DATA PROCESSING METHOD AND ELECTRONIC DEVICE

Abstract

A data processing procedure includes obtaining input data, obtaining descriptive data that constitutes a limiting condition, processing the input data based on a target model, the input data, and the descriptive data to generate output data, and outputting the output data. The input data includes target data that meets the limiting condition. The target model is a generative large language model. The output data does not include the target data that meets the limiting condition.

Description

REFERENCE TO RELATED REGISTRATION

[0001] This application claims priority over Chinese patent application No. 202411808018.4, filed on December 9, 2024, the entire contents of which are hereby incorporated by reference. TECHNICAL AREA

[0002] The present disclosure relates generally to the technical field of data processing and in particular to a data processing method and an electronic device. TECHNICAL BACKGROUND

[0003] When processing input data based on a model, the prior art model typically processes all input data according to a specific processing logic to obtain a processing result. Therefore, neither the model's behavior during input data processing nor the model's output result is controllable. In various implementations, it is often necessary to restrict the model's output result to meet a specific requirement. OVERVIEW OF THE INVENTION

[0004] Embodiments of the present disclosure provide a data processing method. The method comprises obtaining input data, obtaining descriptive data that constitutes a limiting condition, processing the input data based on a target model, the input data, and the descriptive data to generate output data, and outputting the output data. The input data includes target data that conforms to the limiting condition. The target model is a generative large language model. The output data does not include the target data that conforms to the limiting condition.

[0005] Embodiments of the present disclosure provide an electronic device comprising a first input assembly, a second input assembly, one or more processors, and an output assembly. The first input assembly is configured to acquire input data. The second input assembly is configured to acquire descriptive data that constitutes a limiting condition. The input data includes target data that conforms to the limiting condition. The one or more processors are configured to process the input data, based on a target model, the input data, and the descriptive data, to generate output data. The target model is a generative large language model. The output assembly is configured to output the output data. The output data does not include the target data conforming to the limiting condition.

[0006] Embodiments of the present disclosure provide a non-volatile, computer-readable storage medium on which a computer program is stored. When executed by one or more processors, this program causes the one or more processors to obtain input data, obtain descriptive data constituting a limiting condition, process the input data based on a target model, and output the output data to generate output data. The input data includes target data that conforms to the limiting condition. The target model is a generative large language model. The output data does not include the target data conforming to the limiting condition. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 illustrates a schematic flowchart of a data processing procedure according to embodiments of the present disclosure. Fig. Figure 2 illustrates a schematic flowchart of a data processing procedure according to embodiments of the present disclosure. Fig. Figure 3 illustrates a schematic flowchart of a data processing procedure according to embodiments of the present disclosure. Fig. Figure 4 illustrates a schematic flowchart of a data processing procedure according to embodiments of the present disclosure. Fig. Figure 5 illustrates a schematic flowchart of a data processing procedure according to embodiments of the present disclosure. Fig. Figure 6 illustrates a schematic diagram of a summary of content when two parties are involved in a call, according to embodiments of the present disclosure. Fig. Figure 7 illustrates a schematic diagram of a summary of content when two parties are involved in a call, according to embodiments of the present disclosure. Fig. Figure 8 illustrates a schematic flowchart of a data processing procedure when two parties are involved in a call, according to embodiments of the present disclosure. Fig. Figure 9 illustrates a schematic structure diagram of a data processing device according to embodiments of the present disclosure. Fig. Figure 10 illustrates a schematic structure diagram of an electronic device according to embodiments of the present disclosure. Fig. Figure 11 illustrates a schematic diagram of a computer device according to embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EXECUTION FORMS

[0007] To clarify the objectives, technical solutions, and advantages of this disclosure, the technical solutions of this disclosure are described in detail below in conjunction with the accompanying drawings and embodiments. The described embodiments should not be considered a limitation of this disclosure. All other embodiments that can be obtained by a person skilled in the art without creative effort are within the scope of this disclosure.

[0008] In the following description, “some embodiments” refers to a subset of all possible embodiments. “Some embodiments” may refer to the same or different subsets of all possible embodiments, and they may be combined with one another, provided there is no contradiction. The terms “first,” “second,” “third,” etc., are used to distinguish similar objects and do not represent a specific order. “First,” “second,” and “third” in a specific sequence or order may be interchanged to implement the embodiments described in this disclosure in sequences different from the one illustrated or described herein.

[0009] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as generally understood by a person skilled in the art. The terms used herein serve only to describe the present disclosure and not to limit it.

[0010] In the present disclosure, a “target model” can be a machine learning model, a large language model, or a generative large language model.

[0011] A machine learning model can recognize natural language and / or other inputs (such as audio, video, images, tables, etc.) fed into the target model and perform comprehensive language processing tasks such as semantic analysis and question answering, thereby producing an output that relates to and / or answers the input.

[0012] A large language model is an artificial intelligence model based on deep learning techniques. It can learn the properties, rules, and patterns of natural language by training it on a large amount of diverse data to understand and generate natural language texts. A large language model typically has billions to hundreds of billions of parameters, enabling it to grasp complex relationships and patterns in natural language.

[0013] A generative large language model is primarily designed to generate natural language text. From a given input, such as a sentence or prompt, a generative large language model can produce a response—coherent and contextually relevant sentences or even paragraphs based on that input or prompt. The model uses various techniques, including attention mechanisms, transformers, and neural networks, to process the input and generate output that aims to be coherent and contextually appropriate.

[0014] Generative large language models can learn the statistical rules of a language by analyzing large amounts of text data and, based on this learned knowledge and the provided context or prompts, generate text that conforms to grammatical and semantic rules. Generative large language models can improve various applications such as content generation, personalization, and multimodal content generation.

[0015] A generative large language model (GLM) can, for example, include a large language model (LLM), a generative pre-trained transformer (GPT), a visual large model, a multimodal large model, or an expert large model, which are obtained through fine-tuning based on specific needs. The present disclosure is not limited to these examples.

[0016] The output of the target model is not restricted. For example, if personal user data or security-relevant data is input into the target model, the input is processed based on a processing logic determined during training, which can lead to limited controllability of the model's output. This can negatively affect the interaction between the user and the target model.

[0017] To limit the processing result when processing the input data, embodiments of the present disclosure provide a data processing method. The method includes obtaining the input data, obtaining descriptive data that constitutes a limiting condition, processing the input data to generate output data based on the target model, the input data, and the descriptive data, and outputting the output data. The input data may include target data that meets the limiting condition. The target model may be the generative large language model. The output data may not include the target data that meets the limiting condition. During the processing of the input data, the output data can then be limited by the descriptive data of the limiting condition to ensure that the output data of the target model does not include the target data that meets the limiting condition.This allows the output data to be effectively restricted, preventing the target model from producing output and effectively controlling its output behavior. The output data can then meet the requirements.

[0018] Embodiments of the present disclosure provide a data processing method that can be executed by a processor of a computer device. The computer device may be a server, a laptop, a tablet, a desktop computer, a smart TV, a set-top box, a mobile device (e.g., a mobile phone, a portable video player, a personal digital assistant, a dedicated messaging device, and a portable gaming device), or any device with data processing capabilities.

[0019] Fig. Figure 1 illustrates a schematic flowchart of a data processing procedure according to embodiments of the present disclosure. As in Fig. As shown in Figure 1, the procedure includes the following processes.

