Data processing method and electronic equipment

Abstract

The embodiment of the invention discloses a data processing method and electronic equipment. The method comprises the steps of obtaining input data; obtaining description data used for representing a limiting condition, wherein the input data comprises target data corresponding to the limiting condition; based on a target model, the input data and the description data, processing the input data to generate output data, the target model being a generative large language model; and outputting the output data, wherein the output data does not comprise target data corresponding to the limiting condition.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a data processing method and electronic equipment. Background Art

[0002] In related technologies, when processing input data based on a model, the model usually processes all input data according to a specific processing logic to obtain a processing result. In this way, the model's behavior of processing input data or the model's output result is not controllable. In actual processes, it is often necessary to restrict the model's output results so that the output results meet specific requirements. Summary of the invention

[0003] In view of this, the embodiments of the present application at least provide a data processing method and an electronic device.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] On the one hand, an embodiment of the present application provides a data processing method, including: obtaining input data; obtaining descriptive data for characterizing restrictions, the input data including target data corresponding to the restrictions; based on a target model, the input data and the descriptive data, processing the input data to generate output data, the target model being a generative large language model; outputting output data, the output data not including the target data corresponding to the restrictions.

[0006] On the other hand, an embodiment of the present application provides an electronic device, comprising: a first input component, the first input component is used to obtain input data; a second input component, the second input component is used to obtain descriptive data used to characterize a restriction condition, wherein the input data includes target data corresponding to the restriction condition; a processor, the processor is used to process the input data to generate output data based on a target model, the input data and the descriptive data, the target model is a generative large language model; an output component, the output component is used to output the output data, wherein the output data does not include the target data corresponding to the restriction condition.

[0007] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0009] Figure 1 A schematic diagram of an implementation process of a data processing method provided in an embodiment of the present application Figure 1 ;

[0010] Figure 2 A schematic diagram of an implementation process of a data processing method provided in an embodiment of the present application Figure 2 ;

[0011] Figure 3 A schematic diagram of an implementation process of a data processing method provided in an embodiment of the present application Figure 3 ;

[0012] Figure 4 A schematic diagram of an implementation process of a data processing method provided in an embodiment of the present application Figure 4 ;

[0013] Figure 5 A schematic diagram of an implementation process of a data processing method provided in an embodiment of the present application Figure 5 ;

[0014] Figure 6 A summary of the content of a conversation between two parties proposed in the embodiment of the present application Figure 1 ;

[0015] Figure 7 A summary of the content of a conversation between two parties proposed in the embodiment of the present application Figure 2 ;

[0016] Figure 8 A schematic diagram of the data processing flow during a two-way conversation proposed in an embodiment of the present application;

[0017] Fig. 9 A schematic diagram of the structure of a data processing device provided in an embodiment of the present application;

[0018] Fig.10 A schematic diagram of a hardware entity of an electronic device provided in an embodiment of the present application;

[0019] Fig.11 A hardware entity diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions of the present application are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0021] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "first / second / third" involved are merely used to distinguish similar objects and do not represent a specific order for the objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing this application and are not intended to limit this application.

[0023] The target model involved in this application is one of the following: a machine learning model, a large language model, and a generative large language model.

[0024] Among them, the machine learning model can recognize natural language and / or other inputs (such as audio, video, images, tables, etc.) input into the target model, and perform comprehensive language processing tasks such as analyzing semantics and answering questions, and then generate output related to the input and / or respond to the input.

[0025] Among them, the large language model learns the characteristics and patterns of natural language by training a large amount of diverse data, so that it can understand and generate natural language. Large language models usually have parameters ranging from billions to hundreds of billions, and can capture complex relationships and patterns in natural language. The large language model is an artificial intelligence model based on deep learning technology. It learns the laws and patterns of language by analyzing a large amount of text data, so that it can understand and generate natural language text. The main task of the generative large language model is to generate natural language text. It learns the statistical laws of language by analyzing a large amount of text data, and generates text that conforms to grammatical and semantic rules based on this knowledge and given context or prompts.

[0026] Among them, generative language models (GLMs) can include, for example, large language models (LLMs), GPT (Generative Pre-trained Transformer), large visual models, large multimodal models, etc., and can also be expert large models obtained after fine-tuning based on needs, which is not limited in the embodiments of the present application.

[0027] Generative large language models are primarily used to generate natural language text. A generative large language model can take a given input (such as a sentence or prompt) and generate a response: a coherent and contextually relevant sentence or even a paragraph based on the given prompt or input. The model uses a variety of techniques (including attention mechanisms, transformers, and neural networks) to process the input and generate output that is intended to be coherent and contextually appropriate.

[0028] Generative large language models can learn the statistical rules of language by analyzing large amounts of text data, and generate text that conforms to grammatical and semantic rules based on the learned knowledge and the provided context or prompts. Generative large language models can enhance various applications such as content generation, personalization, and multimodal content creation.

[0029] The output results of the above target model are not restricted. For example, if user privacy data or security data is input into the target model, the target model processes the input data based on the processing logic determined by training, resulting in the output results of the target model being not controllable. This is not conducive to the interaction between the user and the target model.

[0030] The embodiment of the present application proposes a data processing method, which includes obtaining input data; obtaining descriptive data for characterizing restriction conditions, wherein the input data includes target data corresponding to the restriction conditions; processing the input data to generate output data based on a target model, the input data, and the descriptive data, wherein the target model is a generative large language model; and outputting the output data, wherein the output data does not include the target data corresponding to the restriction conditions. In this way, when processing the input data, the output data is restricted by the descriptive data of the restriction conditions, so that the output data of the target model does not include the target data corresponding to the restriction conditions, thereby achieving effective restriction on the output data, avoiding arbitrary output results of the target model, and effectively controlling the output behavior of the target model, so that the output data can meet the requirements.

[0031] The present application embodiment provides a data processing method, which can be executed by a processor of a computer device. The computer device may refer to a server, a laptop, a tablet computer, a desktop computer, a smart TV, a set-top box, a mobile device (such as a mobile phone, a portable video player, a personal digital assistant, a dedicated messaging device, a portable gaming device), or other device with data processing capabilities.

