Data processing method and related equipment

Abstract

The embodiment of the invention provides a data processing method and related equipment, and the method comprises the steps: obtaining a probability matrix obtained by a data generation model in a multimedia data generation process, and enabling the probability matrix to comprise a probability set corresponding to each multimedia object in to-be-generated multimedia data, the probability set comprises probabilities of N candidate objects corresponding to the multimedia object; n is a positive integer; updating a probability matrix according to the obtained disturbance key to obtain a probability update matrix, the probability update matrix comprising a probability update set of each multimedia object, the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object; and generating multimedia data according to the probability updating matrix and the N candidate objects of each multimedia object. According to the embodiment of the invention, copyright protection can be better realized for the generated multimedia data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet, in particular to a data processing method and related equipment. BACKGROUND

[0002] With the development of Internet technology, AI (Artificial Intelligence) technology emerges as the times require. At present, the generative large model in the AI scene can generate text, pictures, videos, sounds and other forms of content. The generated content can be referred to as multimedia data. At present, the copyright protection of multimedia data is mainly achieved by directly adding a watermark in the generated multimedia data. However, this method may modify the multimedia data, affecting the quality of the multimedia data. SUMMARY

[0003] The embodiments of the present application provide a data processing method and related equipment, which can better realize copyright protection of generated multimedia data.

[0004] In one aspect, the embodiments of the present application provide a data processing method, which comprises:

[0005] obtaining a probability matrix obtained by a data generation model in the process of generating multimedia data, the probability matrix comprising: a probability set corresponding to each multimedia object in the to-be-generated multimedia data, the probability set comprising probabilities of N candidate objects of the corresponding multimedia object; N is a positive integer;

[0006] updating the probability matrix according to the obtained perturbation key to obtain a probability update matrix, the probability update matrix comprising a probability update set of each multimedia object, the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object being the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object;

[0007] generating the multimedia data according to the probability update matrix and the N candidate objects of each multimedia object.

[0008] In one aspect, the embodiments of the present application provide a data processing method, which comprises:

[0009] obtain multimedia data and a probability update matrix associated with the multimedia data; the multimedia data is determined according to the probability update matrix and N candidate objects of each multimedia object, the probability update matrix is obtained by updating a probability matrix according to an obtained disturbance key, the probability matrix is obtained by a data generation model in a process of generating the multimedia data; N is a positive integer; the probability update matrix includes a probability update set of each multimedia object, and a candidate object corresponding to a maximum update probability in the probability update set of each multimedia object is the same as a candidate object corresponding to a maximum probability in a probability set of the corresponding multimedia object;

[0010] verify the multimedia data according to the obtained verification disturbance key and the probability update matrix, and obtain a verification result of the multimedia data.

[0011] In an aspect, an embodiment of the present application provides a data processing apparatus, which comprises:

[0012] an obtaining unit, configured to obtain a probability matrix obtained by a data generation model in a process of generating multimedia data, the probability matrix comprising: a probability set corresponding to each multimedia object in the multimedia data to be generated, the probability set comprising probabilities of N candidate objects of the corresponding multimedia object; N is a positive integer;

[0013] a processing unit, configured to update the probability matrix according to an obtained disturbance key, and obtain a probability update matrix, the probability update matrix comprising a probability update set of each multimedia object, and a candidate object corresponding to a maximum update probability in the probability update set of each multimedia object being the same as a candidate object corresponding to a maximum probability in a probability set of the corresponding multimedia object;

[0014] the processing unit is further configured to generate the multimedia data according to the probability update matrix and the N candidate objects of each multimedia object.

[0015] When the processing unit updates the probability matrix according to the obtained disturbance key and obtains the probability update matrix, the processing unit can be specifically configured to:

[0016] perform hash processing on the obtained disturbance key, and obtain a hash corresponding to the disturbance key;

[0017] generate probability disturbance information according to the hash corresponding to the disturbance key;

[0018] update the probability matrix according to the probability disturbance information, and obtain the probability update matrix.

[0019] The probability disturbance information comprises M disturbance values associated with each multimedia object, and M is a positive integer; when the processing unit updates the probability matrix according to the probability disturbance information and obtains the probability update matrix, the processing unit can be specifically configured to:

[0020] determining, from the M disturbance values associated with the target multimedia object, a disturbance value corresponding to each probability in the probability set corresponding to the target multimedia object; the target multimedia object being any multimedia object in the multimedia data to be generated;

[0021] multiplying each disturbance value associated with the target multimedia object with the corresponding probability in the probability matrix to obtain a probability update set of the target multimedia object.

[0022] In the process of determining, from the M disturbance values associated with the target multimedia object, a disturbance value corresponding to each probability in the probability set corresponding to the target multimedia object, the processing unit can be specifically configured to:

[0023] obtaining a target disturbance value from the M disturbance values associated with the target multimedia object;

[0024] performing a marking process on the target disturbance value;

[0025] corresponding the marked target disturbance value with the maximum probability in the probability set corresponding to the target multimedia object, and corresponding other disturbance values associated with the target multimedia object with other probabilities in the probability set corresponding to the target multimedia object;

[0026] The other disturbance values refer to disturbance values other than the target disturbance value in the N disturbance values associated with the target multimedia object; and the other probabilities refer to probabilities other than the maximum probability in the probability set corresponding to the target multimedia object.

[0027] In the process of generating the multimedia data according to the probability update matrix and the N candidate objects of each multimedia object, the processing unit can be specifically configured to:

[0028] performing probability sum calculation on the update probabilities in the probability update set corresponding to each multimedia object to obtain a probability sum corresponding to each multimedia object;

[0029] performing normalization processing on the update probabilities in the probability update set corresponding to the corresponding multimedia object according to the probability sum corresponding to each multimedia object to obtain a normalized probability of each candidate object;

[0030] determining a candidate object from the N candidate objects of each multimedia object as the corresponding multimedia object according to the normalized probability of the N candidate objects of each multimedia object;

[0031] generating the multimedia data according to the corresponding multimedia object.

[0032] In one aspect, the embodiments of the present application provide a data processing apparatus, which comprises:

[0033] The acquisition unit is configured to acquire multimedia data and a probability update matrix associated with the multimedia data, wherein the multimedia data is determined according to the probability update matrix and N candidate objects of each multimedia object, the probability update matrix is obtained by updating a probability matrix according to an acquired disturbance key, and the probability matrix is obtained by a data generation model during generation of the multimedia data; N is a positive integer; and the probability update matrix includes a probability update set of each multimedia object, and a candidate object corresponding to a maximum update probability in the probability update set of each multimedia object is the same as a candidate object corresponding to a maximum probability in a probability set of the corresponding multimedia object.

[0034] The processing unit is configured to verify the multimedia data according to the acquired verification disturbance key and the probability update matrix, and obtain a verification result of the multimedia data.

[0035] In a possible implementation, when the processing unit verifies the multimedia data according to the acquired verification disturbance key and the probability update matrix, and obtains the verification result of the multimedia data, the processing unit can be specifically configured to:

[0036] generate verification probability disturbance information according to the acquired verification disturbance key;

[0037] verify the multimedia data according to the verification probability disturbance information and the probability update matrix, and obtain the verification result of the multimedia data.

[0038] In a possible implementation, when the processing unit verifies the multimedia data according to the verification probability disturbance information and the probability update matrix, and obtains the verification result of the multimedia data, the processing unit can be specifically configured to:

[0039] restore the probability update matrix according to the verification probability disturbance information, and obtain an original probability matrix, wherein the original probability matrix includes an original probability set corresponding to each multimedia object in the multimedia data to be generated, and the original probability set includes original probabilities of N candidate objects of the corresponding multimedia object.

[0040] generate original multimedia data according to the original probability matrix and the N candidate objects of each multimedia object.

[0041] If the original multimedia data matches the multimedia data, the processing unit obtains the verification result that the multimedia data is verified successfully.

[0042] In a possible implementation, the verification probability disturbance information includes N verification disturbance values associated with each multimedia object, and the probability disturbance information includes N disturbance values associated with each multimedia object; and the processing unit can be further configured to:

[0043] determine a target verification disturbance value corresponding to each multimedia object from the N verification disturbance values associated with each multimedia object, wherein the target verification disturbance value is a marked verification disturbance value.

[0044] If the target verification perturbation value corresponding to each multimedia object is the same as the target perturbation value corresponding to the corresponding multimedia object in the probability perturbation information, a step of performing restoration processing on the probability update matrix according to the obtained perturbation key to obtain the original probability matrix is performed. The target perturbation value refers to the marked perturbation value.

[0045] In one aspect, an embodiment of the present application provides a computer device, which comprises:

[0046] a processor adapted to execute a computer program;

[0047] a computer readable storage medium, the computer readable storage medium storing a computer program, the computer program being executed by the processor to implement the above data processing method.

[0048] In one aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium storing a computer program, the computer program being loaded and executed by the processor to implement the above data processing method.

[0049] In one aspect, an embodiment of the present application provides a computer program product, the computer program product comprising a computer program or computer instructions, the computer program or computer instructions being executed by the processor to implement the above data processing method.