[0020] The input data is obtained in S101.

[0021] The input data can be the data to be processed.

[0022] In some embodiments, if at least two users are in a voice call, a call assistant on a user's terminal device can record each user's voice audio during the call. The voice audio from all users can then be used as input data for processing the voice audio content.

[0023] If at least two users are in a video call, the user's device's call assistant can, in some implementations, record video call content. The video from all users can then be used as input data for processing the video content.

[0024] If at least two users are in an online meeting, the user's device's call assistant can, in some implementations, record the meeting's voice audio. The voice audio of all users can then be used as input data for processing the meeting's voice audio.

[0025] In some embodiments, the input data from at least one user can be specific data. This data can include speech audio, video, or text that has been pre-saved by the user. For example, the user can pre-save a segment of speech audio as the input data.

[0026] In S102, the descriptive data representing the limiting condition is obtained, and the input data includes the target data that meets the limiting condition.

[0027] The descriptive data can be data about the limiting condition. The limiting condition can correspond to the target data of the input data. The target data can be data from the input data whose processing is not permitted or is unnecessary. The output result can be limited by the descriptive data.

[0028] If at least two users are in a voice call, in some implementations each of the at least two users can propose a limiting condition regarding the processing of the call voice audio. The limiting condition can be used as the descriptive data.

[0029] If at least two users are in the video call, in some implementations each of the at least two users can propose a limiting condition regarding the processing of the video. The limiting condition can be used as the descriptive data.

[0030] If at least two users are in the online meeting, in some implementations either of the at least two users can propose the limiting condition regarding the processing of the meeting's speech audio. The limiting condition can be used as the descriptive data.

[0031] In some embodiments, the descriptive data can be specific data from at least one user. The descriptive data can be predefined by the user for the input data. For example, if the user wants to summarize a stored speech audio segment, the descriptive data can specify the content of the speech audio that must not be summarized.

[0032] If the input data is the call voice audio of at least two users, the descriptive data may include, for example, names and location information from the call voice audio, which must not be aggregated.

[0033] In S103, the input data is processed based on the target model, the input data, and the descriptive data to generate the output data. The target model is a generative large language model.

[0034] After obtaining the input data and the descriptive data of the limiting condition, the input data can be processed based on the target model, the input data and the descriptive data to generate the output data.

[0035] In some embodiments, the target model can be the generative large language model. The generative large language model can process the input data to generate a summary, thus obtaining a summary corresponding to the input data. For example, a text segment can be input into the generative large language model, and the generative large language model can output (generate) the summary corresponding to the text.

[0036] In some embodiments, the target model can be a different neural network model, such as a convolutional neural network, recurrent neural network, etc.

[0037] In some embodiments, the target data can be determined from the input data based on the descriptive data. The input data can then be cleaned to remove the target data and obtain the cleaned input data. This cleaned input data can then be fed into the target model to generate the output data.

[0038] In some embodiments, the input data and the descriptive data can be entered into the target model together. The target model can process the input data based on the descriptive data to generate the output data. The target model can process the input data in any way, as long as it is ensured that the output data does not include the target data that meets the limiting condition.

[0039] Page S104 displays the output data. The output data does not include the target data that meets the restrictive condition.

[0040] After the output data has been generated in the previous step, it may be necessary to output the output data. The output data can be the processed data from the input data and may not include the target data that meets the restrictive condition.

[0041] In one example, the input data could be the voice audio from User 1 and User 2. The generative large language model can summarize the voice audio content. The descriptive data can specify that the phone number and location information in User 1's voice audio should not be summarized. After processing the input voice audio, the output data must not include the phone number and location information in User 1's voice audio.

[0042] In embodiments of the present disclosure, the input data can be obtained. The descriptive data representing the limiting condition can be obtained. The input data can include the target data that conforms to the limiting condition. Based on the target model, the input data, and the descriptive data, the input data can be processed to generate the output data. The target model can be the generative large language model. The output data can be output, and the output data can exclude the target data conforming to the limiting condition. During the processing of the input data, the output data can then be restricted by the descriptive data of the limiting condition to ensure that the output data of the target model does not include the target data conforming to the limiting condition.Therefore, the output data can be effectively restricted, preventing the target model from producing an arbitrary result and effectively controlling the output behavior of the target model. The output data can then meet the requirement.

[0043] Fig. Figure 2 illustrates a schematic flowchart of a data processing method according to embodiments of the present disclosure. The method can be executed by the processor of the computer device. Starting from Fig. 1 are S101 and S102 in Fig. 1 to S201 and S202 have been updated. As in Fig. As shown in section 2, the procedure includes the following processes.

[0044] In S201, information input is obtained from at least two participating objects in order to form the input data.

[0045] If at least two objects are involved in a voice call, the intelligent device, in some embodiments, can record the call audio of the at least two participating objects during the call in order to obtain a call audio recording of each of the at least two participating objects. The call audio can be used as the object's information input to form the input data.

[0046] If at least two objects are involved in a video call, the intelligent device, in some embodiments, can record the video of the at least two participating objects during the call in order to obtain the video of each of the at least two participating objects. This video can then be used as the object's information input to form the input data.

[0047] If at least two objects are participating in an online meeting, the intelligent device, in some embodiments, can record the speech audio of the at least two participating objects during the meeting in order to obtain the speech audio of each of the at least two participating objects. The speech audio can be used as the object's information input to form the input data.

[0048] In some embodiments, the at least two participating objects can be objects of the same device. The input data interface of this device can provide at least two input fields. Each input field can correspond to a participating object. Each object can input the information to be processed into the corresponding input field to convert the information input from the at least two participating objects into input data. The information input by different objects can be differentiated.

[0049] In some embodiments, the at least two participating objects can be objects of the same device. The input data interface of this device can provide a single input field. Each participating object can enter the information to be processed into the input field. The information can be marked for differentiation. The information input from the at least two participating objects can constitute the input data.

[0050] In S202, the restrictive condition entered by at least one participating object from among at least two participating objects is obtained to form the descriptive data.

[0051] At least one of the at least two participating objects can enter the restrictive condition. The entered restrictive condition can form the descriptive data.

[0052] In some embodiments, the at least two participating objects can be the at least two objects in the voice call. During the voice call, at least one participating object can restrict the call voice audio and enter the restricting condition. The entered restricting condition can form the descriptive data.

[0053] In some embodiments, the at least two participating objects can be the at least two objects in a video call. During the video call, at least one participating object can restrict the video content and enter the restrictive condition. The entered restrictive condition can constitute the descriptive data.

[0054] In some embodiments, the at least two participating objects can be at least two objects in an online meeting. During the meeting, at least one participating object can restrict the meeting's voice audio and enter the restricting condition. The restricting condition can form the descriptive data.

[0055] In some embodiments, the restrictive condition entered by the at least one participating object can restrict the information input of all participating objects. For example, if User 1 and User 2 are in a voice call, User 1 can propose the restrictive condition to limit the location information in the voice audio of User 1 and User 2 (i.e., that the output data does not include the location information in the voice audio of User 1 and User 2).

[0056] In some embodiments, the restrictive condition introduced by at least one participating object can restrict only the information input of that object. For example, if User 1 and User 2 are in a voice call, User 1 can propose the restrictive condition. The restrictive condition can only be used to restrict the location information in User 1's voice audio (i.e., the output data does not include the location information in User 1's voice audio), but it cannot be used to restrict the location information in User 2's voice audio (i.e., the output data includes the location information in User 2's voice audio).