[0032] Figure 1 A schematic diagram of an implementation process of a data processing method provided in an embodiment of the present application Figure 1 ,like Figure 1 As shown, the method includes the following steps S101 to S104:

[0033] Step S101: Obtain input data.

[0034] Among them, input data is data that needs to be processed.

[0035] In some embodiments, during a voice call between at least two users, a call assistant in the user's terminal device can record the voice of each user during the call, use the voices of all users as input data, and process the voice content.

[0036] In some embodiments, during a video call between at least two users, a call assistant in the user's terminal device can record the video call content, use all users' videos as input data, and process the video content.

[0037] In some embodiments, during an online conference between at least two users, a conference assistant in a user's terminal device may record the audio of the conference and use the audio of all users as input data to process the conference audio.

[0038] In some embodiments, the input data may be at least one piece of data determined by a user, and the data may be audio, video or text saved by the user. Exemplarily, a segment of audio pre-stored by the user is used as the input data.

[0039] Step S102: Obtain description data for characterizing restriction conditions, wherein the input data includes target data corresponding to the restriction conditions.

[0040] The description data is data about the restriction condition, and the restriction condition corresponds to the target data in the input data, and the target data may be data in the input data that is not allowed / needed to be processed. The final output result can be limited by the description data.

[0041] In some embodiments, during a voice call between at least two users, any user may propose restriction conditions regarding call voice processing and use the restriction conditions as descriptive data.

[0042] In some embodiments, during a video call between at least two users, any user may propose restrictions on video processing and use the restrictions as description data.

[0043] In some embodiments, during an online conference between at least two users, any user may propose restrictions on conference audio processing and use the restrictions as description data.

[0044] In some embodiments, the description data may be at least one user-determined data, which is predetermined by the user for the input data. For example, the user needs to summarize a saved audio segment, and the description data may be content that cannot be summarized for the audio segment.

[0045] Exemplarily, when the input data is a conversation voice of at least two users, the description data may be a name or location information that cannot be summarized in the conversation voice.

[0046] Step S103: Based on the target model, the input data and the description data, the input data is processed to generate output data, wherein the target model is a generative large language model.

[0047] After obtaining the input data and the description data of the restriction conditions, the input data can be processed based on the target model, the input data and the description data to generate the output data.

[0048] In some embodiments, the target model is a generative large language model, which can process input data to generate a summary or abstract, thereby obtaining a summary or abstract corresponding to the input data. Exemplarily, a piece of text is input into the generative large language model, and an abstract corresponding to the text is output (generated).

[0049] In some embodiments, the target model may be other neural network models, such as a convolutional neural network, a recurrent neural network, etc.

[0050] In some embodiments, target data in the input data is determined based on the description data, the target data is cleaned from the input data to obtain cleaned input data, and the cleaned input data is input into a target model to generate output data.

[0051] In some embodiments, the input data and the description data are input into the target model together, and the target model processes the input data based on the description data to generate output data. The target model can process the input data in any manner, as long as the output data does not include the target data corresponding to the restriction condition.

[0052] Step S104: output the output data, wherein the output data does not include target data corresponding to the restriction condition.

[0053] After the output data is generated in the previous step, the output data needs to be output. The output data is the data after the input data is finally processed, and the output data does not include the target data corresponding to the restriction condition.

[0054] Exemplarily, the input data is the speech of user 1 and user 2, and the generative large language model can summarize the content of the speech. The description data is that the mobile phone number and location information in the speech of user 1 are not allowed to be summarized. After processing the input speech, the output data does not include the mobile phone number and location information in the speech of user 1.

[0055] In the embodiment of the present application, input data is obtained; description data for characterizing the restriction condition is obtained, the input data includes target data corresponding to the restriction condition; based on the target model, the input data and the description data, the input data is processed to generate output data, the target model is a generative large language model; the output data is output, and the output data does not include the target data corresponding to the restriction condition. In this way, when processing the input data, the output data is restricted by the description data of the restriction condition, so that the output data of the target model does not include the target data corresponding to the restriction condition, thereby achieving effective restriction on the output data, avoiding arbitrary output results of the target model, effectively controlling the output behavior of the target model, and the output data can meet the demand.

[0056] Figure 2 This is a schematic diagram of the implementation process of a data processing method provided in an embodiment of the present application. Figure 2 , the method can be executed by a processor of a computer device. Figure 1 , Figure 1 S101 can be updated to S201, S102 can be updated to S202, and the combination Figure 2 The steps shown are explained.

[0057] Step S201: Obtain information input from at least two objects to form the input data.

[0058] In some embodiments, at least two objects are at least two objects in a voice call. During the voice call, the smart device records the call voice of at least two objects, obtains the call voice of each object, and uses it as object information input to constitute input data.

[0059] In some embodiments, the at least two objects are at least two objects in a video call. During the video call, the smart device records videos of the at least two objects, obtains the video of each object, and uses it as object information input to constitute input data.

[0060] In some embodiments, the at least two objects are at least two objects in an online meeting. During the meeting, the smart device records the audio of the at least two objects, obtains the audio of each object, and uses it as object information input to constitute input data.

[0061] In some embodiments, at least two objects may be objects from the same device, and at least two input boxes exist on the input data interface of the same device, each input box corresponds to one object, and each object inputs information to be processed in the corresponding input box, and the information input of at least two objects constitutes input data. The information input by different objects can be distinguished.

[0062] In some embodiments, at least two objects may be objects from the same device. There is an input box on the input data interface of the same device. Each object may input information to be processed in the input box and mark the information to distinguish it. The information input of at least two objects constitutes input data.

[0063] Step S202: Obtain the restriction condition input by at least one of the at least two objects to form the description data.

[0064] At least one of the at least two objects will input restriction conditions, and the input restriction conditions constitute description data.

[0065] In some embodiments, at least two objects are at least two objects in a voice call. During the voice call, at least one object will restrict the call voice, input restriction conditions, and form the input restriction conditions into description data.