[0050] In this embodiment, a probability matrix is ​​obtained by the data generation model during the generation of multimedia data. The probability matrix includes: a probability set corresponding to each multimedia object in the multimedia data to be generated, where each probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer; the probability matrix is ​​updated using the obtained perturbation key to obtain a probability update matrix; this probability update matrix includes the probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object; multimedia data is generated based on the probability update matrix and the N candidate objects for each multimedia object. It is evident that during the update of the probability matrix using the perturbation key, it can effectively ensure that the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. In other words, this embodiment can embed a perturbation key while ensuring the output remains unchanged, achieving lossless encryption of the generated multimedia data and guaranteeing the quality of the generated multimedia data to a certain extent. Furthermore, during the multimedia data generation process, the probability matrix obtained during the multimedia data generation process is updated by acquiring the perturbation key, thereby embedding the perturbation key into the probability update matrix and better protecting the copyright of the generated multimedia data through the perturbation key. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A schematic diagram illustrating the process of generating text using a data generation model provided in this application embodiment;

[0053] Figure 2 An architecture diagram of a data processing system provided in this application embodiment;

[0054] Figure 3 This application provides a schematic diagram of a probability matrix update.

[0055] Figure 4 A flowchart illustrating a data processing method provided in an embodiment of this application;

[0056] Figure 5 A schematic diagram of a probability matrix provided for an embodiment of this application;

[0057] Figure 6 A flowchart illustrating another data processing method provided in an embodiment of this application;

[0058] Figure 7 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application;

[0059] Figure 8 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application;

[0060] Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0061] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0062] First, the technical terms involved in the embodiments of this application will be explained.

[0063] I. Data Generation Model

[0064] Generative large-scale language models (GLLMs) are a type of large-scale neural network trained using deep learning techniques to generate various forms of content. Besides generating natural language text, GLLMs also support the generation of images, videos, audio, and other media content. GLLMs are pre-trained on massive amounts of data to learn language structure, semantic relationships, and the characteristics of various media, enabling them to generate coherent and meaningful content based on input. Typical GLLM models include the LLama series, GPT series, DALL-E (for image generation), and Whisper (for audio processing). They achieve the generation and processing of various media content through two stages: pre-training and fine-tuning, and are widely used in fields such as automatic text generation, dialogue systems, machine translation, image generation, and video synthesis.

[0065] The data generation model can include the Transformer model, a deep learning model based on the attention mechanism. Through self-attention and multi-head attention, it achieves efficient parallel computation and long-distance dependency modeling, and is widely used in natural language processing tasks. The attention mechanism is an important technique in deep learning used to improve model performance, especially in natural language processing (NLP) and computer vision (CV). The core idea of ​​the attention mechanism is to calculate the importance of different parts of the input data and assign corresponding weights, enabling the model to focus on the parts most important to the current task.

[0066] It should be understood that the multimedia data involved in the embodiments of this application may include, but is not limited to, text, video, audio, images, etc. For ease of understanding, the following explanation will use text generation by a data generation model as an example. Please refer to [link to relevant documentation]. Figure 1 This is a schematic diagram illustrating a data generation model for generating text, as provided in an embodiment of this application. The data generation model generates text through the following steps: Step 1: The user inputs a prompt message for text generation into the data generation model via a terminal device, such as "I like you". Step 2: The transformation model (e.g., a transformer model) in the data generation model performs matrix operations such as embedding and attention based on the prompt message to obtain a probability matrix. The probability matrix can include a probability set corresponding to each word in the text to be generated, and the probability set corresponding to each word includes the probabilities of N candidate words (i.e., tokens) for that word. The probability set corresponding to each word can also be understood as a one-dimensional vector matrix. Figure 1 As shown, a probability matrix can be obtained through the transformation model. This probability matrix includes a probability set 11 corresponding to a word in the text to be generated. Step 3: Determine the maximum probability from each probability set included in the probability matrix, and determine the candidate word corresponding to the maximum probability, for example... Figure 1 In the probability set 11, the maximum probability is 0.2082, so the candidate word corresponding to the maximum probability is "cat". Step 4: Generate text based on the candidate word corresponding to the maximum probability, which is "I like your cat".

[0067] II. Probability Perturbation

[0068] Probabilistic perturbation refers to a method of adjusting a probability matrix by introducing slight random interference. In this embodiment, the probability matrix obtained by the data generation model during the generation of multimedia data can be probabilistically perturbed without changing the generated content (i.e., without changing the generated multimedia data). This process involves generating probability perturbation information using a perturbation key and adding slight perturbations to the probabilities in the probability matrix using this information, thereby embedding encrypted information (also known as watermark information) while keeping the output content unchanged. This method increases the complexity and security of encryption while ensuring the authenticity and integrity of the generated content.

[0069] III. Normalization

[0070] Normalization refers to the ability to control a set of data within a certain range. In this embodiment, normalization refers to the process of adjusting a set of probabilities after perturbation so that their sum equals 1. Specifically, by calculating the sum of all probabilities after perturbation, and then dividing the perturbed probability of each probability by this sum, it is ensured that the perturbed probabilities are still a valid probability matrix. This process guarantees that the probability matrix output by the model still has mathematical correctness and logical consistency after perturbation. It should be noted that in this embodiment, the perturbed probabilities can be called updated probabilities.

[0071] This application provides a data processing scheme that, while ensuring the generated content remains unchanged, uses a probabilistic perturbation strategy to perturb the probability matrix obtained during the generation of multimedia data, resulting in a probability update matrix. Specifically, probabilistic perturbation information can be generated using a perturbation key, and this information is used to perturb the probability matrix obtained during multimedia data generation, thus integrating watermark information or encryption information into the probability matrix. Then, multimedia data is generated using the probability update matrix. This allows for subsequent verification of the multimedia data using the perturbation key embedded in the probability update matrix and a perturbation key provided by an object (such as a user), thereby verifying the copyright ownership of the multimedia data or whether the multimedia data has been stolen. The multimedia data here can include, but is not limited to, text, images, audio, and video.

[0072] The data processing solution provided in this application has the following beneficial effects:

[0073] (1) It can better protect the generated multimedia data from being copied. The solution of this application embodiment can perform complex probability perturbation on the probability matrix. For example, the perturbation key can be designed to be more complex. In this way, the probability perturbation information generated based on the perturbation key is more complex, and thus the probability matrix is ​​subjected to complex probability perturbation. In this way, the perturbation key can be used to prevent the generated multimedia data from being copied, and better protect the generated multimedia data from being copied.

[0074] (2) Lossless encryption is applied to the generated multimedia data to reduce its impact. The solution provided in this application directly perturbs the probability matrix during the multimedia data generation process, and effectively ensures that the final generated multimedia data remains unchanged during the perturbation process. In other words, it has no impact on the final generated multimedia data. This solves the problem that relying on lossy encryption watermarks can lead to the modification of multimedia data during the encryption process, affecting the quality of multimedia data generation. This application can improve the quality of generated multimedia data to a certain extent.

[0075] (3) To a certain extent, it can save computing resources and time costs, making it more efficient and reliable. Existing lossy encryption watermarking schemes use complex algorithms, such as Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT), which require high computing resources and time costs. When processing large files or real-time data streams, they may significantly reduce system performance and efficiency. However, the data processing scheme provided in this application directly and simply generates probability perturbation information, and generates probability perturbation information simply through the perturbation key, and perturbs the probability matrix with probability perturbation information. It does not require complex algorithms, has low algorithm complexity, and can save computing resources and time costs to a certain extent.

[0076] (4) It can better support dynamic encryption of generative models. The embodiments of this application perform dynamic encryption directly during the generation of multimedia data. This can solve the problem of only being able to encrypt statically, and can better realize real-time encryption of generative models during data generation and processing. It not only achieves efficient encryption of dynamically generated multimedia data, but also solves the security risks of multimedia generated by generative models.

[0077] (5) It has better compatibility and can better adapt to cross-platform use and data sharing scenarios. Existing encryption technologies have significant compatibility differences across different file formats, devices, and operating systems, which limits cross-platform use and data sharing. For example, encrypted data in binary files can be stored in an inode structure in Linux, but compatibility issues may exist on other systems. The data processing scheme provided by the embodiments of this application can encrypt multimedia data by updating the probability matrix through perturbation of the key pair without changing the generated content. This eliminates the need to consider compatibility between different file formats, devices, and operating systems, and does not rely on a specific platform, thus solving the cross-platform compatibility problem and better adapting to cross-platform and data sharing scenarios.

[0078] It should be understood that the data processing solution provided in this application embodiment can provide a solution for encryption in text generation scenarios and other format generation scenarios. That is, the data processing solution provided in this application embodiment includes, but is not limited to: text generation scenarios, image generation scenarios, video generation scenarios, audio and video scenarios, etc. The application of this data processing solution will be illustrated below with two specific scenario examples:

[0079] Application Scenario 1: Data traceability for content creation platforms.