[0057] In one example, in a two-party voice call scenario, the call audio of each participating user can be used as the information input to generate the input data. The restrictive condition entered by a user for the call audio can form the descriptive data.

[0058] In embodiments of the present disclosure, the input information from the at least two participating objects can be obtained to form the input data. The limiting condition input by the at least one participating object from the at least two participating objects can be obtained. The limiting condition can form the descriptive data. Thus, the input data and the descriptive data can be obtained from the at least two participating objects to improve the richness and variety of the input data and the descriptive data.

[0059] In some embodiments, the restrictive condition can be applied from a first participating object to sub-data of the input data that belong to the first participating object.

[0060] The first participating object can be any of the at least two participating objects.

[0061] For the at least two participating objects, each participating object can add corresponding information input. The input data can include sub-data of each participating object. The restrictive condition of the first participating object can only be applied to the sub-data of the input data belonging to the first participating object and cannot be applied to the sub-data of other objects.

[0062] For example, if User 1 and User 2 are in a voice call, the input data can include the voice audio of User 1 and the voice audio of User 2. User 1 can propose the restrictive condition. The restrictive condition can only be applied to the voice audio of User 1 and not to the voice audio of User 2.

[0063] In embodiments of the present disclosure, the restrictive condition can be applied by the first participating object to the sub-data of the input data belonging to that first participating object. The restrictive condition can then be applied only to the sub-data of the input data belonging to the same object, thus avoiding the effect of the restrictive condition on objects other than the object to which the restrictive condition applies. This improves the separation of authority between different users, thereby enhancing the targeting accuracy of the restrictive condition.

[0064] Fig. Figure 3 illustrates a schematic flowchart of a data processing method according to embodiments of the present disclosure. The method can be executed by the processor of the computer device. Starting from Fig. 2 is S201 in Fig. 2 to S301 and S302 have been updated and S202 in Fig. 2 has been updated to S303. As in Fig. As shown in section 3, the procedure includes the following processes.

[0065] In S301, a communication channel is established to at least one participating object.

[0066] At least two participating objects can exchange information. Input can be obtained via the established communication channel to form the input data. It may be necessary for the participating object present to establish the communication channel with the other participating object. Thus, the participating object present can communicate with the other participating object via this communication channel.

[0067] If the at least two participating objects are in the voice call, in some embodiments each participating object can establish a channel for the voice call to the at least one participating object.

[0068] If the at least two participating objects are in the video call, in some embodiments each participating object can establish a channel for the video call to at least one participating object.

[0069] If the at least two participating objects are in the online meeting, in some embodiments each participating object can establish a channel for the meeting to at least one participating object.

[0070] If the at least two participating objects perform an SMS interaction, in some embodiments each participating object can establish a channel for message exchange with at least one participating object.

[0071] In S302, the information input of the participating object present and the information input of the other participating object are obtained to form the input data. The other participating object belongs to the at least one participating object.

[0072] Once the communication channel has been established, the at least two participating objects can communicate via this channel. During communication, the information input of the present participating object and the information input of the other participating object can be obtained via the communication channel to form the input data. The other participating object can belong to the at least one participating object. The present participating object can be the object that needs to process the input data.

[0073] In some embodiments, in a voice call scenario with three participating objects, each participating object can establish a voice call channel to the other two participating objects to conduct the voice call. It may be necessary for one participating object to summarize the voice audio content of the three participating objects. This single participating object can be designated as the present participating object. The intelligent device's call assistant can record the voice audio of the three participating objects based on the communication channel to form the input data.

[0074] If at least two objects are involved in the video call, in some embodiments the call assistant can obtain the video input of the present object and the video input of the other participating object based on the established video call channel in order to form the input data. The present object can be the object that needs to summarize the video call content.

[0075] If at least two participating objects are in the online meeting, in some embodiments the meeting assistant can obtain the meeting voice audio of the participating object present and the meeting voice audio of the other participating object based on the established meeting channel in order to form the input data. The participating object present can be the object that needs to summarize the meeting content.

[0076] In S303, the input data is analyzed to determine the restrictive condition entered by at least one participating object from the present participating object and the other participating object.

[0077] During communication via the communication channel, at least one participating object can insert its limiting condition into the input data. This allows the input data to be analyzed to determine the limiting condition imposed by at least one participating object, based on the existing object and the other participating object. The method for analyzing the input data to obtain the limiting condition can be arbitrary. For example, the input data can be fed into a speech audio analysis model. This model can then process the data to obtain the limiting condition.

[0078] When the at least two participating objects are in a voice call, in some embodiments at least one participating object can add the restrictive condition to the voice audio. The voice audios of the at least two participating objects can be analyzed to determine the restrictive condition entered by the at least one participating object.

[0079] In some embodiments, the at least two participating objects are in a video call, with at least one participating object being able to add the restrictive condition to the video. The videos of the at least two participating objects can be analyzed to determine the restrictive condition entered by the at least one participating object.

[0080] If the at least two participating objects are in the online meeting, in some embodiments at least one participating object can add the limiting condition to the meeting audio. The meeting audio of the at least two participating objects can be analyzed to determine the limiting condition entered by the at least one participating object.

[0081] In embodiments of the present disclosure, a communication channel can be established with at least one participating object. The information input of the present participating object and the information input of the other participating object can be obtained to form the input data. The other participating object can belong to the at least one participating object. Thus, the input data can be obtained based on the communication channel of the at least two participating objects. The input data can be analyzed to determine the limiting condition of the at least one participating object from the present participating object and the other participating object. The limiting condition can then be stored in the input data. The input data can be analyzed to obtain the limiting condition, which simplifies the step of obtaining the limiting condition.

[0082] In some implementations, if the other participating object enters a limiting condition, a notification message indicating that a limiting condition has been entered by the other participating object can be displayed on the current conversation page. This notification can take the form of an icon, the contents of the corresponding limiting condition, or something similar.

[0083] In some embodiments, if the other participating object enters a limiting condition, a notification indicating that the other participating object has entered a limiting condition can be displayed on a page aggregated by an artificial intelligence (AI) agent, as in Fig. 6 or Fig. Figure 7 shows the indicator. The indicator can be a symbol, the contents of the corresponding limiting condition, or something similar. The agent, also referred to as an "assistant," is an application of artificial intelligence technology and can be implemented based on the target model. The agent's behavior can be determined by the target model according to the agent's current state and external inputs. The target model can provide a decision-making basis for the agent through learning and training on large datasets. The agent can use tools, plug-ins, and knowledge bases to support inference, decision-making, and execution. The target model can provide the agent with a decision-making basis.

[0084] In some embodiments, obtaining the restrictive condition entered by the at least one participating object from the at least two participating objects, in order to form the descriptive data, may include: obtaining the restrictive condition entered by the other participating object when the communication channel to the at least one participating object has been successfully established; obtaining the restrictive condition entered by the other participating object when the communication channel to the at least one participating object has been disconnected; and / or obtaining the restrictive condition entered by the other participating object when communication is based on the communication channel to the at least one participating object.