[0066] In some embodiments, the at least two objects are at least two objects in a video call. During the video call, at least one of the objects may restrict the content of the video, input restriction conditions, and use the input restriction conditions to form description data.

[0067] In some embodiments, the at least two objects are at least two objects in an online conference. During the conference, at least one object may restrict the audio of the conference, input restriction conditions, and form the input restriction conditions into description data.

[0068] In some embodiments, the restriction condition input by at least one object can restrict the information input by all objects. For example, for user 1 and user 2 in a voice call, user 1 proposes a restriction condition, and the restriction condition is used to restrict the position information in the voice of user 1 and user 2 (that is, the output data does not include the position information in the voice of user 1 and user 2).

[0069] In some embodiments, the restriction condition of at least one object input can only restrict the information input of the corresponding object. For example, for user 1 and user 2 in a voice call, user 1 proposes a restriction condition, and the restriction condition can only be used to restrict the position information in user 1's voice (that is, the output data does not include the position information in user 1's voice), and cannot restrict the position information in user 2's voice (that is, the output data includes the position information in user 2's voice).

[0070] Exemplarily, in a two-party voice call scenario, the call voice of each user is input as information to constitute input data; and the restriction conditions for the call voice input by one of the users constitute description data.

[0071] In the embodiment of the present application, information input of at least two objects is obtained to form the input data; and restriction conditions of at least one of the at least two objects are obtained to form the description data. In this way, input data and description data can be obtained from at least two objects, thereby improving the richness and diversity of input data and description data.

[0072] In some embodiments, the restriction condition from the first-party object is applied to a portion of the input data belonging to the first-party object.

[0073] The first party object may be any one of the at least two party objects.

[0074] Wherein, for at least two objects, each object has corresponding information input, and the input data includes partial data of each object. The restriction condition of the first object can only act on the partial data belonging to the first object in the input data, but cannot act on the partial data of other objects.

[0075] For example, for user 1 and user 2 in a voice call, the input data includes the voice of user 1 and the voice of user 2. User 1 proposes a restriction condition, which can only apply to the voice of user 1 but not to the voice of user 2.

[0076] In the embodiment of the present application, the restriction condition from the first party object acts on the part of the input data belonging to the first party object. In this way, the restriction condition can only act on part of the input data of the same object, which can avoid the restriction condition from affecting other objects except the object corresponding to the restriction condition, enhance the separation of permissions between different objects, and improve the pertinence of the restriction condition.

[0077] Figure 3 This is a schematic diagram of the implementation process of a data processing method provided in an embodiment of the present application. Figure 3 , the method can be executed by a processor of a computer device. Figure 2 , Figure 2 S201 in can be updated to S301 and S302. Figure 2 S202 in can be updated to S303, which will be combined with Figure 3 The steps shown are explained.

[0078] Step S301: Establish a communication channel with at least one object.

[0079] In the process of communication, at least two objects exchange information and obtain information input based on the established communication channel to form input data. The local object needs to establish a communication channel with at least one object to achieve communication between the local object and at least one object based on the communication channel.

[0080] In some embodiments, when at least two objects are conducting a voice call, each object establishes a voice call channel with at least one object.

[0081] In some embodiments, when at least two objects are in a video call, each object establishes a video call channel with at least one object.

[0082] In some embodiments, when at least two objects are having an online conference, each object establishes a conference channel with at least one object.

[0083] In some embodiments, when at least two objects interact with each other via text messages, each object establishes a text message channel with at least one other object.

[0084] Step S302: Obtain information input of the own object and information input of the other party object to form the input data; the other party object belongs to the at least one object.

[0085] After the communication channel is successfully established, at least two objects can communicate based on the communication channel. During the communication process, information input of the local object and information input of the other party object are obtained based on the communication channel to form input data, and the other party object belongs to at least one object. The local object is the object that needs to process the input data.

[0086] In some embodiments, in a scenario where three parties are conducting a voice call, each party establishes a voice call channel with the other two parties for the voice call. One of the parties needs to summarize the voice content of the three parties, and the one party serves as the local party. The call assistant of the smart device can record the voices of the three parties based on the communication channel to form input data.

[0087] In some embodiments, when at least two objects are making a video call, the call assistant obtains the video input of the object on the party and the video input of the object on the other party based on the established video call channel to form input data. The object on the party is the object for which the video call content needs to be summarized.

[0088] In some embodiments, when at least two parties are having an online conference, the conference assistant obtains the conference audio of the party concerned and the conference audio of the other party concerned based on the established conference channel to form input data. The party concerned is the party that needs to summarize the conference content.

[0089] Step S303: Analyze the input data to determine input restriction conditions of at least one of the own object and the other object.

[0090] In the process of communicating based on the communication channel, at least one object will add the restriction condition of the object to the input data. Therefore, by analyzing the input data, the restriction condition input by at least one object of the own object and the other object can be determined. The way of analyzing the input data to obtain the restriction condition can be arbitrary. Exemplarily, the input data is input into the speech analysis model, and the speech analysis model processes the data to obtain the restriction condition.

[0091] In some embodiments, when at least two parties are conducting a voice call, at least one party may add restrictions in the voice, analyze the voices of the at least two parties, and determine the restrictions input by the at least one party.

[0092] In some embodiments, when at least two parties are conducting a video call, at least one party may add restrictions in the video, and the video of the at least two parties may be analyzed to determine the restrictions input by the at least one party.

[0093] In some embodiments, when at least two parties are conducting an online conference, at least one party may add restrictions to the conference audio, analyze the conference audio of the at least two parties, and determine the restrictions input by the at least one party.

[0094] In the embodiment of the present application, a communication channel is established with at least one object; information input of the own object and information input of the other object are obtained to form the input data; the other object belongs to the at least one object. In this way, input data can be obtained based on the communication channel of at least two objects. The input data is analyzed to determine the restriction conditions of the input of at least one object of the own object and the other object. In this way, the restriction conditions exist in the input data, and the input data is analyzed to obtain the restriction conditions, which simplifies the step of obtaining the restriction conditions.

