Data processing method and device, program product and electronic equipment

By generating the image data of the data transmission node and updating the training data of the signal processing model, the problem of insufficient inference capability of the signal processing model is solved, and data transmission efficiency and encoding and codec reliability in the communication process are improved.

CN118917363APending Publication Date: 2024-11-08CHINA TELECOM CORP LTD TECHNOLOGY INNOVATION CENTER +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411081753.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the existing communication technology, the generation of training data of signal processing models based on large-scale global knowledge leads to insufficient inference capabilities of the signal processing model, which reduces the reliability of the encoding and decoding process and leads to inefficient data transmission efficiency.

Method used

By obtaining the interactive data and feature data of the data transmission node and the generated artificial intelligence model, the image data of the data transmission node is generated, and the training data of the signal processing model is updated based on this to improve the inference ability of the signal processing model and the enrichment of training samples. Spend.

Benefits of technology

It improves the encoding accuracy and decoding reliability of the signal processing model, improves the data transmission efficiency during communication, and enhances the diversity and adaptability of the training sample of the signal processing model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917363A_ABST
    Figure CN118917363A_ABST
Patent Text Reader

Abstract

The invention provides a data processing method and device, a program product and electronic equipment, and relates to the technical field of wireless communication, and the method comprises the steps: obtaining the interaction data of a data transmission node and a generative artificial intelligence model, and the feature data of the data transmission node; inputting the interaction data and the feature data into a training sample generation model, and generating portrait data of the data transmission node based on the output of the training sample generation model; updating original training data corresponding to the signal processing model according to the portrait data to obtain target training data; and training the signal processing model through the target training data, so that the trained signal processing model encodes or decodes transmission information by referring to the portrait data when receiving the transmission information of the data transmission node. According to the invention, the reasoning capability of the signal processing model is improved, and the data transmission efficiency in the communication process is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of wireless communication technology, and in particular to a data processing method, a data processing device, a computer program product, and an electronic device. Background Art

[0002] In the related art, during the communication process, training data for the signal processing model is usually generated based on large-scale global knowledge in the communication field. However, this method leads to insufficient reasoning ability of the trained signal processing model, reduces the reliability of the encoding and decoding process, and causes low data transmission efficiency in the current communication process. Summary of the invention

[0003] The present disclosure provides a data processing method, a data processing device, a computer program product and an electronic device, thereby improving the problem of low data transmission efficiency during the communication process at least to a certain extent.

[0004] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0005] According to a first aspect of the present disclosure, a data processing method is provided, which is applied to a communication system comprising at least a data transmission node and a signal processing model, and the method comprises: obtaining interaction data between the data transmission node and a generative artificial intelligence model, and feature data of the data transmission node; inputting the interaction data and the feature data into a training sample generation model, and generating portrait data of the data transmission node based on the output of the training sample generation model; updating the original training data corresponding to the signal processing model according to the portrait data to obtain target training data; and training the signal processing model through the target training data so that the trained signal processing model, when receiving the transmission information of the data transmission node, encodes or decodes the transmission information with reference to the portrait data.

[0006] Optionally, the training sample generation model includes a first sub-model, a second sub-model and a third sub-model, wherein the second sub-model includes at least the generative artificial intelligence model; the interaction data and the feature data are input into the training sample generation model, and the portrait data of the data transmission node is generated based on the output result of the training sample generation model, including: generating the input data of the second sub-model according to the first sub-model, the feature data, and the interaction data; obtaining the response result data of the generative artificial intelligence model to the input data, and using the second sub-model to generate the data to be verified based on the response result data; verifying the data to be verified through the third sub-model, and if the verification fails, obtaining the correction information output by the third sub-model, so that the first sub-model updates the input data according to the correction information, and regenerates the data to be verified until the verification passes; and if the verification passes, generating the portrait data based on the data to be verified, and the input data corresponding to the data to be verified or the updated input data.