[0085] In some embodiments, the other participating object can know that it is necessary to obtain and process the input data once the communication channel between the present participating object and the at least one participating object has been successfully established. The other participating object can then provide the restrictive condition based on the currently established communication channel or other communication channels. For example, if the two participants are in a voice call, it may be necessary for the present participating object to record the voice call and summarize the contents of the recording. When the present participating object records the voice audio, the other participating object can know that the present participating object has started recording. The other participating object can then provide the restrictive condition (e.g.,, that the voice audio of the other participating object must not be combined) based on the channel of the voice call, or the participating object present can be informed of the restrictive condition via message.

[0086] In some embodiments, the other participating object can know that it is necessary to obtain and process the input data when the communication channel between the present participating object and the at least one participating object has been severed. The other participating object can then provide the restrictive condition based on the currently established communication channel or other communication channels. For example, if the two participants are in a voice call, it may be necessary for the present participating object to record the call and summarize the recording. After the call has ended, the voice call assistant can automatically send a data processing request via message to the other participating object.After the other participating object has received the data processing request, the other participating object can inform the participating object present about the entered restrictive condition via message.

[0087] If the present participating object and the at least one participating object communicate via the established communication channel, the other participating object can, in some embodiments, know that the present participating object is about to summarize the communication content from the communication content of the present participating object. Then, the other participating object can directly inform the present participating object about the restrictive condition input by the other participating object, based on the current communication method. For example, if the two participants are in a voice call, it may be necessary for the present participating object to record the voice call and summarize the contents of the recording. The other participating object may have been informed and can then directly inform the present participating object about the restrictive condition via the voice audio.

[0088] When many people communicate, in some embodiments the restrictive conditions imposed by different other participating objects at different times can be obtained. That is, for the first other participating object, the restrictive condition of the other object can be obtained when the communication channel is successfully established. For the second other object, the restrictive condition of the other object can be obtained while the communication channel is being established. For the third other participating object, the restrictive condition of the other participating object can be obtained when the communication channel is closed.

[0089] In embodiments of the present disclosure, the variety of methods for obtaining the limiting condition can be improved by obtaining the limiting conditions, which were entered by the other participating object at various times during communication based on the communication channel. Therefore, obtaining the limiting condition can be more flexible.

[0090] In some embodiments, obtaining the descriptive data representing the limiting condition may further include obtaining the descriptive data representing the limiting condition in response to a call to the target model.

[0091] It is possible that the input data is not processed immediately after being obtained, and that the descriptive data of the constraint is not needed at that time. The descriptive data of the constraint is only required when it is necessary to call the target model to process the input data. Thus, the descriptive data representing the constraint can be obtained in response to a call to the target model.

[0092] The descriptive data can be the restrictive condition entered by the at least one object, which corresponds to the input data.

[0093] For example, during a voice call between two participating objects, the object present can record the call audio of both participating objects. After the call ends, the content of the voice audio is not necessarily summarized, and the descriptive data does not need to be obtained at that point. Only when it becomes necessary to process the voice audio based on the target model might it be necessary to obtain the descriptive data.

[0094] In some embodiments, after obtaining the descriptive data that constitutes the limiting condition, the subsequent processing process can be terminated if the target data corresponding to the limiting condition is all input data.

[0095] In embodiments of the present disclosure, the descriptive data constituting the limiting condition can be obtained in response to a call to the target model. The descriptive data can then only be obtained when the target model is called, in order to control the precise timing and avoid obtaining invalid descriptive data. This improves the efficiency and accuracy of the data processing.

[0096] Fig. Figure 4 illustrates a schematic flowchart of a data processing method according to some embodiments of the present disclosure. The method can be executed by the processor of the computer device. Starting from Fig. 1 is S103 in Fig. 1 to S401 and S402 have been updated. As in Fig. As shown in section 4, the procedure includes the following processes.

[0097] In S401, the input data is processed based on the descriptive data to obtain vector data that is used as input to the target model. Processing the input data based on the descriptive data includes cleaning the input data by removing the data that satisfies the limiting condition.

[0098] When processing input data, it may be necessary to first process the input data based on the descriptive data to obtain the vector data that can be fed into the target model. Then, the vector data can be processed based on the target model to generate the output data. If the input data is processed based on the descriptive data, it may be necessary to clean the input data by removing data that meets the limiting condition in order to obtain the cleaned vector data.

[0099] In some embodiments, the descriptive data and the input data can first be vectorized to obtain the vectorized descriptive data and input data, respectively. Then, based on the vectorized descriptive data, the vectorized input data can be cleaned to remove data that satisfies the constraint, resulting in the cleaned vector data. Finally, the vector data can be fed into the target model for processing.

[0100] In some embodiments, the input data can first be cleaned based on the descriptive data, removing the data that satisfy the limiting condition, in order to obtain the cleaned input data. Then, the cleaned input data can be vectorized to obtain the vector data for input into the target model.

[0101] The input data can be cleaned using existing data cleansing techniques to remove data that meet the restrictive condition. An existing data cleansing technique could be a data cleansing model.

[0102] If the input data is speech audio, it can be converted into text using, for example, speech-to-text conversion. The descriptive data can also be text. If it is necessary to clean the speech audio of location information, this information can be removed from the corresponding text to obtain the clean text for vectorization, which then yields the vector data to be input into the target model.

[0103] In S402, the vector data is processed based on the target model to generate the output data.

[0104] After obtaining the cleaned vector data, the vector data can be fed into the target model. The target model can be configured to process the vector data to generate the output data. The output data may not include data that meets the limiting condition.

[0105] For example, the vector data could be the call text after location information has been removed. This vector data can be fed into the target model to obtain a summary corresponding to the call text. The summary must not include the location information.

[0106] In embodiments of the present disclosure, the input data can be processed based on the descriptive data to obtain the vector data used as input to the target model. Processing the input data based on the descriptive data can include purifying the input data by removing the data that satisfy the limiting condition. The vector data can then be processed based on the target model to generate the output data. The input data can then be purified by removing the data that meet the limiting condition. The output data of the model can exclude the data that meet the limiting condition. Thus, the output result of the model can be effectively limited.

[0107] Fig. Figure 5 illustrates a schematic flowchart of a data processing method according to embodiments of the present disclosure. The method can be executed by the processor of the computer device. Starting from Fig. 1 is S103 in Fig. 1 to S501 to S503 have been updated. As in Fig. As shown in section 5, the procedure includes the following processes.

[0108] In S501, the input data is processed into the first vector data.

[0109] Once the input data has been obtained, the input data can be vectorized to obtain the first vector data.

[0110] If the input data is, for example, speech audio, the speech audio can first be converted into text. Then the text can be vectorized to obtain the first vector data.

[0111] In S502, the descriptive data is processed for the second vector data.

[0112] Once the descriptive data has been obtained, the descriptive data can be vectorized to obtain the second vector data.

[0113] In some embodiments, the descriptive data can be parsed after the descriptive data has been obtained to yield the parsed text. The text can then be vectorized to obtain the second set of vector data.

[0114] If the descriptive data is text, the text can, for example, be directly parsed and vectorized to obtain the second vector data.

[0115] In S503, the first vector data and the second vector data are processed based on the target model to generate the output data. During output generation, the target model is configured to clean the vector data that meets the limiting condition. The limiting condition is used to restrict the processing of the target model.

[0116] After obtaining the first vector data, corresponding to the input data, and the second vector data, corresponding to the descriptive data, the first and second vector data can be fed into the target model. The target model can be configured to process the second vector data based on the first vector data to obtain the output data. During the output generation process, the target model can be configured to purge the vector data that meets the limiting condition. The limiting condition can be used to restrict the processing capabilities of the target model.