[0095] In some embodiments, if the other party inputs a restriction condition, a prompt mark indicating that the other party has inputted a restriction condition may be displayed on the current call page. The prompt mark may be an icon or the content of the corresponding restriction condition, and so on.

[0096] In some embodiments, if the other party enters a restriction condition, the Figure 6 Or Figure 7The page of the summary of the artificial intelligence (AI) assistant (AI agent) shows a prompt mark that the other party has entered a restriction condition. The prompt mark can be an icon or the content of the corresponding restriction condition, etc. Among them, the agent is an application of artificial intelligence technology, which can be implemented based on the target model. Its behavior is determined by the target model according to the current state of the agent and external input. Among them, the target model provides a decision basis for the agent through learning and training of a large amount of data; the agent can call tools, plug-ins and knowledge bases to provide reasoning, decision-making and execution capabilities. The target model provides decision support for the agent.

[0097] In some embodiments, the restriction conditions input by at least one of the at least two objects are obtained, and the description data comprises at least one of the following: if the communication channel with at least one object is successfully established, the restriction conditions input by the other object are obtained; if the communication channel with at least one object is disconnected, the restriction conditions input by the other object are obtained; in the process of communicating based on the established communication channel with at least one object, the restriction conditions input by the other object are obtained.

[0098] In some embodiments, when the communication channel between the local object and at least one other object is successfully established, the other object will be informed that it will obtain input data and process the input data at this time. At this time, the other object can provide restrictions based on the currently established communication channel or other communication channels. Exemplarily, during a voice call between the two parties, the local object needs to record the call voice and summarize the content of the recording. When recording the voice, the other object will be informed that the local object has turned on recording. At this time, the other object can provide restrictions based on the voice call channel (for example, it is not allowed to summarize the other object's voice), or inform the local object of the input restrictions by text message.

[0099] In some embodiments, when the communication channel between the local object and at least one other object is disconnected, the other object will be informed that it will obtain input data and process the input data. At this time, the other object can provide restriction conditions based on the currently established communication channel or other communication channels. For example, during a voice call between the two parties, the local object needs to record the call voice and summarize the content of the recording. After the call is disconnected, the voice call assistant will automatically send a data processing request to the other object in the form of a text message. After receiving the data processing request, the other object will inform the local object of the input restriction conditions in the form of a text message.

[0100] In some embodiments, during the process of establishing a communication channel between the local object and at least one other object for communication, the other object learns from the communication content of the local object that the local object is going to summarize the communication content, and at this time the other object can directly tell the other object the input restriction conditions based on the current communication method. Exemplarily, during a voice call between the two parties, the local object needs to record the call voice and summarize the content of the recording. At this time, the other object has been informed, and the other object can directly tell the local object the restriction conditions through voice.

[0101] In some embodiments, in a multi-person communication scenario, different restriction conditions input by the other party objects can be obtained at different times, that is, for the first other party object, the restriction conditions of the other party object can be obtained when the communication channel is successfully established; for the second other party object, the restriction conditions of the other party object can be obtained during the process of establishing the communication channel for communication; for the third other party object, the restriction conditions of the other party object can be obtained when the communication channel is disconnected.

[0102] In the embodiment of the present application, considering obtaining the restriction conditions input by the other party object at different times of communication based on the communication channel can increase the diversity of the methods of obtaining the restriction conditions, making the restriction conditions more flexible.

[0103] In some embodiments, obtaining the description data for characterizing the restriction condition further comprises: obtaining the description data for characterizing the restriction condition in response to calling the target model.

[0104] After obtaining the input data, the input data may not be processed immediately, and the description data of the restriction condition is not needed at this time. The description data of the restriction condition is only needed when the target model needs to be called to process the input data. Therefore, in response to calling the target model, the description data for characterizing the restriction condition is obtained.

[0105] The description data is a restriction condition of at least one object input corresponding to the input data.

[0106] For example, during a voice call between two parties, the party records the voice of the call. After the call, the voice may not be summarized, so there is no need to obtain description data. The description data is only required when the voice needs to be processed based on the target model.

[0107] In some embodiments, after obtaining the description data for characterizing the restriction condition, if the target data corresponding to the restriction condition is all the input data, the subsequent processing process is stopped.

[0108] In the embodiment of the present application, in response to calling the target model, description data for characterizing the restriction condition is obtained. In this way, description data is obtained only when the target model is called, which can accurately grasp the timing, avoid obtaining invalid description data, and improve the efficiency and accuracy of data processing.

[0109] Figure 4 This is a schematic diagram of the implementation process of a data processing method provided in an embodiment of the present application. Figure 4 , the method can be executed by a processor of a computer device. Figure 1 , Figure 1 S103 in can be updated to S401 and S402, which will be combined Figure 4 The steps shown are explained.

[0110] Step S401: Process the input data based on the description data to obtain vector data for inputting the target model, wherein the processing of the input data based on the description data is used to clean up data satisfying the restriction condition from the input data.

[0111] When processing the input data, it is necessary to first process the input data based on the description data to obtain vector data that can be input into the target model, and then process the vector data based on the target model to generate output data. When processing the input data based on the description data, it is necessary to clean up the data that meets the restriction conditions from the input data to obtain cleaned vector data.

[0112] In some embodiments, the description data and the input data are first vectorized to obtain vectorized description data and input data, and then based on the vectorized description data, data that meets the restriction condition is cleaned from the vectorized input data to obtain cleaned vector data, and the vector data is input into the target model for processing.

[0113] In some embodiments, data satisfying restriction conditions are first cleaned from input data based on description data to obtain cleaned input data, and then the cleaned input data is vectorized to obtain vector data of the input target model.

[0114] The existing data cleaning technology is used to clean the data satisfying the restriction conditions from the input data. The existing data cleaning technology may be a data cleaning model.

[0115] Exemplarily, the input data is speech, which is converted into text through speech-to-text technology, and the description data is text. For example, the position information in the speech is cleared, and based on the description data, the position information in the text corresponding to the speech is cleared to obtain the cleaned text, which is vectorized to obtain vector data for the input target model.