[0007] Optionally, generating the input data of the second sub-model according to the first sub-model, the feature data, and the interaction data includes: obtaining the stem data from the interaction data through the first sub-model, and obtaining a set of candidate background data associated with the stem data based on the interaction data and the feature data; arranging each of the candidate background data in the candidate background data set in reverse order according to the degree of association with the stem data; when the first sub-model generates the target background data set associated with the stem data for the first time, determining the first N candidate background data in the candidate background data set as the target background data; when the correction information is received, determining the target background data in the candidate background data set based on the correction information, where N is a positive integer; popping out the target background data from the candidate background data set, and adding the popped-up target background data to the target background data set; generating a prompt word for the generative artificial intelligence model in combination with the stem data and the target background data in the target background data set, and determining the input data based on the prompt word.

[0008] Optionally, determining the target background data in the candidate background data set based on the correction information includes: updating the arrangement order of each candidate background data in the candidate background data set based on the correction information and the question stem data, and determining the first N candidate background data as the target background data in the updated candidate background data set.

[0009] Optionally, generating the data to be verified based on the response result data includes: obtaining the stem data corresponding to the response result data, and the standard response data corresponding to the stem data; matching the stem data with the standard response data to obtain a first data pair, and matching the stem data with the response result data to obtain a second data pair, and obtaining the data to be verified based on the first data pair and the second data pair.

[0010] Optionally, the data to be verified includes stem data, and response result data and standard response data corresponding to the stem data; the verification of the data to be verified through the third sub-model includes: obtaining standard response data and response result data corresponding to the same stem data based on the data to be verified; calculating the similarity between the standard response data and the response result data corresponding to the same stem data, so as to verify the data to be verified.

[0011] Optionally, generating the portrait data based on the data to be verified and the input data corresponding to the data to be verified or the updated input data includes: obtaining response result data in the data to be verified; associating the input data with the response result data to obtain an associated data record; and adding the associated data record to a data set corresponding to the portrait data to obtain the portrait data.

[0012] According to a second aspect of the present disclosure, a data processing device is provided, which is applied to a communication system that includes at least a data transmission node and a signal processing model, and the device includes: a data acquisition module, configured to acquire interaction data between the data transmission node and the generative artificial intelligence model, and feature data of the data transmission node; a portrait data generation module, configured to input the interaction data and the feature data into a training sample generation model, and generate portrait data of the data transmission node based on the output of the training sample generation model; a training data update module, configured to update the original training data corresponding to the signal processing model according to the portrait data to obtain target training data; a signal processing model training module, configured to train the signal processing model with the target training data, so that the trained signal processing model, when receiving the transmission information of the data transmission node, encodes or decodes the transmission information with reference to the portrait data.

[0013] According to a third aspect of the present disclosure, a computer program is provided, which, when executed by a processor, implements the data processing method of the first aspect and possible implementation manners thereof.

[0014] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor, wherein the processor is configured to execute the data processing method of the first aspect and possible implementation thereof by executing the executable instructions.

[0015] The technical solution disclosed in this disclosure has the following beneficial effects:

[0016] On the one hand, the portrait data of the data transmission node of the communication system is generated, and the training data of the signal processing model is enriched by the portrait data, so that the trained signal processing model can refer to the portrait data of the data transmission node for encoding or decoding; this solution solves the problem that the signal processing model cannot refer to the characteristics of the transmitting end or the receiving end for encoding or decoding to a certain extent, improves the reasoning ability of the signal processing model, enables the encoder to refer to the portrait data and infer the original intention of the transmission information for encoding, effectively improves the accuracy of the encoding, and enables the decoder to refer to the portrait data of the data transmission node for decoding, so that the decoded information is easier to be understood by the data transmission node, thereby improving the reliability of the decoding process, thereby effectively improving the data transmission efficiency during the communication process. On the other hand, the original training data of the signal processing model is updated by the portrait data of the data transmission node to obtain the target training data, and the signal processing model is trained based on the target training data, effectively improving the richness and diversity of the training samples of the signal processing model.