[0117] In some embodiments, the target model can first process the initial vector data to obtain intermediate data. Then, based on the second set of vector data, the vector data that meet the limiting condition can be removed from the intermediate data to obtain the output data.

[0118] In some embodiments, the target model can first clean the first vector data based on the second vector data, removing the vector data that meet the limiting condition. Then, the cleaned first vector data can be further purified to obtain the output data.

[0119] In embodiments of the present disclosure, the input data can be processed into the first vector data, and the descriptive data can be processed into the second vector data. The first vector data and the second vector data can be processed based on the target model to generate the output data. When generating the output data, the target model can purify the vector data that satisfy the limiting condition. The limiting condition can be used to restrict the processing of the target model. Thus, the generated output data cannot include the data that meets the limiting condition, effectively restricting the output result of the model.

[0120] In some embodiments, the target model is trained based on a large number of training examples, comprising input data examples, descriptive data examples, and output data examples. Each input data example corresponds to one descriptive data example and one output data example. The target model is trained based on multiple sets of input data examples, descriptive data examples, and output data examples to achieve the trained target model.

[0121] In some embodiments, the target model can be trained based on a large number of training examples. A training example can include an input data example, a descriptive data example, and an output data example. Each input data example can be a one-to-one correspondence to the descriptive data example and the output data example. The target model can be trained based on multiple sets of input data examples, descriptive data examples, and output data examples to achieve the trained target model.

[0122] In some implementations, a large number of input data examples and descriptive data examples can be created. Based on the descriptive data examples, the input data examples can be processed to obtain a large number of processed input data examples. Based on this large number of processed input data examples and the output data examples, the target model can be trained to produce the trained target model.

[0123] The applications of the data processing method in actual scenarios are described below.

[0124] For a call assistant based on an AI-based large language model (LLM), current technology allows the speech audio of both participants to be converted into text using automatic speech recognition (ASR). This text content can then be directly entered into the large language model for summarization and recording. Therefore, both participants can use the large language model independently or simultaneously. However, since neither participant is aware of the other's actions and cannot control their behavior, significant risks can arise regarding the privacy, business, and even personal safety of the other participant.

[0125] Fig. Figure 6 illustrates a schematic diagram of a content summary for a two-way conversation according to embodiments of the present disclosure. As in Fig. As shown in Figure 6, a caller (601) and a called party (602) are in a voice call. An AI Call Assistant (603) is provided in the caller's setup's voice call interface. The AI ​​Call Assistant (603) can record the voice audio of both parties, convert it to text, and input the text into the large speech model to obtain a summary of the voice audio content. This summary can then be added to a note. However, the content summary provided by the AI ​​Call Assistant may include sensitive personal information such as telephone numbers.

[0126] In embodiments of the present disclosure, the AI-LLM call assistant can be controlled by both participants. If one participant does not agree, the ability of the other participant's AI model to derive and summarize can be limited or controlled. The other participant's response can be automatically used as part of the input content for the large language model, and the model can be informed about the encoding procedure and given instructions on how to clean the input. Thus, the final derivation and output of the assistant-LLM can be delivered in the direction specified by the user.

[0127] In the prior art, sensitive personal information is typically redacted in some embodiments to prevent data breaches, which can easily lead to inconsistencies in the information. For example, part of the summary may be redacted, and the information may be inconsistent. In embodiments of the present disclosure, the input data and the restrictive condition can be processed based on the model. The output data can exclude the target data corresponding to the restrictive condition. Therefore, the output content of the model can be consistent and does not include personal information. The content output by the model can be controlled. Regardless of whether the output content is disclosed to a third-party user or the third-party user actively obtains the output content, the output content can be secure.Unlike in the prior art, in embodiments of the present disclosure, the limiting information can be entered into the model. The limiting information can originate from any user involved who was not pre-configured by the equipment manufacturer during production.

[0128] Fig. Figure 7 illustrates a schematic diagram of a content summary for a two-way conversation according to embodiments of the present disclosure. As in Fig. As shown in Figure 7, if the two parties are on a voice call, the called party (602) can send an instruction to summarize the call content via SMS. The instruction could read, for example, "This call contains sensitive content; please do not summarize with AI." After the AI ​​call assistant (603) receives the SMS content from the caller's setup (601), it can parse the SMS content to determine that summarizing the content is unnecessary. The AI ​​call assistant interface (703) can then display a message indicating that no extractable information can be found from the call.

[0129] In some embodiments, the solution may include the following implementation steps.

[0130] Step 1: The call between the two parties begins and the party who needs the summary automatically activates the AI ​​call assistant on the other party's mobile device.

[0131] Step 2: During the call, the activated AI call assistant records the voice audio and converts it to text to vectorize the saved recording. The vectorized call content includes text information from both the participant present and the other participant. In one example, the vector content is...<Anrufer: Sprachaudiotext> and<Angerufener: Sprachaudiotext> . Here,<Anrufer: Sprachaudiotext> the caller's voice audio text and<Angerufener: Sprachaudiotext> represents the spoken audio text of the person being called.

[0132] Step 3: Immediately after the call ends, the AI ​​call assistant sends a text message to inform the other participants to prepare for the derivation and summary of the call content.

[0133] Step 4: The other party's response, including whether a summary is not permitted, whether a partial summary is permitted (the call assistant's party), and whether any summary is permitted, is obtained within 5 minutes. The response will be sent via SMS.

[0134] Step 5: The AI ​​call assistant parses the contents of the received SMS messages and vectorizes the SMS content.

[0135] Step 6: Embedded coding can be performed on the received SMS content and the ASR vector (the text vector corresponding to the speech audio) stored in the previous recording. Embedded coding refers to embedding the vectorized SMS content into the text vector of the speech audio to ensure that the object specified by the SMS content matches the object of the speech audio text. For example, the SMS instruction vector for the caller is embedded in the caller's speech audio text vector. The SMS instruction vector for the called party is embedded in the called party's speech audio text vector.

[0136] Step 7: The ASR call content and the content received via SMS are integrated to extract the input information for the final large language model. Once the embedded encoding is complete, the embedded encoding content is further integrated. For example, the embedded encoding is adapted to a predetermined encoding format, or the text vector that must not be aggregated is directly hidden or deleted to obtain the input information for further cleaning and filtering of the large language model.

[0137] Step 8: Cleaning and filtering are performed on the vectorized content (i.e., the input information for the large language model from the last step) according to the instructions. For example, sensitive information is filtered out.

[0138] Step 9: The cleaned content is fed into the large language model to perform derivation and summarization.

[0139] Fig. Figure 8 illustrates a schematic flowchart of a data processing procedure when two parties are involved in a call, according to embodiments of the present disclosure. As in Fig. As shown in section 8, the procedure includes the following processes.

[0140] In S801, the call ends and the AI ​​call assistant is triggered.

[0141] In S802, the ASR data is vectorized.

[0142] After the speech audio has been recorded, it can be converted into text and vectorized to obtain the text vector. For example: Caller vector<Anrufer: Sprachaudiotext> and called vector<Angerufener: Sprachaudiotext> .

[0143] In S803, the SMS text is sent automatically to notify the called party after the call has ended.

[0144] The person being called can be notified via SMS to carry out the derivation and summarization.