[0116] Step S402: Process the vector data based on the target model to generate the output data.

[0117] After obtaining the cleaned vector data, the vector data is input into the target model, and the target model processes the vector data to generate output data. The output data does not include data corresponding to the restriction condition.

[0118] Exemplarily, the vector data is the call text after cleaning the location information. The vector data is input into the target model to obtain a summary corresponding to the call text, which does not include the location information.

[0119] In the embodiment of the present application, the input data is processed based on the description data to obtain vector data for inputting into the target model, and the processing of the input data based on the description data is used to clean up the data satisfying the restriction condition from the input data; the vector data is processed based on the target model to generate the output data. In this way, the data corresponding to the restriction condition in the input data can be cleaned up, and the output data output by the model does not include the data corresponding to the restriction condition, thereby effectively restricting the output result of the model.

[0120] Figure 5 This is a schematic diagram of the implementation process of a data processing method provided in an embodiment of the present application. Figure 5 , the method can be executed by a processor of a computer device. Figure 1 , Figure 1 S103 in can be updated to S501 to S503, which will be combined Figure 5 The steps shown are explained.

[0121] Step S501: Process the input data into first vector data.

[0122] After the input data is obtained, the input data is vectorized to obtain first vector data.

[0123] Exemplarily, the input data is speech, the speech is first converted into text, and then the text is vectorized to obtain first vector data.

[0124] Step S502: Process the description data into second vector data.

[0125] After the description data is obtained, the description data is vectorized to obtain second vector data.

[0126] In some embodiments, after obtaining the description data, the description data is parsed to obtain the parsed text, and then the text is vectorized to obtain the second vector data.

[0127] Exemplarily, the description data is text, and the text is directly parsed and vectorized to obtain second vector data.

[0128] Step S503, processing the first vector data and the second vector data based on the target model to generate the output data, wherein the target model clears the vector data that meets the restriction condition in the process of generating the output data, and the restriction condition is used to limit the processing process of the target model.

[0129] After obtaining the first vector data corresponding to the input data and the second vector data corresponding to the description data, the first vector data and the second vector data are input into the target model, and the target model processes the second vector data based on the first vector data to obtain output data. In the process of generating output data, the target model cleans up the vector data that meets the restriction conditions, and the restriction conditions are used to limit the processing process of the target model.

[0130] In some embodiments, the target model first processes the first vector data to obtain intermediate data, and then cleans the vector data corresponding to the restriction condition from the intermediate data based on the second vector data to obtain output data.

[0131] In some embodiments, the target model first cleans the vector data corresponding to the restriction condition from the first vector data based on the second vector data, and then processes the cleaned first vector data to obtain output data.

[0132] In the embodiment of the present application, the input data to be processed is the first vector data, the description data to be processed is the second vector data, the first vector data and the second vector data are processed based on the target model to generate output data, and the target model removes the vector data that meets the restriction condition in the process of generating the output data, and the restriction condition is used to restrict the processing process of the target model. In this way, the generated output data does not include the data corresponding to the restriction condition, and the output result of the model can be effectively restricted.

[0133] In some embodiments, the target model is trained based on a large number of training samples, and the training samples include input data samples, description data samples, and output data samples. Each input data sample corresponds to a description data sample and an output data sample one by one. Based on multiple groups of input data samples, description data samples, and output data samples, the target model is trained to obtain a trained target model.

[0134] In some embodiments, a large number of input data samples and description data samples are constructed, and based on the description data samples, the input data samples are processed to obtain a large number of processed input data samples. A target model is trained based on the large number of processed input data samples and output data samples to obtain a trained target model.

[0135] The following describes the application of the data processing method provided in the embodiment of the present application in a practical scenario.

[0136] In the related technology of artificial intelligence AI big language model call assistant, the voices of both parties in the call will be converted into text through automatic speech recognition technology (Automatic Speech Recognition, ASR), and the text content will be directly summarized and recorded by the big model. In this way, both parties use the big model separately or simultaneously, and the other party is unaware and cannot control the behavior of the other party's big model, which greatly brings risks to the other party's privacy, business and even personal safety.

[0137] Figure 6 A summary of the content of a conversation between two parties proposed in the embodiment of the present application Figure 1 .like Figure 6 As shown, the caller 601 and the called party 602 have a voice call. In the voice call interface of the caller's device, there is an AI call assistant 603. The AI ​​call assistant will record the voices of both parties and convert them into text. The text is input into the big model to obtain a summary of the voice content, and the summary can be further added to the notes. However, the content summary output by the AI ​​call assistant includes some sensitive privacy information, such as mobile phone numbers.

[0138] In the embodiment of the present application, the call assistant of the AI ​​big model is duplex controlled. If any party has an objection, the reasoning and summarizing ability of the other party's AI model needs to be restricted or controlled; the other party's reply will automatically be used as part of the input content of the big model, and the model will be informed of the input cleaning and encoding methods in the form of instructions, thereby assisting the big model's final reasoning output to be output in the direction confirmed by the user.

[0139] In some embodiments, in order to prevent information leakage in the related art, privacy-sensitive information is usually blurred, which can easily lead to information incoherence. For example, blurring part of the content in a summary will cause the summary content to be incoherent. The embodiment of the present application processes the input data and restrictions based on the model, and the output data does not contain the target data corresponding to the restrictions. In this way, the output content of the model is coherent and does not contain privacy information. The content output by the model is controlled, and the output content is safe whether it is shared with a third-party user or actively obtained by a third-party user. Compared with the related art, the embodiment of the present application can input restriction information into the model, and the restriction information can come from any user, rather than the restriction information set by the device manufacturer when the device leaves the factory.

[0140] Figure 7 A summary of the content of a conversation between two parties proposed in the embodiment of the present application Figure 2 .like Figure 7As shown, when the two parties are having a voice call, the called party 602 can send an instruction to summarize the call content via SMS, for example, the instruction can be "This call contains sensitive content, please do not summarize it by AI". After receiving the SMS content, the AI ​​call assistant 603 of the calling party 601 device parses the SMS content and determines that the content will not be summarized. A prompt message pops up on the AI ​​call assistant interface, indicating that no extractable information was found from the call.