[0017] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0019] Figure 1 A flow chart showing a data processing method according to the exemplary embodiment;

[0020] Figure 2 A flowchart showing a method for generating portrait data in this exemplary embodiment;

[0021] Figure 3 A flow chart showing a method for obtaining input data in this exemplary embodiment;

[0022] Figure 4A flow chart showing a verification method in this exemplary embodiment is shown;

[0023] Figure 5 A flow chart showing a method for acquiring portrait data based on the data to be verified and the input data in this exemplary embodiment is shown;

[0024] Figure 6 A schematic diagram showing the structure of a training sample generation model in this exemplary embodiment;

[0025] Figure 7 A flowchart showing another data processing method in this exemplary embodiment;

[0026] Figure 8 A schematic diagram showing the structure of a semantic communication system to which the data processing method in this exemplary embodiment is applied;

[0027] Fig. 9 A schematic diagram showing the structure of a data processing device in this exemplary embodiment is shown;

[0028] Fig.10 A schematic structural diagram of an electronic device in this exemplary embodiment is shown. DETAILED DESCRIPTION

[0029] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0030] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0031] In the related art, training data for signal processing models is usually generated based on large-scale global knowledge in the communication field. However, this method leads to insufficient reasoning ability of the signal processing model, reduces the reliability of the encoding and decoding process, and causes low data transmission efficiency in the current communication process.

[0032] In view of one or more of the above problems, an exemplary embodiment of the present disclosure first provides a data processing method. The data processing method in this exemplary embodiment can be applied to a communication system including at least a data transmission node and a signal processing model.

[0033] Combine the following Figure 1 The data processing method is described. Figure 1 An exemplary process of a data processing method is shown, including the following steps S110 to S140:

[0034] Step S110, acquiring interaction data between the data transmission node and the generative artificial intelligence model, and feature data of the data transmission node;

[0035] Step S120, inputting the interaction data and the feature data into a training sample generation model, and generating the portrait data of the data transmission node based on the output of the training sample generation model;

[0036] Step S130, updating the original training data corresponding to the signal processing model according to the portrait data to obtain target training data;

[0037] Step S140, training the signal processing model through target training data, so that the trained signal processing model can encode or decode the transmission information with reference to the portrait data when receiving the transmission information from the data transmission node.

[0038] Based on the above method, on the one hand, the portrait data of the data transmission node of the communication system is generated, and the training data of the signal processing model is enriched by the portrait data, so that the trained signal processing model can refer to the portrait data of the data transmission node for encoding or decoding; this solution solves the problem that the signal processing model cannot refer to the characteristics of the transmitting end or the receiving end for encoding or decoding to a certain extent, improves the reasoning ability of the signal processing model, enables the encoder to refer to the portrait data and infer the original intention of the transmission information for encoding, effectively improves the accuracy of the encoding, and enables the decoder to refer to the portrait data of the data transmission node for decoding, so that the decoded information is easier to be understood by the data transmission node, thereby improving the reliability of the decoding process, thereby effectively improving the data transmission efficiency during the communication process. On the other hand, the original training data of the signal processing model is updated by the portrait data of the data transmission node to obtain the target training data, and the signal processing model is trained based on the target training data, effectively improving the richness and diversity of the training samples of the signal processing model.

[0039] Below Figure 1 Each step in the procedure is described in detail.

[0040] refer to Figure 1 , in step S110, the interaction data between the data transmission node and the generative artificial intelligence model, as well as the feature data of the data transmission node are obtained.

[0041] Among them, the above-mentioned communication system may include a semantic communication system. The core of semantic communication (Semantic Communication, SemCom) is to extract the "original meaning" of the transmitted information, that is, the semantic communication system enables the communication system to have a certain understanding ability, rather than simply transmitting a bit stream; at the same time, the semantic communication system may include data transmission nodes such as the sender and the receiver, and by using the semantic knowledge bases corresponding to the sender and the receiver respectively, the accuracy of information transmission is improved. It can be seen that semantic communication is essentially a communication solution based on artificial intelligence.