[0080] In content creation platforms, data generation models are widely used to generate multimedia data such as text, images, and videos. By introducing the data processing scheme provided in this application, a perturbation key can be embedded during the content generation stage to encrypt the generated content and protect its copyright. For example, an online literature creation platform allows users to use generative large models (such as GPT-4) to generate literary works such as novels, essays, and poems. To protect the author's copyright and prevent plagiarism, the data processing scheme provided in this application obtains a probability matrix when generating text, generates probability perturbation information using a perturbation key, perturbs the probability matrix using the probability perturbation information, and then generates text based on the updated probability matrix. This text and the updated probability matrix can be stored together, thus embedding watermark information or encryption information into the text, thereby protecting the copyright of the generated text.

[0081] Application Scenario 2: Plagiarism prevention in academic papers.

[0082] In the process of academic paper writing and generation, using generative large models can significantly improve efficiency, but it is also necessary to ensure the originality of the content and copyright protection. The data processing scheme provided in this application can use a perturbation key to generate probability perturbation information to perturb the probability matrix during paper generation. The resulting probability update matrix embeds encrypted information, ensuring the originality of the paper and preventing plagiarism. For example, on an academic paper generation platform, researchers input prompts or frameworks, and the data generation model generates a paper and embeds an encrypted watermark. When the generated paper is downloaded or submitted to an academic institution for review, it includes a probability update matrix. That is, not only the paper is provided, but also a probability update matrix with embedded watermark information is provided to ensure originality and copyright protection. When receiving the paper, the academic institution verifies the source and authenticity of the paper through the probability update matrix with embedded watermark information, thus preventing plagiarism.

[0083] The data processing system provided in the embodiments of this application will be described in detail below.

[0084] Please see Figure 2 This is an architecture diagram of a data processing system provided in an embodiment of this application. The data processing system may include: terminal device 201, terminal device 202, and server 203; it should be understood that this embodiment does not limit the number of terminal devices and servers; terminal devices 201 and 202 can interact with server 203 via wired or wireless means. The following description uses terminal device 201 as an example:

[0085] Terminal device 201 can be a device used by any object. This object can be an object that needs to generate multimedia data. For example, if the multimedia data is a paper, then the object can be a researcher; if the multimedia data is video, then the object can be a video creator. Terminal device 201 can be, but is not limited to, smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, in-vehicle terminals, smart wearable devices, etc.

[0086] Server 203 can be a server that provides technical services to terminal device 201. Server 203 can call a data generation model to generate multimedia data based on the prompt information sent by terminal device 201. Server 203 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0087] In one embodiment, the data processing flow between the terminal device 201 and the server 203 includes: an encryption end (or a watermarking end) and a decryption end (or a watermark decryption end); wherein:

[0088] I. Encryption Terminal:

[0089] ① Prompt message input: The object inputs a prompt message through the terminal device 201. The terminal device 201 can respond to the prompt message input by the object and send the prompt message to the server 203. For example, the prompt message is "I like you".

[0090] ② Probability Matrix Acquisition: After receiving the prompt information from the terminal device 201, the server 203 can call the transformation model (such as the transformer model) in the data generation model to generate a probability matrix. This probability matrix includes the probability set corresponding to each multimedia object in the multimedia data to be generated, and the probability set includes the probabilities of N candidate objects corresponding to the multimedia object. For example... Figure 3 In the multimedia data to be generated, there is one multimedia object, and the probability set 11 corresponding to the multimedia object includes the probabilities of 8 (i.e., N=8) candidate objects corresponding to the multimedia object.

[0091] ③ Server 203 obtains the perturbation key and generates probabilistic perturbation information using the perturbation key. For example, if the perturbation key is "This is generated by huanyuan", probabilistic perturbation information can be generated based on "This is generated by huanyuan". This probabilistic perturbation information can be used to perturb the probability matrix. Probabilistic perturbation refers to updating the probability matrix based on this probabilistic perturbation information, thereby adding subtle interference.

[0092] ④ With the multimedia data generated by the data generation model remaining unchanged, server 203 performs probability perturbation on the probability matrix using probability perturbation information to obtain a probability update matrix. The probability update matrix includes a probability update set for each multimedia object, and the probability update set includes the update probabilities of candidate objects.

[0093] The statement that the multimedia data generated by the data generation model remains unchanged means that, after updating the probability matrix, the candidate object corresponding to the highest update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the highest probability in the probability set of the corresponding multimedia object. For example, Figure 3In the process, the probability matrix is ​​updated using probability perturbation information to obtain a probability update matrix. In the probability update set 31 of the multimedia objects included in the probability update matrix, the candidate object corresponding to the highest update probability is a cat; and the candidate object corresponding to the highest probability in probability matrix 11 is also a cat. Therefore, in the multimedia data generated based on the probability matrix and the multimedia data generated based on the probability update matrix, the multimedia object is a cat. This means that the multimedia data generated by the data generation model remains unchanged. Of course, when the multimedia data to be generated includes multiple multimedia objects, the candidate object corresponding to the highest update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the highest probability in the probability set of the corresponding multimedia object. This means that the corresponding multimedia objects determined by the probability sets and probability update sets of each multimedia object are the same. Thus, after updating the probability matrix using probability perturbation information, the final multimedia data remains unchanged.

[0094] Alternatively, the probabilities in the probability set can be arranged according to the target order (e.g., from largest to smallest), and the update probabilities in the probability update set can also be arranged according to the target order. The statement that the multimedia data generated by the data generation model remains unchanged can mean that the position of the update probability of each candidate object in the update probability set of the multimedia object does not change from the position of the update probability of the corresponding candidate object in the corresponding probability set of the multimedia object. For example, Figure 3 In the probability set 11, the candidate objects are arranged from largest to smallest, and the candidate objects in the probability update set 31 are also arranged from largest to smallest. However, the position of the probability of the cat in the probability set 11 is the same as the position of the updated probability of the cat in the probability update set 31, the position of the probability of the dog in the probability set 11 is the same as the position of the updated probability of the dog in the probability update set 31, and so on.

[0095] ⑤ Server 203 generates multimedia data based on the probability update matrix and N candidate objects for each multimedia object. For example... Figure 3 In the process, the probability update set of the multimedia object includes the update probabilities of 8 candidate objects. The candidate object with the highest update probability, "cat", is selected from these 8 candidate objects as the multimedia object. Finally, multimedia data is generated based on the multimedia object "cat", such as "I like your cat".

[0096] As can be seen, in this embodiment, while ensuring that the final generated multimedia data remains unchanged, probability perturbation information can be generated by perturbation key to update the probability matrix obtained in the process of generating multimedia data. This embeds the perturbation key into the probability matrix, thereby adding watermark information in the process of dynamically generating multimedia data. Without loss to the generated multimedia data, it can effectively prove the copyright and integrity of the multimedia data and prevent the multimedia data from being copied to a certain extent.

[0097] ⑥ Server 203 associates and stores the multimedia data with the probability update matrix, and outputs the generated multimedia data to terminal device 201. Terminal device 201 can store the multimedia data or send it to terminal device 202.

[0098] II. Decryption Terminal:

[0099] ① Server 203 can receive multimedia data to be verified and a probability update matrix associated with the multimedia data to be verified sent by terminal device 201 or terminal device 202. For ease of description, the multimedia data sent by terminal device 202 to terminal device 201 will be used as an example.

[0100] ② Server 203 obtains the verification perturbation key and generates verification probability perturbation information based on the verification perturbation key.

[0101] ③ Server 203 verifies the multimedia data based on the verification probability perturbation information and the probability update matrix. Specifically, the probability update matrix can be restored based on the verification probability perturbation information to obtain the original probability matrix. If the original multimedia data generated based on the original probability matrix is ​​consistent with the multimedia data, it means that the verification perturbation key is consistent with the perturbation key embedded in the probability update matrix, and the multimedia data belongs to the provider of the verification perturbation key, i.e., the multimedia data verification is successful. If the original multimedia data generated based on the original probability matrix is ​​inconsistent with the multimedia data, it means that the verification perturbation key is inconsistent with the perturbation key embedded in the probability update matrix, and the multimedia data does not belong to the provider of the verification perturbation key, i.e., the multimedia data verification fails. In this case, it can be considered that the verification perturbation key is incorrect, and the multimedia data may have been tampered with or copied.

[0102] In summary, at the decryption end, verification probability perturbation information can be generated based on the verification perturbation key. Then, the probability update matrix can be restored based on the verification probability perturbation information. By judging whether the original multimedia data generated by the restored original probability matrix is ​​consistent with the multimedia data to be verified, the copyright ownership and integrity of the multimedia data can be effectively proved. This can prevent the multimedia data from being copied to a certain extent and effectively ensure the security of the multimedia data.

[0103] It should be noted that in this embodiment, encryption and decryption can be performed simultaneously by server 203. Of course, in some optional embodiments, encryption and decryption can be performed by different terminal devices. For example, encryption can be performed by terminal device 201, and the server 203 can send the multimedia data to be verified and the associated probability update matrix to terminal device 202, which will then perform decryption. This embodiment does not impose any limitations on this. Furthermore, in some optional implementations, the data generation model may also include an independent encryption module. That is, generating probability perturbation information using a perturbation key and perturbing the probability matrix using this perturbation information can be performed by an independent encryption module in the data generation model.