[0145] Step S804 determines whether a response is obtained within 5 minutes. If so, the process continues with step S805; if not, with step S812.

[0146] In S805, the answer is parsed and vectorized.

[0147] It may be necessary to parse the other participant's response in order to identify the content of the response and vectorize the parsed content.

[0148] In S806, the response vector is embedded in the ASR vector.

[0149] Step S807 determines whether the ASR vector can be entered into the large language model. If yes, the process continues with step S808; if no, it continues with step S812.

[0150] In S808, the final input data for the large language model is extracted.

[0151] The final input data for the large language model can include instructions (the reply SMS content) and speech audio text (content to be summarized).

[0152] In S809, the input data for the large language model is cleaned and filtered.

[0153] The speech audio text can be cleaned and filtered based on the instructions.

[0154] Step S810 determines whether the cleaning and filtering was successful. If so, it proceeds to step S811; if not, to step S812.

[0155] In S811, the cleaned data is entered into the large language model.

[0156] In S812: The End.

[0157] In some embodiments, the input content received by the large speech model may, in a first situation, contain no input and the called party does not allow the large speech model to perform derivation and summarization; in a second situation, contain only the contents of the caller {SMS instruction of the called party: voice audio text of the caller}; or in a third situation, contain the contents of both parties {SMS instruction of the called party: voice audio text of the caller and voice audio text of the called party}.

[0158] In some implementations, the called party's SMS instruction can be used as part of the large speech model to clean up the input content. This informs the large speech model about the type and extent of adjustment or content filtering to be performed when deriving and summarizing the speech audio content, preventing the large speech model from performing arbitrary summarization and output. For example, if the received reply SMS includes the instruction "Remove phone number and location information from the call content of the person present," the large speech model can clean up and filter out the called party's sensitive and private information.

[0159] In embodiments of the present disclosure, the behavior of the AI-LLM of both parties in the call can be assessed and accepted by the other party, and the respective responses of both parties can be used as part of the input for the large language model to effectively manage its behavior. The large language model can then behave as required, and malicious output can be prevented. The interests of both parties in the call can be effectively protected, and furthermore, data breaches of the user's business and financial information can be prevented.

[0160] Based on the foregoing, embodiments of the present disclosure further provide a data processing device. The device may comprise units and modules contained within the units, which can be implemented by one or more processors of the computer device or a specific logic circuit. During implementation, the one or more processors may include a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), etc.

[0161] Fig. Figure 9 illustrates a schematic structure diagram of a data processing device 900 according to embodiments of the present disclosure. As in Fig. As shown in Figure 9, the data processing unit 900 has a first reference module 910, a second reference module 920, a processing module 930 and an output module 940.

[0162] The first reference module 910 can be configured to obtain the input data.

[0163] The second reference module 920 can be configured to retrieve the descriptive data that defines the limiting condition. The input data can include the target data that meets the limiting condition.

[0164] The 930 processing module can be configured to process the input data based on the target model, the input data, and the descriptive data to generate the output data. The target model can be a generative large language model.

[0165] The output module 940 can be configured to output the output data. The output data cannot include the target data that meets the restrictive condition.

[0166] In some embodiments, the first reference module 910 can further be configured to obtain the information input from the at least two participating objects in order to form the input data. The second reference module 920 can further be configured to obtain the restrictive condition entered by the at least one participating object from the at least two participating objects in order to form the descriptive data.

[0167] In some embodiments, the limiting condition can be applied from the first participating object to a portion of the input data belonging to the first participating object.

[0168] In some embodiments, the first reference module 910 can further be configured to establish the communication channel to the at least one participating object and to obtain the information input of the present participating object and the information input of the other participating object in order to form the input data. The present participating object can belong to the at least one participating object. The second reference module 920 can further be configured to analyze the input data in order to determine the restrictive condition entered by the at least one participating object from the present participating object and the other participating object.

[0169] In some embodiments, the reference module 920 may further be configured to obtain, when the communication channel to the at least one participating object has been successfully established, the restrictive condition entered by the other participating object; when the communication channel to the at least one participating object has been disconnected, the restrictive condition entered by the other participating object; and during communication based on the communication channel to the at least one participating object, the restrictive condition entered by the other participating object.

[0170] In some embodiments, the second reference module 920 may also be configured to obtain, in response to a call to the target model, the descriptive data that represent the limiting condition.

[0171] In some embodiments, the processing module 930 can further be configured to process the input data based on the descriptive data in order to obtain the vector data to be input into the target model. Processing the input data based on the descriptive data can include purifying the input data by removing the data that satisfy the limiting condition. The processing module 930 can further be configured to process the vector data based on the target model in order to generate the output data.

[0172] In some embodiments, the processing module 930 can further be configured to process the input data for the first vector data, process the descriptive data for the second vector data, and process the first and second vector data based on the target model to generate the output data. Once the output data has been generated, the target model can be configured to clean the vector data according to the limiting condition. The limiting condition can be used to restrict the processing of the target model.

[0173] The description of the above device embodiments may be similar to the description of the above method embodiments and exhibits similar advantageous effects. In some embodiments, the functions or modules included in the device according to embodiments of this disclosure may be used to perform the methods described above. For technical details not disclosed in the device embodiments, reference may be made to the description of the method embodiments.

[0174] In embodiments of the present disclosure, if the data processing method is implemented in the form of functional software modules and sold or used as an independent product, the method may also be stored on a computer-readable storage medium. Based on this understanding, the essence of the technical solutions of the present disclosure, or the part of the technical solutions that contributes to the prior art, may be embodied as software products. The software products may be stored on a storage medium and comprise several instructions used to cause a computer device (e.g., a personal computer, a server, or a network device, etc.) to execute all or part of the methods according to embodiments of the present disclosure.The storage medium can include a USB flash drive, a portable hard drive, read-only memory (ROM), magnetic disks, optical disks, and other media capable of storing program code. Therefore, the embodiments described in this disclosure are not limited to specific hardware, software, firmware, or any combination thereof.

[0175] Embodiments of the present disclosure provide a computer device comprising a memory and a processor. The memory may contain a computer program which, when executed by the processor, causes the processor to perform some or all of the steps of the methods described above.

[0176] Embodiments of the present disclosure provide the computer-readable storage medium on which a computer program is stored which, when executed by the processor, causes the processor to implement all or part of the steps of the above methods. The computer-readable storage medium can be volatile or non-volatile.

[0177] Embodiments of the present disclosure provide a computer program comprising computer-readable code. When the computer-readable code runs on the computer device, the processor of the computer device can execute all or part of the steps of the above procedures.

[0178] Embodiments of the present disclosure provide a computer program product. The computer program product may include a non-volatile, computer-readable storage medium on which the computer program is stored. When read and executed by a computer, the computer program causes the computer to perform all or part of the steps of the above procedures. The computer program product may be implemented by hardware, software, or a combination thereof. In some embodiments, the computer program product may be embodied as a computer storage medium. In some other embodiments, the computer program product may be embodied as a software product, for example, as a software development kit (SDK).

[0179] The descriptions of the various embodiments above tend to emphasize the differences between them. Identical or similar parts may refer to one another. The descriptions of the above device, storage medium, computer program, and computer program product embodiments are similar to the description of the method embodiments above and exhibit similar advantageous effects. For technical details not disclosed in the device, storage medium, computer program, and computer program product embodiments of this disclosure, reference may be made to the description of the method embodiments of this disclosure.