[0141] In some embodiments, the steps of implementing the protocol include:

[0142] Step 1: The call between the two parties begins, and the party who needs to be summarized automatically activates the AI ​​call assistant on his mobile terminal device.

[0143] Step 2: The activated AI call assistant will use ASR to record the voice and convert it into text during the call, and finally vectorize and save the record. The vectorized call content includes the text information of the caller and the other party. For example, the vector content is<caller:voice text> and<callee:voice text> ,<caller:voice text> Represents the voice text of the calling party,<callee:voice text> Represents the voice text of the called party.

[0144] Step 3. After the call ends, the AI ​​call assistant will automatically send a text message to notify the other party and prepare to make an inference and summary of the call content.

[0145] Step 4: Ensure that you get a response from the other party within 5 minutes, including not allowing summary, allowing partial content summary (referring to the call assistant), and allowing full content summary. The response is sent via SMS.

[0146] Step 5: The AI ​​call assistant parses the content of the received SMS and vectorizes the SMS content.

[0147] Step 6: embed the received SMS encoded content and the previously recorded ASR vector (text vector corresponding to the speech); embedding coding means embedding the vectorized SMS content into the text vector of the speech so that the object indicated by the SMS content can match the object of the speech text. For example, the SMS instruction vector for the caller's speech will be embedded into the caller's speech text vector, and the SMS instruction vector for the called party will be embedded into the called party's speech text vector.

[0148] Step 7: Integrate the ASR call content and the text message content received to extract the final large model input information; after completing the embedded coding, the embedded coding content can be further integrated, for example, the embedded coding can be adjusted to a preset coding format, or for text vectors that cannot be summarized, directly hide or delete them to obtain the input information of the large model that needs further cleaning and filtering.

[0149] Step 8: Clean and filter the quantized content (large model input information in the previous step) according to the instruction information, for example, filter out sensitive information.

[0150] Step 9: Input the final cleaned content into the large model for reasoning and summary.

[0151] Figure 8 This is a schematic diagram of the data processing flow when two parties are talking in a conversation according to the embodiment of the present application. Figure 8 As shown, it includes steps S801 to S812:

[0152] Step S801, start a call and trigger the AI ​​call assistant;

[0153] Step S802, ASR data vectorization;

[0154] After recording the voice, the voice is converted into text and vectorized to obtain a text vector.<caller:voice text> , called direction quantity<callee:voice text> .

[0155] Step S803: After the call is over, a text message is automatically sent to notify the called party.

[0156] The called party is informed via text message to make an inference and summary.

[0157] Step S804: Determine whether a response is received within 5 minutes. If yes, go to step S805; if no, go to step S812.

[0158] Step S805: parse and vectorize the response.

[0159] Among them, it is necessary to parse the other party's response to identify the content of the response and vectorize the parsed content.

[0160] Step S806: embed the response vector into the ASR vector;

[0161] Step S807: Determine whether it is allowed to input the ASR vector into the large language model. If yes, go to step S808; if not, go to step S812.

[0162] Step S808: extract the final large model input data.

[0163] Among them, the final large model input data includes instructions (response SMS content) and voice text (content to be summarized).

[0164] Step S809: clean and filter the input data of the large model.

[0165] Among them, the voice text is cleaned and filtered based on the instructions.

[0166] Step S810, determine whether the cleaning and filtering is successful. If yes, go to step S811; if not, go to step S812.

[0167] Step S811: input the cleaned data into the large language model.

[0168] Step S812, end.

[0169] In some embodiments, the input content that the large model may receive includes three situations: the first situation is that there is no input and the called party does not allow the large model to infer and summarize; the second situation is that there is only the content of the calling party: {called party SMS instruction: calling party voice text}; the third situation includes the content of both parties of the call: {called party SMS instruction: calling party voice text and called party voice text}.

[0170] In some embodiments, the SMS command of the called party will be used as part of the large language model to clean the input content, telling the large model how and to what extent to adjust or filter the content when exercising voice content inference and summary, so as to prevent the large model from arbitrarily summarizing the output; for example: if the reply SMS received contains the command "remove the mobile phone number and location information in my call content", the large model will clean and filter out the privacy-sensitive content in the called party.

[0171] In the embodiment of the present application, the AI ​​big model behavior of the two parties in the call can be judged and agreed by the other party, and the replies of both parties are used as part of the big model input, effectively managing the behavior of the big model; enabling the big model to act on demand and prevent malicious output; and effectively protecting the interests of both parties in the call, further preventing the leakage of user privacy or business property information from the source.

[0172] Based on the foregoing embodiments, an embodiment of the present application provides a data processing device, which includes the units included and the modules included in the units, which can be implemented by a processor in a computer device; of course, it can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0173] Fig. 9 A schematic diagram of the structure of a data processing device provided in an embodiment of the present application is shown in FIG. Fig. 9 As shown, the data processing device 900 includes: a first acquisition module 910, a second acquisition module 920, a processing module 930 and an output module 940, wherein:

[0174] A first acquisition module 910, used to obtain input data;

[0175] A second acquisition module 920 is used to obtain description data for characterizing the restriction condition, wherein the input data includes target data corresponding to the restriction condition;

[0176] A processing module 930 is used to process the input data to generate output data based on a target model, the input data and the description data, wherein the target model is a generative large language model;

[0177] The output module 940 is used to output the output data, where the output data does not include the target data corresponding to the restriction condition.

[0178] In some embodiments, the first acquisition module 910 is further used to obtain information input from at least two objects to constitute the input data; the second acquisition module 920 is further used to obtain restriction conditions of the input from at least one of the at least two objects to constitute the description data.

[0179] In some embodiments, the restriction condition from the first-party object is applied to a portion of the input data belonging to the first-party object.