[0042] The data transmission node may include a transmitting end and a receiving end in a communication system. For example, the data transmission node may include a user of the transmitting end, and the user may have specific communication requirements.

[0043] The signal processing model may include a model for processing the transmission information sent by the data transmission node, such as a semantic encoding model (SEM). In a semantic communication system, the semantic encoding model can be used to convert the transmission information into a symbolic representation with specific semantic content; the model can combine the semantic knowledge base and the interaction data with the generative artificial intelligence model to extract the core semantics from the transmission information and encode it in a compact and easy-to-transmit form to ensure that the original semantics are retained to the maximum extent during the information transmission process. Semantic communication has pioneered a paradigm shift in the communication process from bit-oriented transmission to semantic-oriented transmission, and the semantic encoding model and the semantic decoding model can accurately extract and restore the core semantics of the transmission message with the help of the semantic knowledge base. The knowledge base (KB) refers to a data set that stores and organizes background information, context information, and semantic content related to a specific communication task; it is used in the process of semantic communication to support the semantic encoding model and decoder to more accurately understand and reconstruct the semantic information of the message during the information transmission process; the knowledge base includes a variety of methods for representing data, such as knowledge graphs, semantic tags, and triples. In the process of semantic communication driven by neural networks, the knowledge base can be regarded as a training dataset for the neural networks (signal processing models) in the encoder and decoder.

[0044] Generative artificial intelligence (GAI) is an artificial intelligence technology that uses complex algorithms, models and rules to learn from large-scale data sets to create new original content. Through GAI, various types of content such as text, pictures, sounds, videos and codes can be created. Compared with traditional models, GAI has stronger data processing and analysis capabilities.

[0045] The interaction data includes the context data generated by the interaction between the data transmission node and GAI. The feature data comes from the knowledge base corresponding to the data transmission node.

[0046] Continue to refer Figure 1 In step S120, the interaction data and the feature data are input into the training sample generation model, and the portrait data of the data transmission node is generated based on the output of the training sample generation model.

[0047] By generating the portrait data of the data transmission node through the interaction data and the feature data, and adding the portrait data as a supplementary training sample to the training sample of the signal processing model in the subsequent step, the signal processing model can refer to the information transmission habits of the data transmission node when encoding or decoding the transmission information, so as to facilitate the signal processing model to encode or decode while understanding the core semantics of the transmission information. The above-mentioned portrait data is mainly used to describe the specific mode of the data transmission node in the process of transmitting information. For example, the portrait data may include the specific communication scenario in which the data transmission node is located, as well as the information transmission style, etc.

[0048] Since GAI has content creation capabilities similar to humans, portrait data can be generated through GAI. In one embodiment, the training sample generation model includes a first sub-model, a second sub-model and a third sub-model, wherein the second sub-model at least includes a generative artificial intelligence model; the interaction data and feature data are input into the training sample generation model, and based on the output results of the training sample generation model, the portrait data of the data transmission node is generated, referring to Figure 2 As shown, steps S210 to S230 may be included:

[0049] Step S210, generating input data of the second sub-model according to the first sub-model, feature data, and interaction data.

[0050] The input data may include prompts, specifically, stem data and background data corresponding to the stem data; the stem data may be used to describe the content of the question, and the background data may include contextual information such as relevant preconditions, historical information, environment settings or basic data of the question described by the stem data, so that GAI can more easily understand the intention of the input data. For example, each input data may include a stem data q∈Q and the target background data h corresponding to the stem data. q ∈H, Q is the set of all stem data, and H is the set of all background data.

[0051] The first sub-model is mainly used to adjust the content of the input data, so that the final input data can express the core semantics of the question with the minimum amount of data, so that GAI can generate response result data similar to the standard response data, where the standard response data can include the actual answer to the question described by the input data. The answer result data may include the answer generated by GAI based on the question described in the input data. q .

