one of a linear model, a random forest, an xgboost, and a neural network.

[0057] The model generator 503 is configured to apply the new set of fine-tuning parameters to the base model to generate a new personalized model. The content generator 504 is configured to generate new personalized content that meets the multiple conditions using the new personalized model.

[0058] In some embodiments of the present disclosure, at least one of the multiple conditions includes being of interest to at least one of the multiple users. The new personalized content is personalized content that all of the multiple users are interested in. In some embodiments of the present disclosure, at least one of the multiple conditions includes having at least one of multiple personalized features. The new personalized content has the multiple personalized features.

[0059] One or more embodiments of the present disclosure can provide one or more of the following technical effects: improving the usability of the basic model, providing personalized AIGC services that meet multiple conditions, improving the stickiness of AIGC service users, building an AIGC sharing community ecosystem, and better utilizing existing resource libraries.

[0060] Some embodiments of the present disclosure also provide an electronic device. Figure 6 shows a block diagram of an electronic device 600 according to some embodiments of the present disclosure. As shown in Figure 6, the electronic device 6 of this embodiment includes: a memory 601 and a processor 602 coupled to the memory 601. It should be noted that the components of the electronic device 600 shown in Figure 6 are merely exemplary and non-restrictive. Depending on actual application requirements, the electronic device 600 may also have other components. The processor 602 can control other components in the electronic device 600 to perform desired functions.

[0061] In some embodiments, the memory 601 is configured to store one or more computer-readable instructions. When the processor 602 is configured to execute the computer-readable instructions, the computer-readable instructions are executed by the processor 602 to implement the method according to any of the above embodiments, such as the method described with respect to FIG. 3 or 4 . The specific implementation and related explanations of each step of the method can be found in the above embodiments, and any repetitive details are omitted here.

[0062] For example, the processor 602 and the memory 601 may communicate with each other directly or indirectly. For example, the processor 602 and the memory 601 may communicate with each other via a network. The network may include a wireless network, a wired network, and / or any combination of wireless networks and wired networks. The processor 602 and the memory 601 may also communicate with each other via a system bus, which is not limited in this disclosure.

[0063] For example, the processor 602 can be embodied as various appropriate processors, processing devices, etc., such as a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The central processing unit (CPU) can be an X86 or ARM architecture, etc. For example, the memory 601 can include any combination of various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The memory 601 can include, for example, a system memory, which stores, for example, an operating system, an application, a boot loader (Boot Loader), a database, and other programs. Various applications and various data can also be stored in the storage medium.

[0064] In addition, according to some embodiments of the present disclosure, when various operations / processes according to the present disclosure are implemented through software and / or firmware, the programs constituting the software can be installed from a storage medium or a network to a computer system having a dedicated hardware structure, such as the computer system 700 shown in FIG. 7 . When the various programs are installed, the computer system can perform various functions, including functions such as the parameter extractor 501, parameter generator 502, model generator 503, and content generator 504 described with respect to FIG. FIG. 7 shows a block diagram of an example structure of a computer system 700 that can be used in some embodiments of the present disclosure.

[0065] In Figure 7, a central processing unit (CPU) 701 performs various processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage section 708 to a random access memory (RAM) 703. In the RAM 703, data required when the CPU 701 performs various processes, etc., is also stored as needed. The central processing unit is merely exemplary and may also be other types of processors, such as the various processors described above. The ROM 702, RAM 703, and storage section 708 may be various forms of computer-readable storage media, as described below. It should be noted that although ROM 702, RAM 703, and storage device 708 are shown separately in Figure 7, one or more of them may be combined or located in the same or different memory or storage modules.

[0066] The CPU 701, the ROM 702, and the RAM 703 are connected to one another via a bus 704. An input / output interface 707 is also connected to the bus 704.