[0180] Fig. Figure 10 illustrates a schematic structure diagram of an electronic device 1000 according to embodiments of the present disclosure. As in Fig. As shown in Figure 10, the electronic device 1000 has a first input assembly 1010, a second input assembly 1020, a processor 1030 and an output assembly 1040.

[0181] The first input module 1010 can be configured to obtain the input data.

[0182] The second input module 1020 can be configured to retrieve the descriptive data that defines the limiting condition. The input data can include the target data that meets the limiting condition.

[0183] The 1030 processor can be configured to process input data based on the target model, the input data, and the descriptive data to generate the output data. The target model can be the generative large language model.

[0184] Output module 1040 can be configured to output the output data. The output data cannot include the target data that meets the restrictive condition.

[0185] The electronic device can be any device capable of performing the data processing procedure. For example, the electronic device can include a smartphone, a tablet, a computer, etc.

[0186] The input data can be the data to be processed. The descriptive data can be data about the limiting condition. The limiting condition can correspond to the target data in the input data. The target data can be data in the input data whose processing is not permitted or is unnecessary. The descriptive data can limit the final output result.

[0187] In some embodiments, the input data and the descriptive data can be obtained by different input assemblies. For example, in a voice call scenario with at least two participating objects, the present participating object can establish a voice audio channel to the other participating object and conduct a voice call based on this voice audio channel. The first input assembly can be configured to obtain the voice audio of the present participating object and the other participating object in the voice audio channel in order to obtain the input data. The other participating object can propose the descriptive data, which represents the limiting condition, to the present participating object. For example, the other participating object can send the descriptive data to the present participating object via SMS.The descriptive data representing the limiting condition can then be obtained through the second input module. The first input module can be the speech audio parser module in the electronic device and can be configured to parse the speech audio of both participants in the call to obtain the input data. The second input module can be the data parser module of the electronic device and can be configured to parse the SMS sent by the other participating object to obtain the descriptive data.

[0188] In a voice call scenario with at least two participating objects, in some embodiments the first input assembly can be configured to obtain the voice audio of the present participating object and the other participating object in the voice audio channel in order to obtain the input data. The other participating object can suggest the descriptive data that represents the limiting condition. For example, the present participating object can select whether to limit the voice audio via a prompt that pops up on the touchscreen of the electronic device. The descriptive data of the present participating object can be obtained by the second input assembly. The second input assembly can be the touchscreen. In another example, the electronic device can be a computer.The participating object can determine the descriptive data through mouse or keyboard input to the electronic device. The second input assembly can be an assembly that responds to the mouse or keyboard.

[0189] In some embodiments, in a voice call scenario with at least two participating objects, the call content of both participants can include the descriptive data that represents the limiting condition. For example, the other participating object can directly state during the call that no identity information may be processed in the voice call. In this case, the input data and the descriptive data are derived from the voice call, and only a single input assembly is required to obtain the input data and the descriptive data that represent the limiting condition.

[0190] In some embodiments, the input data and the descriptive data can originate from the same user. The input data can be speech audio, video, or text stored by the user, and the descriptive data can be specified by the user for the speech audio, video, or text to be processed. For example, if the input data and the descriptive data are speech audio segments, a single input assembly can be configured to retrieve both. This input assembly can be a speech audio parser assembly. In another example, if the input data is speech audio and the descriptive data is text, the first input assembly can be configured to retrieve the input data, and the second input assembly can be configured to retrieve the descriptive data.The first input module can be a speech audio parser module and the second input module can be a text parser module.

[0191] In some embodiments, the processor can be communicatively connected to the first input assembly and the second input assembly. After the first input assembly has received the input data and the second input assembly has received the descriptive data, the processor can be configured to process the input data based on the target model, the input data, and the descriptive data to generate the output data. The target model can be the generative large language model. The generative large language model can be configured to summarize the input data to obtain the summary corresponding to the input data. For example, the processor can be configured to call the target model, input the input data and the descriptive data into the target model, execute the target model, and obtain the output data.

[0192] In some embodiments, after the processor has processed the input data, the output assembly can be configured to output the output data. The output data can be the final processed data of the input data and may not include the target data corresponding to the limiting condition. For example, the output data can be a summary text, and the output assembly can be a text generation assembly of the electronic device.

[0193] In some embodiments, the electronic device may further include a memory used to store the target model. The processor 1030 may also be configured to load and execute the target model from the memory.

[0194] The target model can be stored in the memory of the electronic device. The processor can access the memory, load the target model from memory into internal memory, and execute the target model.

[0195] In some embodiments, the first input assembly 1010 can further be configured to obtain the information input from the at least two participating objects in order to form the input data. The second input assembly 1020 can be configured to obtain the restrictive condition entered by the at least one participating object from the at least two participating objects in order to form the descriptive data.

[0196] In some embodiments, the restrictive conditions can be applied by the first participant to the portion of the input data belonging to the first participating object.

[0197] In some embodiments, the first input assembly 1010 may further be configured to establish the communication channel to the at least one participating object, to obtain the information input of the present participating object and the information input of the other participating object in order to form the input data, and to analyze the input data in order to determine the restrictive condition entered by the at least one participating object from the present participating object and the other participating object.

[0198] In some embodiments, the first input assembly 1010 may further be configured to obtain, when the communication channel to the at least one participating object has been successfully established, the restrictive condition entered by the other participating object; when the communication channel to the at least one participating object has been disconnected, the restrictive condition entered by the other participating object; and during communication based on the communication channel to the at least one participating object, the restrictive condition entered by the other participating object.

[0199] In some embodiments, the second input assembly 1020 may also be configured to obtain, in response to a call to the target model, the descriptive data that represent the limiting condition.

[0200] In some embodiments, the 1030 processor can further be configured to process the input data based on the descriptive data in order to obtain the vector data to be fed into the target model. Processing the input data based on the descriptive data can include purifying the input data by removing the data that satisfies the limiting condition. The 1030 processor can further be configured to process the vector data based on the target model in order to generate the output data.

[0201] In some embodiments, the 1030 processor can further be configured to process the input data for the first vector data, process the descriptive data for the second vector data, and process the first and second vector data based on the target model to generate the output data. When generating the output data, the target model can purify the vector data that satisfy the limiting condition. The limiting condition can be used to restrict the processing of the target model.

[0202] Fig. Figure 11 illustrates a schematic diagram of a computer device 1100 according to embodiments of the present disclosure. As in Fig.As shown in Figure 11, the hardware unit of the computer system 1100 comprises a processor 1101 and a memory 1102. A computer program is stored in the memory 1102, which, when executed by the processor 1101, causes the processor 1101 to carry out the steps of the procedure described above.

[0203] Memory 1102 can store the computer program that can be executed by the processor. Memory 1102 can be configured to store instructions and applications that can be executed by the processor 1101 and to temporarily store data that is to be processed or has already been processed by the processor 1101 and various modules in the computer unit 1100 (e.g., image data, speech audio data, speech audio communication data, and video communication data), which can be stored in flash memory (FLASH) or random access memory (RAM).

[0204] When processor 1101 executes the program, the steps of any of the above data processing procedures can be implemented. Processor 1101 usually controls the overall operation of computer system 1100.