[0052] In one embodiment, the input data of the second sub-model is generated according to the first sub-model, the feature data, and the interaction data, referring to Figure 3 As shown, steps S310 to S340 may be included:

[0053] Step S310, obtain the stem data from the interactive data through the first sub-model, and obtain a set of candidate background data associated with the stem data based on the interactive data and the feature data; each candidate background data in the candidate background data set is arranged in reverse order according to the degree of association with the stem data.

[0054] The candidate background data set may be a virtual container for storing background data related to the question stem data; the step of obtaining the candidate background data set associated with the question stem data based on the interaction data and the feature data may include the following steps:

[0055] Extract keywords from the question data, and obtain multiple background data based on feature data and interaction data;

[0056] Obtain similarity matching results between the keywords and each background data, and determine the degree of association between the background data and the keywords based on the similarity matching results and the frequency of occurrence of the keywords in the question stem data;

[0057] The background data corresponding to the correlation degree meeting the preset conditions is determined as candidate background data;

[0058] The candidate background data are arranged in reverse order according to the degree of association to obtain a candidate background data set.

[0059] Among them, the above-mentioned correlation degree is used to characterize the relevance between background data and question data; the preset conditions are used as the basis for evaluating whether the background data belongs to candidate background data. For example, the preset conditions may include a correlation degree greater than a preset correlation degree threshold, or selecting a preset number of background data with the highest correlation degree as candidate background data.

[0060] For example, the content can be split based on the feature data and the interaction data to obtain multiple clauses, each of which is a background data; and keywords are extracted from the question stem data to obtain the word vector corresponding to each keyword, and the frequency of occurrence of each keyword in the question stem data is calculated; by calculating the cosine distance between the word vector corresponding to each keyword and each background data, similarity matching is performed on the keywords and the background data to obtain the similarity matching result of each background data; finally, the frequency of occurrence corresponding to any keyword and the similarity matching result (such as similarity) between any keyword and each background data are combined to jointly determine the degree of association between any keyword and each background data, such as the product of the similarity corresponding to the background data and the frequency of occurrence of the keyword corresponding to the background data to determine the degree of association of the background data; among all the background data, the background data corresponding to the degree of association that meets the preset conditions is selected as the candidate background data, and these candidate background data are arranged from large to small according to the degree of association to obtain the candidate background data set T q =(δ1,δ2,…,δ n ), n is a positive integer.

[0061] By determining the candidate background data set based on the degree of correlation between the stem data and the background data, the background data most relevant to the stem data can be screened out, providing the possibility for quickly screening out the target background data from the candidate background data set in subsequent steps, effectively improving the efficiency of obtaining the target background data.

[0062] Continue to refer Figure 3 In step S320, when the first sub-model generates a target background data set associated with the question stem data for the first time, the first N candidate background data in the candidate background data set are determined as target background data; when correction information is received, the target background data is determined in the candidate background data set based on the correction information, and N is a positive integer.

[0063] That is to say, when the target background data corresponding to the question stem data is obtained for the first time, the first N candidate background data in the candidate background set can be directly determined as the target background data; if the data to be verified obtained through the input data passes the verification after the input data is generated based on the question stem data and the target background data, the portrait data is directly obtained based on the input data; if the data to be verified obtained through the input data does not pass the verification, the third sub-model will generate correction information so that the first sub-model continues to obtain the target background data in the candidate background data set based on the correction information. The above correction information can be used to instruct the first sub-model to adjust the arrangement order of the candidate background data.

[0064] In one implementation, the above-mentioned determining the target background data in the candidate background data set based on the correction information may include the following steps:

[0065] The arrangement order of each candidate background data in the candidate background data set is updated based on the correction information and the question stem data, and the first N candidate background data in the updated candidate background data set are determined as the target background data.