[0067] The following components are connected to the input / output interface 705: an input portion 706, such as a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; an output portion 707, including a display, such as a cathode ray tube (CRT), liquid crystal display (LCD), speaker, vibrator, etc.; a storage portion 708, including a hard disk, magnetic tape, etc.; and a communication portion 709, including a network interface card, such as a LAN card, modem, etc. The communication portion 709 allows communication processing to be performed via a network, such as the Internet. It will be readily understood that although FIG7 shows that the various devices or modules in the computer system 700 communicate via the bus 704, they may also communicate via a network or other means, where the network may include a wireless network, a wired network, and / or any combination of wireless and wired networks.

[0068] A drive 710 is also connected to the input / output interface 705 as needed. A removable medium 711 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is mounted on the drive 710 as needed so that a computer program read therefrom is installed in the storage section 708 as needed.

[0069] In the case of realizing the above-described series of processing by software, a program constituting the software can be installed from a network such as the Internet or a storage medium such as the removable medium 711 .

[0070] According to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network via the communication device 709, or installed from the storage device 708, or installed from the ROM 702. When the computer program is executed by the CPU 701, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0071] It should be noted that in the context of the present disclosure, a computer-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. A computer-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 computer-readable storage media may include, but are not limited to, an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0072] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0073] In some embodiments, a computer program is further provided, comprising: instructions, which, when executed by a processor, cause the processor to perform the method of any embodiment. For example, the instructions may be embodied as computer program codes.

[0074] In embodiments of the present disclosure, computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or combinations thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0075] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the prescribed logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the prescribed function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0076] The modules, components, or units described in the embodiments of the present disclosure may be implemented in software or hardware. The names of the modules, components, or units do not necessarily limit the modules, components, or units themselves.

[0077] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, and without limitation, exemplary hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0078] The above descriptions are merely some embodiments of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the present disclosure.

[0079] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the present disclosure may be practiced without these specific details. In other cases, well-known methods, structures, and techniques are not presented in detail in order not to obscure the understanding of the description.

[0080] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details have been included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0081] Although some specific embodiments of the present disclosure have been described in detail by way of examples, those skilled in the art will appreciate that the above examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Those skilled in the art will appreciate that modifications may be made to the above embodiments without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

[0082] The present disclosure also includes the following embodiments.

[0083] 1. A method for generating personalized content, comprising:

[0084] Acquire multiple sets of fine-tuning parameters for multiple personalized models generated based on the same basic model, each of the multiple personalized models being used to generate personalized content that satisfies at least one of multiple conditions;

[0085] generating a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters;

[0086] Applying the new set of fine-tuning parameters to the base model to generate a new personalized model;

[0087] The new personalized model is used to generate new personalized content that meets the multiple conditions.

[0088] 2. The method according to item 1, wherein

[0089] At least one of the plurality of conditions includes at least one user among the plurality of users being interested;

[0090] The new personalized content is personalized content that is of interest to the multiple users.

[0091] 3. The method according to item 1, wherein

[0092] At least one of the plurality of conditions includes at least one personalized feature among a plurality of personalized features;

[0093] The new personalized content has the multiple personalized features.

[0094] 4. The method according to item 1, wherein

[0095] Each set of fine-tuning parameters in the multiple sets of fine-tuning parameters is obtained by fine-tuning the basic model using the same fine-tuning method and different training data.

[0096] 5. The method according to item 4, wherein

[0097] The fine-tuning method includes at least one of a low-rank adapter, text reversal, and prompt word engineering.

[0098] 6. The method according to item 1, wherein

[0099] Each set of fine-tuning parameters in the multiple sets of fine-tuning parameters includes at least one of a low-rank matrix, an embedding, and a prompt word.

[0100] 7. The method of clause 1, wherein the new set of fine-tuning parameters is of the same type as the multiple sets of fine-tuning parameters.

[0101] 8. The method of clause 1, wherein generating a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters comprises:

[0102] The multiple sets of fine-tuning parameters are input into a trained parameter generator to generate the new set of fine-tuning parameters.

[0103] 9. The method of item 8, wherein the parameter generator comprises at least one of a linear model, a random forest, an xgboost, and a neural network.