[0205] Embodiments of the present disclosure provide a computer storage medium on which one or more programs are stored which, when executed by the processor, cause the processor to implement the steps of the above data processing procedures.

[0206] The descriptions of the above storage medium and equipment embodiments are similar to the descriptions of the above process embodiments and exhibit similar advantageous effects. For technical details not disclosed in the storage medium and equipment embodiments of this disclosure, reference may be made to the description of the process embodiments in this disclosure.

[0207] The processor described above may include an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing unit (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a control unit, a microcontroller, and / or a microprocessor. The electronic device implementing the processor's functions may also include other types; these are not limited here.

[0208] The above computer storage medium or memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical discs or CD-ROMs, as well as various terminal devices that incorporate any of the above memory or any combination thereof, for example, mobile phones, computers, tablets, personal digital assistants, etc.

[0209] The above description merely illustrates some embodiments of the present disclosure. The scope of the present disclosure is not limited thereto. A person skilled in the art can easily devise modifications or substitutions of embodiments of the present disclosure. These modifications and substitutions fall within the scope of the present disclosure. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] CN 202411808018.4

[0001]

Claims

[1] Data processing procedures, comprehensive: Obtaining input data; Obtaining descriptive data that constitutes a limiting condition, where the input data includes target data that meet the limiting condition; based on a target model, the input data and the descriptive data, processing the input data to generate output data, where the target model is a generative large language model; and Outputting the output data, where the output data does not include the target data corresponding to the restrictive condition. [2] Method according to claim 1, wherein: Obtaining the input data includes: Obtaining information input from at least two participating objects in order to form the input data; and Obtaining the descriptive data used to represent the limiting condition includes: Obtaining the restrictive condition entered by at least one of the two participating objects in order to form the descriptive data. [3] The method of claim 2, further comprising: Applying a restrictive condition from a first participating object to a portion of the input data belonging to the first participating object. [4] Method according to claim 3, wherein: Obtaining the information input from at least two participating objects in order to form the input data includes: Establishing a communication channel to at least one participating object; and Obtaining an information input from a present participating object and an information input from another participating object in order to form the input data, wherein the other participating object belongs to the at least one participating object; and Obtaining the restrictive condition entered by the at least one participating object from the at least two participating objects, in order to form the descriptive data, includes: Analyzing the input data to determine the restrictive condition entered by the at least one participating object from the present participating object and the other participating object. [5] Method according to claim 3, wherein: Obtaining the information input from at least two participating objects in order to form the input data includes: Establishing a communication channel to the at least one object involved; and Obtaining an information input from the present participating object and an information input from the other participating object in order to form the input data, wherein the other participating object belongs to the at least one participating object; and Obtaining the restrictive condition entered by the at least one participating object from the at least two participating objects, in order to form the descriptive data, includes: as a reaction to the successful establishment of the communication channel to at least one participating object, obtaining the restrictive condition entered by the other participating object; as a reaction to the communication channel to at least one of the involved objects being severed, obtaining the restrictive condition entered by the other involved object; and / or Obtaining the restrictive condition entered by the other participating object during communication based on the communication channel established to the at least one participating object. [6] The method of claim 1, wherein obtaining the descriptive data used to represent the limiting condition further comprises: in response to a call to the target model, obtaining the descriptive data used to represent the limiting condition. [7] Method according to claim 1, wherein the processing of the input data based on the target model, the input data and the descriptive data to generate the output data comprises: Processing the input data based on the descriptive data to obtain vector data used as input to the target model, which includes cleaning the input data of data that satisfy the restrictive condition; and Processing the vector data based on the target model to generate the output data. [8] Method according to claim 1, wherein the processing of the input data based on the target model, the input data and the descriptive data to generate the output data comprises: Processing the input data into initial vector data; Processing the descriptive data into second vector data; and Processing the first vector data and the second vector data based on the target model to generate the output data, wherein the target model is set up to clean vector data that satisfy the limiting condition when generating the output data, wherein the limiting condition is used to restrict a processing process of the target model. [9] Electronic device comprising: a first input assembly that is configured to receive input; a second input assembly configured to obtain descriptive data that constitutes a limiting condition, wherein the input data includes target data that conforms to the limiting condition; one or more processors configured to process the input data, based on a target model, input data, and descriptive data, to generate output data, where the target model is a generative large language model; and an output assembly configured to output the output data, wherein the output data does not include the target data corresponding to the restrictive condition. [10] Device according to claim 9, further comprising one or more memories in which the target model is stored, wherein: which are further configured to load and execute the target model from one or more memories. [11] Device according to claim 9, wherein: the first input assembly is further set up for this purpose: to obtain information input from at least two participating objects in order to form the input data; and the second input assembly is further set up for this purpose: to obtain the restrictive condition entered by at least one of the two participating objects in order to form the descriptive data. [12] Device according to claim 11, wherein the one or more processors are further configured to: to apply a restrictive condition from a first participating object to a portion of the input data that belongs to the first participating object. [13] Non-volatile, computer-readable storage medium on which a computer program is stored which, when executed by one or more processors, causes the one or more processors to: Obtain input data; to obtain descriptive data that constitute a limiting condition, where the input data includes target data that meet the limiting condition; based on a target model, the input data, and the descriptive data that process the input data to generate output data, where the target model is a generative large language model; and Output the output data, excluding the target data that meets the restrictive condition. [14] Storage medium according to claim 13, wherein the one or more processors are further configured to: to obtain information input from at least two participating objects in order to form the input data; and to obtain the restrictive condition entered by at least one of the two participating objects in order to form the descriptive data. [15] Storage medium according to claim 14, wherein the one or more processors are further configured to: to apply a restrictive condition from a first participating object to a portion of the input data that belongs to the first participating object. [16] Storage medium according to claim 15, wherein the one or more processors are further configured to: to establish a communication channel to at least one of the objects involved; to obtain an information input from one participating object present and an information input from another participating object in order to form the input data, wherein the other participating object belongs to the at least one participating object; and to analyze the input data in order to determine the restrictive condition entered by the at least one participating object from the one participating object and the other participating object. [17] Storage medium according to claim 14, wherein the one or more processors are further configured to: Establishing a communication channel to the at least one object involved; and Obtaining an information input from the present participating object and an information input from the other participating object in order to form the input data, wherein the other participating object belongs to the at least one participating object; and, as a reaction to the successful establishment of the communication channel to at least one participating object, obtaining the restrictive condition entered by the other participating object; as a reaction to the communication channel to at least one of the involved objects being severed, obtaining the restrictive condition entered by the other involved object; or Obtaining the restrictive condition entered by the other participating object during communication based on the communication channel established to the at least one participating object. [18] Storage medium according to claim 13, wherein the one or more processors are further configured to: to obtain the descriptive data used to represent the limiting condition in response to a call to the target model. [19] Storage medium according to claim 13, wherein the one or more processors are further configured to: to process the input data based on the descriptive data in order to obtain vector data used for input into the target model, which includes cleaning the input data of data that satisfy the restrictive condition; and to process the vector data based on the target model in order to generate the output data. [20] Storage medium according to claim 13, wherein the one or more processors are further configured to: to process the input data into initial vector data; to process the descriptive data into second vector data; and to process the first vector data and the second vector data based on the target model to generate the output data, wherein the target model is set up to clean vector data that satisfy the limiting condition when generating the output data, wherein the limiting condition is used to restrict a processing process of the target model.

Images