[0066] Updating the arrangement order of candidate background data through correction information is conducive to selecting candidate background data required to make the response result data close to the standard response data as target background data, thereby improving the efficiency of obtaining the target background data.

[0067] Continue to refer Figure 3 In step S330, the target background data is popped out from the candidate background data set, and the popped-out target background data is added to the target background data set.

[0068] That is to say, in each process of acquiring target background data, the candidate background data that is most associated with the question stem data and is conducive to generating standard response data can be added to the target background data set.

[0069] Step S340, combining the question stem data with the target background data in the target background data set, generating prompt words for the generative artificial intelligence model, and determining input data based on the prompt words.

[0070] After obtaining the question stem data and the target background data set, the question stem data and the target background data can be combined into a coherent prompt word to obtain input data.

[0071] based on Figure 3 The method can extract the target background data that is most conducive to generating standard answer data from the background data related to the question stem data, eliminate the redundant background data in the input data that has a low correlation with the question stem data, and realize the rapid acquisition of the core background data required for the standard answer data; and to a certain extent solve the information redundancy problem that exists in the process of GAI in related technologies generating answer result data based on the global knowledge in the public domain.

[0072] Continue to refer Figure 2 In step S220, the response result data of the generative artificial intelligence model to the input data is obtained, and the second sub-model is used to generate the data to be verified based on the response result data.

[0073] Exemplarily, the second sub-model M can obtain the given question stem data q and the target background data by parsing the input data, and obtain the standard answer data The probability distribution of And the response result data a q The probability distribution p(q,a q |h q , M); to determine the response result data a through the probability distribution of the response result data q .

[0074] In one implementation, the above-mentioned generation of the data to be verified based on the response result data may include the following steps:

[0075] Obtain the question stem data corresponding to the answer result data, and the standard answer data corresponding to the question stem data;

[0076] The stem data is matched with the standard answer data to obtain a first data pair, and the stem data is matched with the answer result data to obtain a second data pair, and the data to be verified is obtained based on the first data pair and the second data pair.

[0077] It should be noted that after obtaining the response result data, the data set can also be sampled through the third sub-model to obtain standard response data for the question stem data, thereby generating data to be verified based on the response result data and the standard response data corresponding to the same question stem data. In other words, the data to be verified in the present disclosure can be generated by the second sub-model or by the third sub-model.

[0078] The above-mentioned standard response data may include standard response results of the stem data obtained from the actual communication scenario. The first data pair and the second data pair are respectively obtained through the stem data and its corresponding standard response data and response result data to obtain the data to be verified, which is conducive to speeding up the verification process and thus improving the efficiency of generating portrait data.

[0079] Continue to refer Figure 2 In step S230, the data to be verified is verified by the third sub-model. If the verification fails, correction information output by the third sub-model is obtained, so that the first sub-model updates the input data according to the correction information and regenerates the data to be verified until the verification passes; and if the verification passes, the portrait data is generated based on the data to be verified and the input data corresponding to the data to be verified or the updated input data.

[0080] That is to say, the third sub-model can generate portrait data when the verification data passes the verification, and generate correction information when the verification fails, and feed back the correction information to the first sub-model, so that the first sub-model adjusts the target background data set. Among them, the third sub-model is equivalent to a discriminator, which is used to score the generated response result data based on the standard response data to drive the first sub-model to optimize the target background data set, ensure that there is a high correlation between the input data and the response result data, thereby improving the overall performance of the semantic communication system. The portrait data can include the response result data when the verification passes and its corresponding input data, which is used to add a personalized feature description of the data transmission node to the original training data to enhance the original training data.

[0081] The above-mentioned data to be verified includes the question stem data, and the answer result data and standard answer data corresponding to the question stem data; the data to be verified is verified by the third sub-model, referring to Figure 4 As shown, steps S410 to S420 may be included:

[0082] Step S410, obtaining standard answer data and answer result data corresponding to the same question stem data based on the data to be verified;

[0083] Step S420, calculate the similarity between the standard response data and the response result data corresponding to the same question stem data to verify the data to be verified.