[0104] 10. The method of clause 8, wherein the parameter generator is trained by:

[0105] a) obtaining multiple sets of specific fine-tuning parameters for multiple specific personalized models generated based on the base model, each of the multiple specific personalized models being used to generate personalized content that satisfies at least one specific condition among multiple specific conditions;

[0106] b) inputting the multiple sets of specific fine-tuning parameters into the parameter generator to generate a new set of specific fine-tuning parameters;

[0107] c) applying the new set of specific fine-tuning parameters to the base model to generate a new specific personalized model;

[0108] d) obtaining target personalized content for training that meets the multiple specific conditions;

[0109] e) training the base model using the target personalized content to generate a reference model;

[0110] f) comparing the personalized content generated by the new specific personalized model with the personalized content generated by the reference model and calculating the similarity;

[0111] g) If the similarity does not meet the requirement, adjust the model parameters of the parameter generator and return to step b).

[0112] 11. A device for generating personalized content, comprising:

[0113] processor; and

[0114] A memory having program instructions stored thereon, which, when executed by a processor, cause the processor to implement the method as described in any one of items 1 to 11.

[0115] 12. A non-transitory computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, cause the processor to implement the method according to any one of items 1 to 11.

[0116] 13. A computer program product comprising program instructions which, when executed by a processor, cause the processor to implement the method according to any one of clauses 1 to 11.

Claims

1. A method for generating personalized content, comprising: Acquire multiple sets of fine-tuning parameters for multiple personalized models generated based on the same basic model, each of the multiple personalized models being used to generate personalized content that satisfies at least one of multiple conditions; generating a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters; Applying the new set of fine-tuning parameters to the base model to generate a new personalized model; New personalized content that meets the multiple conditions is generated using the new personalized model.

2. The method of claim 1, wherein At least one of the plurality of conditions includes at least one user among the plurality of users being interested; The new personalized content is personalized content that is of interest to the multiple users.

3. The method of claim 1, wherein At least one of the plurality of conditions includes at least one personalized feature among a plurality of personalized features; The new personalized content has the multiple personalized features.

4. The method of claim 1, wherein Each set of fine-tuning parameters in the multiple sets of fine-tuning parameters is obtained by fine-tuning the basic model using the same fine-tuning method and different training data.

5. The method of claim 4, wherein The fine-tuning method includes at least one of a low-rank adapter, text reversal, and prompt word engineering.

6. The method of claim 1, wherein Each set of fine-tuning parameters in the multiple sets of fine-tuning parameters includes at least one of a low-rank matrix, an embedding, and a prompt word. The method of claim 1 , wherein the new set of fine-tuning parameters is of the same type as the multiple sets of fine-tuning parameters.

8. The method of claim 1 , wherein generating a new set of fine-tuning parameters based on the multiple sets of fine-tuning parameters comprises: The multiple sets of fine-tuning parameters are input into a trained parameter generator to generate the new set of fine-tuning parameters.

9. The method of claim 8, wherein the parameter generator comprises at least one of a linear model, a random forest, an xgboost, and a neural network.

10. The method of claim 8, wherein the parameter generator is trained by: a) obtaining multiple sets of specific fine-tuning parameters for multiple specific personalized models generated based on the base model, each of the multiple specific personalized models being used to generate personalized content that satisfies at least one specific condition among multiple specific conditions; b) inputting the multiple sets of specific fine-tuning parameters into the parameter generator to generate a new set of specific fine-tuning parameters; c) applying the new set of specific fine-tuning parameters to the base model to generate a new specific personalized model; d) obtaining target personalized content for training that meets the multiple specific conditions; e) training the base model using the target personalized content to generate a reference model; f) comparing the personalized content generated by the new specific personalized model with the personalized content generated by the reference model and calculating the similarity; g) If the similarity does not meet the requirement, adjust the model parameters of the parameter generator and return to step b).

11. A device for generating personalized content, comprising: processor; as well as A memory having program instructions stored thereon, wherein when the program instructions are executed by a processor, the processor is caused to implement the method according to any one of claims 1 to 11. 12 . A non-transitory computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, cause the processor to implement the method according to claim 1 .

13. A computer program product comprising program instructions, which, when executed by a processor, cause the processor to implement the method according to any one of claims 1 to 11.

Images