[0104] The data processing method provided in the embodiments of this application will be described in detail below.

[0105] Please see Figure 4 This is a flowchart illustrating a data processing method provided in an embodiment of this application. The data processing method can be executed by a computer device, which can be the aforementioned server or terminal device. The data processing method may include at least steps S401-S403:

[0106] S401. Obtain the probability matrix obtained by the data generation model during the process of generating multimedia data. The probability matrix includes: the probability set corresponding to each multimedia object in the multimedia data to be generated, and the probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer.

[0107] In a specific implementation, prompt information can be obtained, which can be used to prompt the data generation model to generate multimedia data. Then, the data generation model is invoked to generate multimedia data based on the prompt information, and the probability matrix obtained by the data generation model in the process of generating multimedia data based on the prompt information is obtained. The data generation model may include a transformation model (such as a transformer model). The transformation model in the data generation model is invoked to generate a probability matrix associated with the multimedia data to be generated based on the prompt information. This transformation model is a deep learning model based on an attention mechanism, such as a transformer. That is, the embodiments of this application can obtain the probability matrix of the data generation model based on the attention mechanism in the process of generating multimedia data.

[0108] The multimedia object can be one or more, and the candidate objects for each multimedia object can be words (i.e., tokens). For example, the probability matrix can be as follows: Figure 3As shown, the probability matrix includes a probability set 11 corresponding to a multimedia object. Probability set 11 includes the probabilities of eight word units: "cat" (0.2082), "dog" (0.1914), "bird" (0.1560), "cow" (0.1079), "sheep" (0.1024), "smile" (0.0993), "eyes" (0.0890), and "guide" (0.0540). For example, the probability matrix can be like... Figure 5 As shown, the probability matrix includes the probability set corresponding to each of the three multimedia objects, such as... Figure 5 In the above, the probability set corresponding to the first multimedia object is probability set 51, the probability set corresponding to the second multimedia object is probability set 52, and the probability set corresponding to the third multimedia object is probability set 53.

[0109] It should be understood that in some embodiments, if the multimedia data to be generated is text, then the multimedia object can be a word (i.e., a token), and the N candidate objects of the multimedia object are also words. That is, the probability matrix includes the probabilities of the N words of the multimedia object. In other embodiments, if the multimedia data to be generated is an image, then the multimedia object can be a pixel, and the N candidate objects of the multimedia object are words. That is, the probability matrix includes the probabilities of the N words of the multimedia object, and the corresponding multimedia object is determined based on the correspondence between words and pixels. Similarly, if the multimedia data to be generated is video, the process is similar to that of images, and will not be elaborated further here.

[0110] S402. Update the probability matrix based on the obtained perturbation key to obtain the probability update matrix.

[0111] The probability update matrix includes a probability update set for each multimedia object. The candidate object corresponding to the highest update probability in each multimedia object's probability update set is the same as the candidate object corresponding to the highest probability in the corresponding multimedia object's probability set. For example... Figure 5 After the probability matrix in the probability update set is updated using probability perturbation information, a probability update matrix is ​​obtained. Specifically, the candidate object corresponding to the maximum update probability in the probability update set of the first multimedia object is the same as the candidate object corresponding to the maximum probability in the probability update set of the first multimedia object; the candidate object corresponding to the maximum update probability in the probability update set of the second multimedia object is the same as the candidate object corresponding to the maximum probability in the probability update set of the second multimedia object; and the candidate object corresponding to the maximum update probability in the probability update set of the third multimedia object is the same as the candidate object corresponding to the maximum probability in the probability update set of the third multimedia object.

[0112] In some implementations, a perturbation key is defined, and a pseudo-random number generator is used to generate probability perturbation information. The probability matrix is ​​then updated based on this perturbation information. Specific implementations of step S402 may include: ① hashing the acquired perturbation key to obtain a hash corresponding to the perturbation key; ② generating probability perturbation information based on the hash corresponding to the perturbation key; ③ updating the probability matrix based on the probability perturbation information to obtain a probability update matrix.

[0113] The generation of probability perturbation information based on the hash corresponding to the perturbation key can include: determining the probability perturbation seed of the pseudo-random number generator based on the hash corresponding to the perturbation key, and generating probability perturbation information by controlling the pseudo-random number generator. For example, the pseudo-random number generator can be controlled using Python's uniform methods. One implementation is to directly use the hash corresponding to the perturbation key as the probability perturbation seed of the pseudo-random number generator. Another implementation is to perform a modulo operation on the hash corresponding to the perturbation key, and then use the modulo result as the probability perturbation seed of the pseudo-random number generator. The modulo operation here can be: performing a modulo operation of 2 to the power of P (e.g., P = 32) on the hash corresponding to the perturbation key.

[0114] The probability perturbation information includes M perturbation values ​​associated with each multimedia object, where M is a positive integer. These M perturbation values ​​can be generated within a certain range by controlling a pseudo-random number generator, such as generating M perturbation values ​​between 0.95 and 1.05. M can be less than or equal to N, or M can be greater than N; this embodiment does not limit the relationship between M and N. Furthermore, the M perturbation values ​​associated with each multimedia object can be the same or different; this embodiment also does not limit this.

[0115] In other implementations, probabilistic perturbation information is generated by defining specific rules. For example, for M perturbation values ​​associated with a multimedia object, a specific rule can specify that the hash corresponding to the perturbation key is processed in an equal-decreasing manner to obtain M perturbation values. If the hash corresponding to the perturbation key is 1, then processing 1 in an equal-decreasing manner will yield perturbation values ​​of 0.8, 0.6, 0.4, etc.

[0116] The probability perturbation information includes M perturbation values ​​associated with each multimedia object. These perturbation values ​​are multiplied one by one with the corresponding probabilities in the probability matrix to form a probability update matrix. For ease of understanding, taking the target multimedia object as an example, the probability matrix is ​​updated based on the probability perturbation information to obtain the probability update matrix, including steps ①-②:

[0117] Step ①: From the M perturbation values ​​associated with the target multimedia object, determine the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object; the target multimedia object is any multimedia object in the multimedia data to be generated.

[0118] In one implementation, the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object can be determined from the M perturbation values ​​associated with the target multimedia object, based on the principle that the candidate object corresponding to the highest update probability in the probability update set of the updated target multimedia object is the same as the candidate object corresponding to the highest probability in the probability set of the corresponding multimedia object. Determining the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object from the M perturbation values ​​associated with the target multimedia object can include the following cases:

[0119] Scenario 1: If M = 1, then the perturbation value associated with the target multimedia object can be directly determined as the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object. That is, each probability in the probability set corresponding to the target multimedia object corresponds to the same perturbation value. For example, suppose the perturbation value is 1.02 (i.e., M = 1), and the probability set includes probability 1, probability 2, and probability 3, where probability 1 is greater than probability 2, and probability 2 is greater than probability 3; in this case, probability 1 corresponds to the perturbation value 1.02; probability 2 corresponds to the perturbation value 1.02, and probability 3 also corresponds to the perturbation value 1.02.

[0120] Scenario 2: If M is not equal to 1, then the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object can be determined from the M perturbation values ​​associated with the target multimedia object. As one implementation, M and N are the same, then each probability in the probability set corresponding to the target multimedia object corresponds to one perturbation value; that is, one probability corresponds to one perturbation value. (Illustrative example follows.) Figure 3In the probability matrix, there is a probability set 11 for the multimedia object. The probability set 11 includes the probability of 8 candidate objects, and there are also 8 perturbation values ​​associated with the target multimedia object. These 8 perturbation values ​​are represented as p = [1.00, 1.02, 0.98, 1.03, 0.99, 1.01, 0.97, 1.04]. The first perturbation value is 1.00, the second perturbation value is 1.02, and so on, with the 8th perturbation value being 1.04. Then, the perturbation value corresponding to each probability in the probability set of the target multimedia object can be determined from these 8 perturbation values. For example, the probability of "cat" (0.2082) corresponds to a perturbation value of 1.00; the probability of "dog" (0.1914) corresponds to a perturbation value of 1.02; the probability of "bird" (0.1560) corresponds to a perturbation value of 0.98; the probability of "cow" corresponds to a perturbation value of 1.03; the probability of "sheep" (0.1024) corresponds to a perturbation value of 0.99; the probability of "smiling" (0.0993) corresponds to a perturbation value of 0.01, and so on. As another implementation, since M and N are different, the probabilities in the probability set corresponding to the target multimedia object may correspond to the same perturbation value, or they may correspond to different perturbation values. For example, suppose there are 4 perturbation values ​​(M=4) represented as p=[1.00, 1.02, 0.98, 1.03], and the probability set includes probability 1, probability 2, and probability 3, where probability 1 is greater than probability 2, probability 2 is greater than probability 3; in this case, probability 1 corresponds to a perturbation value of 1.00, probability 2 corresponds to a perturbation value of 1.02, and probability 3 corresponds to a perturbation value of 0.98. As another example, suppose there are 3 perturbation values ​​(M=3) represented as p=[1.00, 1.02, 0.98], and the probability set includes probability 1, probability 2, and probability 3, where probability 1 is greater than probability 2, probability 2 is greater than probability 3, and probability 3 is greater than probability 4; in this case, probability 1 corresponds to a perturbation value of 1.00, probability 2 corresponds to a perturbation value of 1.02, probability 3 corresponds to a perturbation value of 0.98, and probability 4 corresponds to a perturbation value of 0.98. It is clear that probability 3 and probability 4 correspond to the same perturbation value.