[0084] Exemplarily, the data to be verified is input into the third sub-model, including the answer result data a of the question data q q and standard response data The third sub-model can score the response result data by calculating the similarity between the standard response data and the response result data. For example, the score of the response result data can be between 0 and 1. The higher the score, the closer the response result data is to the standard response data. In addition, the third sub-model can be used to vectorize the question stem data and its corresponding standard response data and response result data, and then input the generated vector into the middle layer with ReLU activation function in the third sub-model, and the output layer with sigmoid function as activation function to obtain the verification result and correction information; the third sub-model uses a loss function as shown in formula (1), and the loss function l d Update the weight w of the third sub-model according to the direction of gradient ascent g :

[0085]

[0086] The above verification may pass, and the score obtained by the similarity between the standard answer data and the answer result data corresponding to the same question stem data may be greater than a preset score threshold; the present disclosure does not specifically limit the specific value of the score threshold, for example, the score threshold may be 0.9.

[0087] In one embodiment, the above-mentioned image data is generated based on the data to be verified and the input data corresponding to the data to be verified or the updated input data, referring to Figure 5 As shown, steps S510 to S530 may be included:

[0088] Step S510, obtaining response result data in the data to be verified;

[0089] Step S510, matching the input data with the response result data to obtain a data record;

[0090] Step S510, adding the data record to the data set corresponding to the portrait data to obtain the portrait data.

[0091] based on Figure 5 The method can generate corresponding data records according to the verified data to be verified, and obtain the portrait data of the data transmission node based on all the data records corresponding to the data transmission node, so that the portrait data can describe the personalized information of the data transmission node with a smaller data volume.

[0092] For example, the structure of a training sample generation model disclosed in the present invention can refer to Figure 6 As shown, the generation stage may include a generator composed of a first sub-model and a second sub-model, which is used to generate data records in the portrait data; the third sub-model included in the discrimination stage is equivalent to a discriminator, which is used to score the generated response result data. The generator and the discriminator run alternately to promote adversarial training, thereby generating portrait data; the objective function of the GAI network D included in the second sub-model can refer to formula (2):

[0093]

[0094] Among them, E represents expectation, represents the probability distribution of the standard response data corresponding to the data transmission node, P(H,Q) represents the probability distribution of the response result data, is the probability of correctly distinguishing the standard answer data output by the third sub-model, which can be obtained by sampling the data labels in the data set storing the standard answer data, Q(a q ) represents the probability that the third sub-model identifies the response result data as standard response data.

[0095] based on Figure 2 The method continuously updates the input data through the correction information fed back by the third sub-model, removes the redundant background data in the input data, and generates the target background data required to obtain the standard response data. To a certain extent, it solves the problem in the related technology that it is difficult to generate supplementary training data according to the characteristics of the data transmission node, so that the obtained portrait data can accurately describe the communication scenario of the data transmission node with a smaller amount of data.

[0096] Continue to refer Figure 1 In step S130, the original training data corresponding to the signal processing model is updated according to the portrait data to obtain target training data.

[0097] Among them, the original training data may include training samples (TS), which are original data sets used to train signal processing models, such as semantic coding models and decoding models. The training samples do not include portrait data of data transmission nodes. For example, the original training data may include actual communication data, artificially generated communication data, and background data corresponding to the public domain to help the signal processing model learn to extract and reconstruct transmission information.

[0098] For example, the portrait data can be combined with the original training data to obtain a customized knowledge base corresponding to the data transmission node; or the original training data can be replaced by the portrait data to obtain the target training data. By training the encoding model with the knowledge base, the encoding model can fully understand the core meaning of the transmission information in the process of encoding the transmission information of the data transmission node.

[0099] After obtaining the target training data, continue to refer to Figure 1 In step S140, the signal processing model is trained using target training data so that the trained signal processing model, when receiving transmission information from a data transmission node, encodes or decodes the transmission information with reference to the portrait data.