[0180] In some embodiments, the first acquisition module 910 is also used to establish a communication channel with at least one object; obtain information input of the own object and information input of the other object to constitute the input data; the other object belongs to the at least one object; the second acquisition module 920 is also used to analyze the input data and determine the restriction conditions of the input of at least one of the own object and the other object.

[0181] In some embodiments, the second acquisition module 920 is also used to obtain the restriction conditions input by the other party if the communication channel with at least one party is successfully established; to obtain the restriction conditions input by the other party if the communication channel with at least one party is disconnected; and to obtain the restriction conditions input by the other party based on the process of communicating through the established communication channel with at least one party.

[0182] In some embodiments, the second acquisition module 920 is further used to obtain description data for characterizing restriction conditions in response to calling the target model.

[0183] In some embodiments, the processing module 930 is also used to process the input data based on the description data to obtain vector data for inputting the target model, wherein the processing of the input data based on the description data is used to clean up data that meets the restriction conditions from the input data; and the processing of the vector data based on the target model generates the output data.

[0184] In some embodiments, the processing module 930 is also used to process the input data into first vector data; process the description data into second vector data; process the first vector data and the second vector data based on the target model to generate the output data, and the target model clears the vector data that meets the restriction condition in the process of generating the output data, and the restriction condition is used to limit the processing process of the target model.

[0185] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiment of the present application can be used to execute the method described in the above method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.

[0186] It should be noted that in the embodiment of the present application, if the above-mentioned data processing method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific hardware, software or firmware, or any combination of hardware, software, and firmware.

[0187] An embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0188] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, some or all of the steps in the above method are implemented. The computer-readable storage medium can be transient or non-transient.

[0189] An embodiment of the present application provides a computer program, including a computer-readable code. When the computer-readable code is run in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0190] The embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, some or all of the steps in the above method are implemented. The computer program product can be implemented specifically by hardware, software or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium, and in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK) and the like.

[0191] It should be noted here that the description of the various embodiments above tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. The description of the above device, storage medium, computer program and computer program product embodiments is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiments of the device, storage medium, computer program and computer program product of this application, please refer to the description of the method embodiment of this application for understanding.

[0192] Fig.10 A hardware entity diagram of an electronic device provided in an embodiment of the present application, such as Fig.10 As shown, the electronic device 1000 includes a first input component 1010, a second input component 1020, a processor 1030 and an output component 1040;

[0193] The first input component 1010 is used to obtain input data;

[0194] The second input component 1020 is used to obtain description data for characterizing the restriction condition, wherein the input data includes target data corresponding to the restriction condition;

[0195] The processor 1030 is used to process the input data to generate output data based on a target model, the input data and the description data, wherein the target model is a generative large language model;

[0196] The output component 1040 is used to output the output data, wherein the output data does not include target data corresponding to the restriction condition.

[0197] The electronic device may be any device capable of executing the data processing method, for example, a smart phone, a tablet, a computer, or the like.

[0198] The input data is the data that needs to be processed. The description data is the data about the restriction conditions. The restriction conditions correspond to the target data in the input data. The target data can be the data that is not allowed / needed to be processed in the input data. The final output result can be limited by the description data.

[0199] In some embodiments, the input data and the description data are obtained through different input components. Exemplarily, in a scenario where at least two objects are conducting a voice call, the local object establishes a voice channel with the other object and conducts a voice call based on the voice channel. The first input component obtains the voice of the local object and the other object in the voice channel to obtain the input data. The other object proposes descriptive data for characterizing the restriction to the local object. For example, the other object sends the descriptive data to the local object in the form of a text message. At this time, the descriptive data for characterizing the restriction needs to be obtained through the second input component. The first input component can be an audio analysis component in an electronic device, which can analyze the voices of both parties in the call to obtain input data; the second input component can be a data analysis component in an electronic device, which can analyze the text messages sent by the other object to obtain descriptive data.

[0200] In some embodiments, in a scenario where at least two objects are conducting a voice call, the first input component obtains the voice of the local object and the other object in the voice channel to obtain input data. The local object proposes descriptive data for characterizing the restriction condition. For example, the local object chooses whether to restrict the voice through a prompt box that pops up on the touch screen of the electronic device. At this time, the descriptive data of the local object is obtained through the second input component, and the second input component can be a touch screen. For another example, the electronic device is a computer, and the local object determines the descriptive data through the input of the computer's mouse or keyboard. At this time, the second input component can be a component that responds to the mouse or keyboard input.

[0201] In some embodiments, in a scenario where at least two parties are conducting a voice call, the call content between the two parties includes descriptive data used to characterize the restriction conditions. For example, the other party directly states in the call that the identity information in the voice call cannot be processed. At this time, the input data and the descriptive data both come from the voice call, and only one input component is required to obtain the input data and the descriptive data used to characterize the restriction conditions.

[0202] In some embodiments, the input data and the description data come from the same user, the input data may be audio, video or text saved by the user, and the description data may be set by the user for the audio, video or text to be processed. For example, the input data and the description data are both a segment of audio, in which case the input data and the description data are obtained through the same input component, and the input component may be an audio parsing component. For another example, the input data is audio and the description data is text, in which case the first input component obtains the input data and the second input component obtains the description data, the first input component is an audio parsing component, and the second input component is a text parsing component.

[0203] In some embodiments, the processor is in communication with the first input component and the second input component. After the first input component obtains input data and the second input component obtains description data, the processor can process the input data based on the target model, the input data, and the description data to generate output data. The target model is a generative large language model, and the generative large language model can summarize the input data to obtain a summary corresponding to the input data. Exemplarily, the processor calls the target model, inputs the input data and the description data into the target model, runs the target model, and obtains output data.

[0204] In some embodiments, after the processor processes the input data, the output component can output the output data, the output data being the data after the input data is finally processed, and the output data does not include the target data corresponding to the restriction condition. Exemplarily, the output data is a summary text, and the output component can be a text generation component in the electronic device.

[0205] In some embodiments, the electronic device further includes a memory, wherein the memory is used to store the target model; the processor 1030 is further used to load and execute the target model from the memory.