[0121] Step ②: Multiply each perturbation value associated with the target multimedia object by the corresponding probability in the probability matrix to obtain the updated probability set of the target multimedia object. For example, in the above example, multiplying the probability of "cat" by the perturbation value 1.00 gives the updated probability of "cat," and multiplying the probability of "dog" by the perturbation value 1.02 gives the updated probability of "dog." Figure 3 The probability set 11 in the set is updated to the probability update set 31.

[0122] It should be understood that the probability set corresponding to each multimedia object in the multimedia data to be generated can be updated in the same way as the probability set corresponding to the target multimedia object described above. Finally, the probability update set of each multimedia object constitutes the probability update matrix. In this way, watermark information (i.e., probability perturbation information or perturbation key) can be embedded without significantly changing the probability matrix.

[0123] In some implementations, a marker can be embedded in the probability perturbation information, which can then be used to verify whether the perturbation key used to generate the probability perturbation information has been tampered with. Specifically, determining the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object from the M perturbation values ​​associated with the target multimedia object can include: ① Obtaining the target perturbation value from the M perturbation values ​​associated with the target multimedia object. ② Marking the target perturbation value; by marking the target perturbation value, it can be directly verified later to determine whether the perturbation key used to generate the probability perturbation information is mismatched or has been tampered with. One implementation is to place the target perturbation value at a marker position; the marker position here can be determined based on the maximum probability in the probability set corresponding to the target multimedia object. If the probabilities in the probability set corresponding to the target multimedia object are arranged in the order of the target, with higher probabilities appearing earlier, then the target perturbation value is placed in the first position (i.e., the marking position) to mark the target perturbation value. For example, if the target perturbation value is 1.00, then the target perturbation value 1.00 can be placed in the first position. Alternatively, if the probabilities in the probability set corresponding to the target multimedia object are arranged in the order of the target, with higher probabilities appearing later, then the target perturbation value is placed in the last position (i.e., the marking position) to mark the target perturbation value. As another implementation, special symbols (such as *, #) can be directly used to mark the target perturbation value, and this application embodiment does not limit this in any way. ③ The marked target perturbation value is mapped to the highest probability in the probability set corresponding to the target multimedia object, and other perturbation values ​​associated with the target multimedia object are mapped to other probabilities in the probability set corresponding to the target multimedia object. Other perturbation values ​​refer to perturbation values ​​other than the target perturbation value among the N perturbation values ​​associated with the target multimedia object; other probabilities refer to probabilities other than the highest probability in the probability set corresponding to the target multimedia object.

[0124] It should be understood that the embodiments of this application can select any perturbation value for marking, as long as the candidate object corresponding to the maximum update probability in the probability update set of the multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object, or in other words, as long as the multimedia data determined before and after adding probability perturbation information does not change.

[0125] S403. Generate multimedia data based on the probability update matrix and the N candidate objects for each multimedia object.

[0126] In this embodiment, the perturbation key is embedded into the probability update matrix, and the multimedia data and the probability update matrix are stored together. This is equivalent to adding watermark information or encryption information to the multimedia data, which can be used for copyright protection or copyright proof of the multimedia data.

[0127] In one implementation, step S403 may specifically include: determining a candidate object as the corresponding multimedia object from N candidate objects for each multimedia object based on the probability update matrix; specifically, determining the candidate object with the highest probability among the N candidate objects for each multimedia object, and using the candidate object with the highest probability as the corresponding multimedia object. For example... Figure 3 In the probability update matrix, there is a probability update set 31, which includes the update probabilities of 8 candidate objects of the multimedia object. The candidate object with the highest update probability is "cat". Therefore, "cat" with the highest update probability can be used as the multimedia object. Then, multimedia data is generated based on the multimedia object "cat", that is, the multimedia data is "I like your cat".

[0128] In another implementation, the probability update matrix can be normalized first to ensure that it remains a valid probability matrix. The specific implementation of step S403 can include: ① calculating the sum of the update probabilities in the probability update set corresponding to each multimedia object to obtain the probability sum for each multimedia object; ② normalizing the update probabilities in the probability update set corresponding to each multimedia object based on the probability sum for each multimedia object to obtain the normalized probability of each candidate object; ③ determining the candidate object as the corresponding multimedia object from the N candidate objects based on their normalized probabilities; specifically, determining the maximum normalized probability among the N candidate objects for each multimedia object and using the candidate object with the maximum normalized probability as the corresponding multimedia object; ④ generating multimedia data based on the corresponding multimedia object.

[0129] In this embodiment, a probability matrix obtained by the data generation model during the generation of multimedia data can be acquired. The probability matrix includes a probability set corresponding to each multimedia object in the multimedia data to be generated, where each probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer. Based on the acquired perturbation key, the probability matrix is ​​updated to obtain a probability update matrix. This probability update matrix includes the probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. Multimedia data is generated based on the probability update matrix and the N candidate objects for each multimedia object. It is evident that during the update of the probability matrix based on the perturbation key, it can be effectively ensured that the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. In other words, this embodiment can embed a perturbation key while ensuring the output remains unchanged, achieving lossless encryption of the generated multimedia data and guaranteeing the quality of the generated multimedia data to a certain extent. Furthermore, during the multimedia data generation process, the probability matrix obtained during the multimedia data generation process is updated by acquiring the perturbation key, thereby embedding the perturbation key into the probability update matrix. In this way, the copyright ownership of the generated multimedia data can be verified through the perturbation key in the probability update matrix. That is, the perturbation key achieves better copyright protection for the generated multimedia data without loss to the multimedia data.

[0130] Please see Figure 6 This is a flowchart illustrating a data processing method provided in an embodiment of this application. The data processing method can be executed by a computer device, which can be a server or a terminal device. The data processing method may include steps S601-S602:

[0131] S601. Obtain multimedia data and the probability update matrix associated with the multimedia data; the multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the obtained perturbation key. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer.

[0132] S602. Based on the obtained verification perturbation key and probability update matrix, verify the multimedia data to obtain the verification result of the multimedia data.

[0133] In one implementation, the methods for obtaining the verification perturbation key may include, but are not limited to, several methods:

[0134] Method 1: If the generation and verification of multimedia data are performed on the same computer device, the verification perturbation key can be stored in the trusted execution environment of the computer device, and the verification perturbation key can be directly obtained from the trusted execution environment of the computer device.

[0135] Method 2: If the generation of multimedia data and the verification of multimedia data are performed on different computer devices, then the perturbation key is encrypted on the computer device corresponding to the encryption end. Obtaining the verification perturbation key can be done by the computer device at the decryption end obtaining the encrypted verification perturbation key and decrypting it to obtain the verification perturbation key.

[0136] Method 3: Verify the perturbation key by storing it on the blockchain. Obtaining the verification perturbation key can be done by retrieving it from the blockchain. Based on the immutability and transparency of the blockchain, the verification perturbation key can be protected from being leaked, thus providing strong support for the copyright protection of generated multimedia data.

[0137] Method 4: If the generation and verification of multimedia data are performed on different computer devices, the verification perturbation key is encrypted on the computer device corresponding to the encryption end. Homomorphic encryption is used to encrypt the verification perturbation key. Homomorphic encryption allows computations to be performed on encrypted data (such as the perturbation key), ensuring that the verification perturbation key remains encrypted during use, thereby enhancing its security. Specifically, at the encryption end, the verification perturbation key is encrypted using a homomorphic encryption algorithm to generate ciphertext. The computer device at the decryption end can perform calculations (such as addition, multiplication, etc.) on the ciphertext. If the result of performing the same calculation operation on the decrypted verification perturbation key is the same, then the decryption is successful, and the verification perturbation key is obtained.

[0138] Method 5: Encrypt the verification perturbation key using differential privacy technology. Differential privacy technology protects privacy by adding noise to the data. Watermark information can be embedded during the generation of multimedia data while maintaining privacy. Obtaining the verification perturbation key in this case involves: acquiring the encrypted verification perturbation key and removing noise from it to obtain the final verification perturbation key.

[0139] In summary, the embodiments of this application can employ at least one method to protect the verification perturbation key, thereby preventing the leakage of the verification perturbation key and the resulting multimedia data from being stolen or tampered with, and can effectively ensure the security of the multimedia data verification process to a certain extent.

[0140] In one implementation, step S602 may include: generating verification probability perturbation information based on the obtained verification perturbation key. Specifically, the obtained verification perturbation key is hashed to obtain a hash corresponding to the verification perturbation key, and verification probability perturbation information is generated based on the hash corresponding to the verification perturbation key. Then, the probability update matrix is ​​restored based on the verification probability perturbation information and the probability update matrix to obtain the original probability matrix. The generation method of the verification probability perturbation information is similar to that of the probability perturbation information and will not be described in detail here. It should be understood that when the verification perturbation key is the same as the perturbation key, the verification probability perturbation information generated based on the verification perturbation key is the same as the probability perturbation information generated based on the perturbation key. When the verification perturbation key is different from the perturbation key, the verification probability perturbation information generated based on the verification perturbation key may be the same as or different from the probability perturbation information generated based on the perturbation key.

[0141] The process of verifying multimedia data based on the verification probability perturbation information and the probability update matrix to obtain the verification result of the multimedia data may include steps s11-s13:

[0142] s11. Based on the verification probability perturbation information, the probability update matrix is ​​restored to obtain the original probability matrix. The original probability matrix includes the original probability set corresponding to each multimedia object in the multimedia data to be generated. The original probability set includes the original probabilities of N candidate objects corresponding to the multimedia object.

[0143] In one implementation, it is necessary to first check whether the markers embedded during the encryption process are correct, and then restore the probability update matrix based on the obtained verification probability perturbation information to verify whether the multimedia data is legitimate. The verification probability perturbation information includes N verification perturbation values ​​associated with each multimedia object. Checking whether the markers embedded during the encryption process are correct includes: determining the target verification perturbation value corresponding to each multimedia object from the N verification perturbation values ​​associated with each multimedia object. The target verification perturbation value refers to the marked verification perturbation value, such as the verification perturbation value placed in the marked position. If the target verification perturbation value corresponding to each multimedia object is the same as the target perturbation value corresponding to the corresponding multimedia object in the probability perturbation information, it is determined that the verification perturbation key used to generate the verification probability perturbation information is consistent with the perturbation key and has not been tampered with, and step s11 is executed. If the target verification perturbation value corresponding to a multimedia object differs from the target perturbation value corresponding to the same multimedia object in the probability perturbation information, then the verification perturbation key used to generate the verification probability perturbation information is inconsistent with the perturbation key embedded in the probability update matrix. For example, the target perturbation value corresponding to the multimedia object in the probability perturbation information is 1.00; check if the target verification perturbation value in the verification probability perturbation information is the same. If they are different or not similar, it indicates that the perturbation keys do not match or the perturbation key has been tampered with. If they are the same or similar, it indicates that the verification perturbation key matches the perturbation key and has not been tampered with.

[0144] The verification probability perturbation information includes M verification perturbation values ​​associated with each multimedia object. Based on this information, the probability update matrix is ​​restored to obtain the original probability matrix. This process includes: determining the perturbation value corresponding to each probability in the probability set for each multimedia object from the M verification perturbation values ​​associated with each object; and dividing each updated probability in each probability update set of the probability update matrix by its corresponding verification perturbation value to obtain the original probability set for each multimedia object. By restoring the probability update matrix to obtain the original probability set for each multimedia object, the integrity of the multimedia data can be effectively verified, and the encrypted probability can be successfully recovered.

[0145] s12. Generate the original multimedia data based on the original probability matrix and the N candidate objects for each multimedia object.

[0146] In one implementation, step s12 may include: determining candidate objects as corresponding multimedia objects from N candidate objects for each multimedia object based on the original probability matrix, and generating original multimedia data based on the corresponding multimedia objects. Specifically, the candidate object with the highest original probability in the original probability set corresponding to each multimedia object in the original probability matrix may be determined, and the candidate object with the highest original probability may be used as the corresponding multimedia object. For example, if the candidate object with the highest original probability is a dog, then the dog may be used as the corresponding multimedia object, and the original multimedia data "I like your dog" may be generated.

[0147] s13. If the original multimedia data matches the multimedia data, then the verification result of successful multimedia data verification is obtained.

[0148] Specifically, the process checks if the original multimedia data matches the new multimedia data. If they don't match, the verification fails. This means the verification perturbation key doesn't match the one embedded in the probability update matrix, indicating the multimedia data doesn't belong to the provider of the verification perturbation key and may have been copied or stolen. Conversely, if they match, the verification succeeds. Again, this means the verification perturbation key doesn't match the one embedded in the probability update matrix, indicating the multimedia data belongs to the provider of the verification perturbation key and hasn't been copied or stolen. For example, in the above example, the multimedia data is "I like your cat," and the verification key is "This is generated by hunyuan." However, the original multimedia data is "I like your dog." This mismatch means an incorrect verification perturbation key was used, not "This is generated by hunyuan," implying the multimedia data has been tampered with or stolen. It is evident that by verifying the original probability matrix restored from the perturbation key, the copyright ownership of multimedia data can be determined, thereby achieving copyright protection for the generated multimedia data.

[0149] It should be understood that, taking multimedia data as text as an example, text consists of multiple segments of text. For many segments of text, a large probability matrix will be formed, such as... Figure 5 As shown, Figure 5The matrix in the code represents the probability matrix of one of the paragraphs. Each generated text segment contains a probability matrix for that paragraph. Each column of this probability matrix corresponds to a set of probabilities for a multimedia object. The probability update matrix corresponding to this paragraph's probability matrix is ​​downloaded along with the text. When text verification is required, verification probability perturbation information or a verification perturbation key (e.g., "This is generated by hunyuan") is used to generate verification probability perturbation information. This information is then used to restore each column of the probability update matrix. If the restored update matrix retains the position corresponding to the highest probability in each column, decryption is successful. This proves that the verification perturbation key is the same as the perturbation key embedded in the probability update matrix, meaning the verification perturbation key is correct and the generated text is valid. Otherwise, it fails, indicating that the verification perturbation key is incorrect, i.e., the verification perturbation key is not "This is generated by hunyuan".

[0150] In this embodiment, multimedia data and a probability update matrix associated with the multimedia data are obtained. The multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the obtained probability perturbation information. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer. The multimedia data is verified based on the obtained verification perturbation key and the probability update matrix to obtain the verification result of the multimedia data. Therefore, this embodiment can verify the copyright ownership of multimedia data by jointly verifying the verification perturbation key and the probability update matrix embedded with the perturbation key, thereby better protecting the copyright of the generated multimedia data.

[0151] The data processing apparatus provided in the embodiments of this application will be described in detail below.

[0152] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application. The data processing device can be a computer program (including program code) in a computer device; for example, the data processing device can be application software in a computer device. The data processing device can be used to execute... Figure 4 Some or all of the steps in the method embodiments shown. Please refer to [link / reference]. Figure 7 The data processing device includes the following units:

[0153] The acquisition unit 701 is used to acquire the probability matrix obtained by the data generation model in the process of generating multimedia data. The probability matrix includes: the probability set corresponding to each multimedia object in the multimedia data to be generated, and the probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer.

[0154] The processing unit 702 is used to update the probability matrix according to the acquired perturbation key to obtain a probability update matrix. The probability update matrix includes a probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object.

[0155] The processing unit 702 is also used to generate multimedia data based on the probability update matrix and N candidate objects for each multimedia object.

[0156] Specifically, when processing unit 702 updates the probability matrix based on the acquired perturbation key to obtain the probability update matrix, it can be used for:

[0157] The obtained perturbation key is hashed to obtain the hash corresponding to the perturbation key;

[0158] Based on the hash corresponding to the perturbation key, generate probability perturbation information;

[0159] The probability matrix is ​​updated based on the probability perturbation information to obtain the probability update matrix.

[0160] The probability perturbation information includes M perturbation values ​​associated with each multimedia object, where M is a positive integer. When processing unit 702 updates the probability matrix based on the probability perturbation information to obtain the probability update matrix, it can be specifically used for:

[0161] From the M perturbation values ​​associated with the target multimedia object, determine the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object; the target multimedia object is any multimedia object in the multimedia data to be generated;

[0162] Each perturbation value associated with the target multimedia object is multiplied by the corresponding probability in the probability matrix to obtain the probability update set of the target multimedia object.

[0163] Specifically, when determining the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object from the M perturbation values ​​associated with the target multimedia object, the processing unit 702 can be used for:

[0164] Obtain the target perturbation value from the M perturbation values ​​associated with the target multimedia object;

[0165] The target disturbance value is marked.

[0166] The marked target perturbation value is mapped to the maximum probability in the probability set corresponding to the target multimedia object, and other perturbation values ​​associated with the target multimedia object are mapped to other probabilities in the probability set corresponding to the target multimedia object;

[0167] Among them, other perturbation values ​​refer to the perturbation values ​​other than the target perturbation value among the N perturbation values ​​associated with the target multimedia object; other probabilities refer to the probabilities other than the maximum probability in the probability set corresponding to the target multimedia object.

[0168] Specifically, when generating multimedia data based on the probability update matrix and N candidate objects for each multimedia object, the processing unit 702 can be used for:

[0169] The probability sum of each multimedia object is obtained by summing the update probabilities in the probability update set corresponding to each multimedia object.