[0206] The target model is stored in a memory of the electronic device, and the processor can access the memory, load the target model from the memory, load the target model into the internal memory, and execute the target model.

[0207] In some embodiments, the first input component 1010 is further used to obtain information input from at least two objects to constitute the input data; the second input component 1020 is further used to obtain restriction conditions on the input of at least one of the at least two objects to constitute the description data.

[0208] In some embodiments, the restriction condition from the first-party object is applied to a portion of the input data belonging to the first-party object.

[0209] In some embodiments, the first input component 1010 is also used to establish a communication channel with at least one object; obtain information input of the own object and information input of the other object to constitute the input data; the other object belongs to the at least one object; analyze the input data to determine the restriction conditions of the input of at least one of the own object and the other object.

[0210] In some embodiments, the first input component 1010 is further used to obtain the restriction conditions of the input of the other party if the communication channel with at least one object is successfully established; to obtain the restriction conditions of the input of the other party if the communication channel with at least one object is disconnected; and to obtain the restriction conditions of the input of the other party based on the process of communicating through the established communication channel with at least one object.

[0211] In some embodiments, the second input component 1020 is further used to obtain description data for characterizing restriction conditions in response to calling the target model.

[0212] In some embodiments, the processor 1030 is further used to process the input data based on the description data to obtain vector data for inputting the target model, wherein the processing of the input data based on the description data is used to clean up data that meets the restriction conditions from the input data; and the vector data is processed based on the target model to generate the output data.

[0213] In some embodiments, the processor 1030 is also used to process the input data into first vector data; process the description data into second vector data; process the first vector data and the second vector data based on the target model to generate the output data, and the target model clears the vector data that meets the restriction condition in the process of generating the output data, and the restriction condition is used to limit the processing process of the target model.

[0214] Fig.11 A hardware entity diagram of a computer device provided in an embodiment of the present application is shown in FIG. Fig.11 As shown, the hardware entity of the computer device 1100 includes: a processor 1101 and a memory 1102, wherein the memory 1102 stores a computer program that can be run on the processor 1101, and the processor 1101 implements the steps in the method of any of the above embodiments when executing the program.

[0215] The memory 1102 stores computer programs that can be run on the processor. The memory 1102 is configured to store instructions and applications executable by the processor 1101. It can also cache data to be processed or processed by the processor 1101 and various modules in the computer device 1100 (for example, image data, audio data, voice communication data, and video communication data). This can be achieved through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0216] When the processor 1101 executes the program, the steps of any of the above data processing methods are implemented. The processor 1101 generally controls the overall operation of the computer device 1100.

[0217] An embodiment of the present application provides a computer storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps of the data processing method of any of the above embodiments.

[0218] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0219] The processor may be at least one of an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a Central Processing Unit (CPU), a controller, a microcontroller, and a microprocessor. It is understandable that the electronic device that implements the functions of the processor may also be other, and the embodiments of the present application are not specifically limited.

[0220] The above-mentioned computer storage medium / memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM) and the like; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0221] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A data processing method, comprising: Get input data; Obtaining description data for characterizing a restriction condition, wherein the input data includes target data corresponding to the restriction condition; Based on a target model, the input data and the description data, processing the input data to generate output data, wherein the target model is a generative large language model; The output data is output, wherein the output data does not include the target data corresponding to the restriction condition.

2. The method according to claim 1, wherein obtaining input data comprises: Obtaining information input from at least two objects to form the input data; The obtaining of description data for characterizing the restriction condition includes: Obtaining the restriction condition of at least one of the at least two objects input to form the description data.

3. The method according to claim 2, further comprising: The restriction condition from the first-party object acts on a portion of the input data that belongs to the first-party object.

4. The method according to claim 3, wherein obtaining information input from at least two objects to form the input data comprises: Establishing a communication channel with at least one object; Obtaining information input of the own object and information input of the other party object to form the input data; The other party object belongs to the at least one party object; The restriction condition for obtaining the input of at least one of the at least two objects, constituting the description data, includes: The input data is analyzed to determine input restriction conditions of at least one of the own object and the other object.

5. The method according to claim 3, wherein the step of obtaining information input from at least two objects and forming the input data comprises: Establishing a communication channel with at least one object; Obtaining information input of the own object and information input of the other party object to form the input data; The other party object belongs to the at least one party object; The restriction condition for obtaining the input of at least one of the at least two objects, constituting the description data, includes at least one of the following: If the communication channel with at least one object is successfully established, obtaining the restriction condition input by the other object; If the communication channel with at least one object is disconnected, obtaining the restriction condition input by the other object; In the process of establishing a communication channel with at least one object for communication, a restriction condition input by the other object is obtained.

6. The method according to claim 1, wherein obtaining description data for characterizing the restriction condition further comprises: In response to calling the target model, description data for characterizing the constraint condition is obtained.

7. The method according to claim 1 or 6, wherein the step of processing the input data to generate output data based on the target model, the input data and the description data comprises: Processing the input data based on the description data to obtain vector data for inputting into the target model, wherein the processing of the input data based on the description data is used to clean up data satisfying the restriction condition from the input data; The vector data is processed based on the target model to generate the output data.

8. The method according to claim 1 or 6, wherein the step of processing the input data to generate output data based on the target model, the input data and the description data comprises: Processing the input data into first vector data; Processing the description data into second vector data; The first vector data and the second vector data are processed based on the target model to generate the output data. The target model clears the vector data that meets the restriction condition in the process of generating the output data. The restriction condition is used to limit the processing process of the target model.

9. An electronic device, comprising: a first input component, the first input component being used to obtain input data; A second input component, the second input component is used to obtain description data for characterizing the restriction condition, wherein the input data includes target data corresponding to the restriction condition; A processor, the processor being configured to process the input data to generate output data based on a target model, the input data, and the description data, wherein the target model is a generative large language model; An output component is used to output the output data, wherein the output data does not include target data corresponding to the restriction condition.

10. The electronic device according to claim 9, further comprising a memory for storing the target model; The processor is further configured to load and execute the target model from the memory.

Images