[0170] Based on the sum of probabilities corresponding to each multimedia object, the update probabilities in the probability update set corresponding to the corresponding multimedia object are normalized to obtain the normalized probabilities of each candidate object.

[0171] Based on the normalized probabilities of N candidate objects for each multimedia object, candidate objects are determined from the N candidate objects for each multimedia object as the corresponding multimedia object;

[0172] Generate multimedia data based on the corresponding multimedia object.

[0173] In this embodiment, a probability matrix is ​​obtained by the data generation model during the generation of multimedia data. The probability matrix includes: a probability set corresponding to each multimedia object in the multimedia data to be generated, where each probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer; the probability matrix is ​​updated using the obtained perturbation key to obtain a probability update matrix; this probability update matrix includes the probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object; multimedia data is generated based on the probability update matrix and the N candidate objects for each multimedia object. It is evident that during the update of the probability matrix using the perturbation key, it can effectively ensure that the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. In other words, this embodiment can embed a perturbation key while ensuring the output remains unchanged, achieving lossless encryption of the generated multimedia data and guaranteeing the quality of the generated multimedia data to a certain extent. Furthermore, during the multimedia data generation process, the probability matrix obtained during the multimedia data generation process is updated by acquiring the perturbation key, thereby embedding the perturbation key into the probability update matrix. This allows for better copyright protection of the generated multimedia data through the perturbation key.

[0174] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application. The data processing device can be a computer program (including program code) in a computer device; for example, the data processing device can be application software in a computer device. The data processing device can be used to execute... Figure 6 Some or all of the steps in the method embodiments shown. Please refer to [link / reference]. Figure 8 The data processing device includes the following units:

[0175] The acquisition unit 801 is used to acquire multimedia data and a probability update matrix associated with the multimedia data. The multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the acquired perturbation key. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data. N is a positive integer. The probability update matrix includes the probability update set for each multimedia object. The candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object.

[0176] The processing unit 802 is used to verify the multimedia data based on the obtained verification perturbation key and probability update matrix, and obtain the verification result of the multimedia data.

[0177] Specifically, when processing unit 802 verifies multimedia data based on the acquired verification perturbation key and probability update matrix to obtain the verification result of multimedia data, it can be used for:

[0178] Based on the obtained verification perturbation key, generate verification probability perturbation information;

[0179] Based on the verification probability perturbation information and the probability update matrix, the multimedia data is verified to obtain the verification result of the multimedia data.

[0180] Specifically, when processing unit 802 verifies multimedia data based on verification probability perturbation information and probability update matrix to obtain the verification result of multimedia data, it can be used for:

[0181] Based on the verification probability perturbation information, the probability update matrix is ​​restored to obtain the original probability matrix. The original probability matrix includes the original probability set corresponding to each multimedia object in the multimedia data to be generated. The original probability set includes the original probabilities of N candidate objects corresponding to the multimedia object.

[0182] Based on the original probability matrix and N candidate objects for each multimedia object, generate the original multimedia data;

[0183] If the original multimedia data matches the multimedia data, a verification result indicating successful multimedia data verification is obtained.

[0184] The verification probability perturbation information includes N verification perturbation values ​​associated with each multimedia object, and the probability perturbation information includes N perturbation values ​​associated with each multimedia object; the processing unit 802 can also be used for:

[0185] From the N verification perturbation values ​​associated with each multimedia object, determine the target verification perturbation value corresponding to each multimedia object. The target verification perturbation value refers to the marked verification perturbation value.

[0186] If the target verification perturbation value corresponding to each multimedia object is the same as the target perturbation value corresponding to the corresponding multimedia object in the probability perturbation information, then the step of restoring the probability update matrix according to the obtained perturbation key is executed to obtain the original probability matrix; the target perturbation value refers to the marked perturbation value.

[0187] In this embodiment, multimedia data and a probability update matrix associated with the multimedia data are obtained. The multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the obtained probability perturbation information. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer. The multimedia data is verified based on the obtained verification perturbation key and the probability update matrix to obtain the verification result of the multimedia data. Therefore, this embodiment can verify the copyright ownership of multimedia data by jointly verifying the verification perturbation key and the probability update matrix embedded with the perturbation key, thereby better protecting the copyright of the generated multimedia data.

[0188] The computer device provided in the embodiments of this application will be described in detail below.

[0189] Furthermore, this application also provides a schematic diagram of the structure of a computer device, which can be found in [reference needed]. Figure 9 The computer device may include a processor 901, an input device 902, an output device 903, and a memory 904. The processor 901, input device 902, output device 903, and memory 904 are connected via a bus. The memory 904 stores computer programs, which include program instructions, and the processor 901 executes the program instructions stored in the memory 904.

[0190] In one embodiment, processor 901 performs the following operations by executing program instructions stored in memory 904:

[0191] The probability matrix obtained by the data generation model during the generation of multimedia data includes: the probability set corresponding to each multimedia object in the multimedia data to be generated, and the probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer.

[0192] The probability matrix is ​​updated based on the obtained perturbation key to obtain the probability update matrix. The probability update matrix includes the probability update set of each multimedia object, the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object, and the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object.

[0193] Multimedia data is generated based on the probability update matrix and N candidate objects for each multimedia object.

[0194] Specifically, when the processor 901 updates the probability matrix based on the acquired perturbation key to obtain the probability update matrix, it can perform the following steps:

[0195] The obtained perturbation key is hashed to obtain the hash corresponding to the perturbation key;

[0196] Based on the hash corresponding to the perturbation key, generate probability perturbation information;

[0197] The probability matrix is ​​updated based on the probability perturbation information to obtain the probability update matrix.

[0198] The probability perturbation information includes M perturbation values ​​associated with each multimedia object, where M is a positive integer. When the processor 901 updates the probability matrix based on the probability perturbation information to obtain the probability update matrix, it can specifically perform the following steps:

[0199] From the M perturbation values ​​associated with the target multimedia object, determine the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object; the target multimedia object is any multimedia object in the multimedia data to be generated;

[0200] Each perturbation value associated with the target multimedia object is multiplied by the corresponding probability in the probability matrix to obtain the probability update set of the target multimedia object.

[0201] Specifically, when processor 901 determines the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object from the M perturbation values ​​associated with the target multimedia object, it may perform the following steps:

[0202] Obtain the target perturbation value from the M perturbation values ​​associated with the target multimedia object;

[0203] The target disturbance value is marked.

[0204] The marked target perturbation value is mapped to the maximum probability in the probability set corresponding to the target multimedia object, and other perturbation values ​​associated with the target multimedia object are mapped to other probabilities in the probability set corresponding to the target multimedia object;

[0205] Among them, other perturbation values ​​refer to the perturbation values ​​other than the target perturbation value among the N perturbation values ​​associated with the target multimedia object; other probabilities refer to the probabilities other than the maximum probability in the probability set corresponding to the target multimedia object.

[0206] Specifically, when generating multimedia data based on the probability update matrix and the N candidate objects for each multimedia object, the processor 901 can perform the following steps:

[0207] The probability sum of each multimedia object is obtained by summing the update probabilities in the probability update set corresponding to each multimedia object.

[0208] Based on the sum of probabilities corresponding to each multimedia object, the update probabilities in the probability update set corresponding to the corresponding multimedia object are normalized to obtain the normalized probabilities of each candidate object.

[0209] Based on the normalized probabilities of N candidate objects for each multimedia object, candidate objects are determined from the N candidate objects for each multimedia object as the corresponding multimedia object;

[0210] Generate multimedia data based on the corresponding multimedia object.

[0211] In this embodiment, a probability matrix is ​​obtained by the data generation model during the generation of multimedia data. The probability matrix includes: a probability set corresponding to each multimedia object in the multimedia data to be generated, where each probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer; the probability matrix is ​​updated using the obtained perturbation key to obtain a probability update matrix; this probability update matrix includes the probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object; multimedia data is generated based on the probability update matrix and the N candidate objects for each multimedia object. It is evident that during the update of the probability matrix using the perturbation key, it can effectively ensure that the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. In other words, this embodiment can embed a perturbation key while ensuring the output remains unchanged, achieving lossless encryption of the generated multimedia data and guaranteeing the quality of the generated multimedia data to a certain extent. Furthermore, during the multimedia data generation process, the probability matrix obtained during the multimedia data generation process is updated by acquiring the perturbation key, thereby embedding the perturbation key into the probability update matrix. This allows for better copyright protection of the generated multimedia data through the perturbation key.

[0212] In another embodiment, processor 901 performs the following operations by running program instructions stored in memory 904:

[0213] Obtain multimedia data and the probability update matrix associated with the multimedia data; the multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the obtained perturbation key. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer; the probability update matrix includes the probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object.

[0214] Based on the obtained verification perturbation key and probability update matrix, the multimedia data is verified to obtain the verification result of the multimedia data.

[0215] Specifically, when processor 901 verifies multimedia data based on the obtained verification perturbation key and probability update matrix, and obtains the verification result of multimedia data, it can perform the following operations:

[0216] Based on the obtained verification perturbation key, generate verification probability perturbation information;

[0217] Based on the verification probability perturbation information and the probability update matrix, the multimedia data is verified to obtain the verification result of the multimedia data.

[0218] When processor 901 verifies multimedia data based on verification probability perturbation information and probability update matrix to obtain the verification result of multimedia data, it can specifically perform the following operations:

[0219] Based on the verification probability perturbation information, the probability update matrix is ​​restored to obtain the original probability matrix. The original probability matrix includes the original probability set corresponding to each multimedia object in the multimedia data to be generated. The original probability set includes the original probabilities of N candidate objects corresponding to the multimedia object.

[0220] Based on the original probability matrix and N candidate objects for each multimedia object, generate the original multimedia data;

[0221] If the original multimedia data matches the multimedia data, a verification result indicating successful multimedia data verification is obtained.

[0222] The verification probability perturbation information includes N verification perturbation values ​​associated with each multimedia object, and the probability perturbation information includes N perturbation values ​​associated with each multimedia object; the processor 901 can also perform the following operations:

[0223] From the N verification perturbation values ​​associated with each multimedia object, determine the target verification perturbation value corresponding to each multimedia object. The target verification perturbation value refers to the marked verification perturbation value.

[0224] If the target verification perturbation value corresponding to each multimedia object is the same as the target perturbation value corresponding to the corresponding multimedia object in the probability perturbation information, then the step of restoring the probability update matrix according to the obtained perturbation key is executed to obtain the original probability matrix; the target perturbation value refers to the marked perturbation value.

[0225] In this embodiment, multimedia data and a probability update matrix associated with the multimedia data are obtained. The multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the obtained probability perturbation information. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer. The multimedia data is verified based on the obtained verification perturbation key and the probability update matrix to obtain the verification result of the multimedia data. Therefore, this embodiment can verify the copyright ownership of multimedia data by jointly verifying the verification perturbation key and the probability update matrix embedded with the perturbation key, thereby better protecting the copyright of the generated multimedia data.

[0226] In this application, the term "unit" refers to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more units. Furthermore, each unit can be part of an overall unit that includes the functionality of that unit.

[0227] Furthermore, it should be noted that this application also provides a computer-readable storage medium storing a computer program, which includes program instructions. When a processor executes these program instructions, it can execute the aforementioned... Figure 4 or Figure 6 The methods described in the corresponding embodiments are therefore not repeated here. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application. As an example, program instructions may be deployed on a computer device, executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected through a communication network.

[0228] According to one aspect of this application, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, enabling the computer device to perform the aforementioned... Figure 4 or Figure 6 The methods described in the corresponding embodiments are therefore not repeated here.

[0229] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0230] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A data processing method, characterized in that, include: The probability matrix obtained by the data generation model during the generation of multimedia data includes: a probability set corresponding to each multimedia object in the multimedia data to be generated, and the probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer. The probability matrix is ​​updated according to the obtained perturbation key to obtain a probability update matrix. The probability update matrix includes the probability update set of each multimedia object. The candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. Multimedia data is generated based on the probability update matrix and the N candidate objects for each multimedia object.

2. The method as described in claim 1, characterized in that, The step of updating the probability matrix based on the acquired perturbation key to obtain a probability update matrix includes: The obtained perturbation key is hashed to obtain the hash corresponding to the perturbation key; Based on the hash corresponding to the perturbation key, probability perturbation information is generated; The probability matrix is ​​updated based on the probability perturbation information to obtain the probability update matrix.

3. The method as described in claim 2, characterized in that, The probability perturbation information includes M perturbation values ​​associated with each multimedia object, where M is a positive integer; updating the probability matrix based on the probability perturbation information to obtain a probability update matrix includes: From M perturbation values ​​associated with the target multimedia object, determine the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object; the target multimedia object is any multimedia object in the multimedia data to be generated; Each perturbation value associated with the target multimedia object is multiplied by the corresponding probability in the probability matrix to obtain the probability update set of the target multimedia object.

4. The method as described in claim 3, characterized in that, The step of determining the perturbation value corresponding to each probability in the probability set corresponding to the target multimedia object from the M perturbation values ​​associated with the target multimedia object includes: Obtain the target perturbation value from the M perturbation values ​​associated with the target multimedia object; The target disturbance value is marked. The marked target perturbation value is mapped to the maximum probability in the probability set corresponding to the target multimedia object, and other perturbation values ​​associated with the target multimedia object are mapped to other probabilities in the probability set corresponding to the target multimedia object; Wherein, the other perturbation values ​​refer to the perturbation values ​​other than the target perturbation value among the N perturbation values ​​associated with the target multimedia object; the other probabilities refer to the probabilities other than the maximum probability in the probability set corresponding to the target multimedia object.

5. The method as described in claim 1, characterized in that, The step of generating multimedia data based on the probability update matrix and the N candidate objects for each multimedia object includes: The probability sum of each multimedia object is obtained by summing the update probabilities in the probability update set corresponding to each multimedia object. Based on the sum of probabilities corresponding to each multimedia object, the update probabilities in the probability update set corresponding to the corresponding multimedia object are normalized to obtain the normalized probabilities of each candidate object. Based on the normalized probabilities of the N candidate objects for each multimedia object, a candidate object is determined as the corresponding multimedia object from the N candidate objects for each multimedia object; Generate multimedia data based on the corresponding multimedia object.

6. A data processing method, characterized in that, The method includes: Acquire multimedia data and a probability update matrix associated with the multimedia data; the multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object; the probability update matrix is ​​obtained by updating the probability matrix based on the acquired perturbation key; the probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer; the probability update matrix includes a probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object; The multimedia data is verified based on the obtained verification perturbation key and the probability update matrix to obtain the verification result of the multimedia data.

7. The method as described in claim 6, characterized in that, The step of verifying the multimedia data based on the obtained verification perturbation key and the probability update matrix to obtain the verification result of the multimedia data includes: Based on the obtained verification perturbation key, generate verification probability perturbation information; The multimedia data is verified based on the verification probability perturbation information and the probability update matrix to obtain the verification result of the multimedia data.

8. The method as described in claim 7, characterized in that, The step of verifying the multimedia data based on the verification probability perturbation information and the probability update matrix to obtain the verification result of the multimedia data includes: Based on the verification probability perturbation information, the probability update matrix is ​​restored to obtain the original probability matrix. The original probability matrix includes the original probability set corresponding to each multimedia object in the multimedia data to be generated. The original probability set includes the original probabilities of N candidate objects corresponding to the multimedia object. Based on the original probability matrix and the N candidate objects for each multimedia object, generate the original multimedia data; If the original multimedia data matches the multimedia data, a verification result indicating successful verification of the multimedia data is obtained.

9. The method as described in claim 8, characterized in that, The verification probability perturbation information includes N verification perturbation values ​​associated with each multimedia object, and the probability perturbation information includes N perturbation values ​​associated with each multimedia object; the method further includes: From the N verification perturbation values ​​associated with each multimedia object, determine the target verification perturbation value corresponding to each multimedia object, where the target verification perturbation value refers to the marked verification perturbation value; If the target verification perturbation value corresponding to each multimedia object is the same as the target perturbation value corresponding to the corresponding multimedia object in the probability perturbation information, then the step of restoring the probability update matrix according to the obtained perturbation key to obtain the original probability matrix is ​​executed; the target perturbation value refers to the marked perturbation value.

10. A data processing apparatus, characterized in that, include: The acquisition unit is used to acquire the probability matrix obtained by the data generation model during the process of generating multimedia data. The probability matrix includes: a probability set corresponding to each multimedia object in the multimedia data to be generated, and the probability set includes the probabilities of N candidate objects corresponding to the multimedia object; N is a positive integer. The processing unit is used to update the probability matrix according to the acquired perturbation key to obtain a probability update matrix. The probability update matrix includes a probability update set for each multimedia object. The candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. The processing unit is used to generate multimedia data based on the probability update matrix and N candidate objects for each multimedia object.

11. A data processing apparatus, characterized in that, include: An acquisition unit is used to acquire multimedia data and a probability update matrix associated with the multimedia data; The multimedia data is determined based on the probability update matrix and N candidate objects for each multimedia object. The probability update matrix is ​​obtained by updating the probability matrix based on the acquired perturbation key. The probability matrix is ​​obtained by the data generation model during the generation of multimedia data; N is a positive integer. The probability update matrix includes a probability update set for each multimedia object, and the candidate object corresponding to the maximum update probability in the probability update set of each multimedia object is the same as the candidate object corresponding to the maximum probability in the probability set of the corresponding multimedia object. The processing unit is used to verify the multimedia data based on the obtained verification perturbation key and the probability update matrix, and obtain the verification result of the multimedia data.

12. A computer device, characterized in that, include: A processor is used to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, performs the data processing method according to any one of claims 1-9.

13. A computer-readable storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, performs the data processing method according to any one of claims 1-9.

14. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the data processing method according to any one of claims 1-9.

Images