[0100] Based on the above method, the original training data can be supplemented by the portrait data of the data transmission node to obtain the target training data. By training the signal processing model with the target training data, the reasoning ability of the signal processing model can be improved, the accuracy of the encoding is effectively improved, and the reliability of the decoding process is improved. It further improves the data transmission efficiency in the communication process, which helps to improve the robustness and adaptability of the communication system.

[0101] In one embodiment, the data transmission node includes a transmitting end and a receiving end, then refer to Figure 7 As shown, the data processing method of this exemplary embodiment may include steps S701 to S712:

[0102] Step S701, obtaining historical interaction data or real-time interaction data of a data transmission node, where the real-time interaction data includes question content input by the data transmission node to the GAI in real time;

[0103] Step S702, obtaining question stem data from historical interaction data or real-time interaction data through the first sub-model, and obtaining a set of candidate background data corresponding to the question stem data;

[0104] Step S703, determining the first candidate background data in the candidate background data set as the target background data, and adding the target background data to the target background data set;

[0105] Step S704, generating input data based on the question data and the target background data set, and inputting the input data into the GAI to obtain answer result data;

[0106] Step S705, obtaining the standard response data corresponding to the question stem data, and obtaining the question stem data vector, and its corresponding standard response data vector, and response result data vector; generating the data to be verified based on the question stem data vector, and its corresponding standard response data vector, and response result data vector;

[0107] Step S706, input the data to be verified into the third sub-model, and the third sub-model scores the response result data vector according to the similarity between the standard response data vector and the response result data vector;

[0108] Step S707: If the score of the response result data vector is greater than the preset score threshold, jump to step S708; otherwise, jump to step S709;

[0109] Step S708, generating portrait data based on the input data and the response result data;

[0110] Step S709, updating the arrangement order of each candidate background data in the candidate background data set according to the correction information;

[0111] Step S710, merging the updated portrait data with the original training data to obtain target training data;

[0112] Step S711, training the semantic encoding model and the semantic decoding model using target training data to obtain trained semantic encoding model and semantic decoding model;

[0113] Step S712, the semantic coding model encodes the transmission information of the sending end with reference to the portrait information of the sending end to obtain a symbol stream S; the semantic decoding model decodes the symbol stream S with reference to the portrait information of the receiving end to obtain reconstructed information.

[0114] based on Figure 7 The method can make the generated symbol stream S reflect the core meaning of the transmitted information, so that the reconstructed information can be understood by the receiving end as the above core meaning, thereby improving the reliability of the communication process.

[0115] In one embodiment, the data processing method in this exemplary embodiment can be applied to Figure 8 The semantic communication system shown in Figure 8 As shown, the data transmission nodes in the semantic communication system include the terminal device U used by the sending end user. a , and the terminal device U used by the receiving user b ; That is, the sending end user can use the terminal device U a Send transmission information W, the receiving user can use the terminal device U b Receive reconstruction information The terminal device can be a smart phone, tablet computer or other electronic device; a With terminal equipment U b The signal processing model can include a semantic encoding model and a semantic decoding model trained on the cloud; Figure 8 The end-to-end semantic communication model of the semantic communication system can be expressed as the following formula (3) and formula (4):

[0116]

[0117]

[0118] Where T is the original training data; represents the background data of the sender learned from the original training data; represents the background data of the receiver learned from the original training data; S represents the semantic encoding model C . (·) The symbol stream obtained by encoding the transmission information W; S1 represents the symbol stream received by the receiving end; Represents the semantic decoding model C b (·) Information obtained by decoding based on the symbol stream S1.

[0119] refer to Figure 8The semantic communication system shown can obtain the feature data of the sender and the receiver respectively according to the sender knowledge base and the receiver knowledge base, and generate a common knowledge base according to the feature data of the sender and the receiver. For example, a common knowledge base can be generated according to the intersection of the feature data of the sender and the feature data of the receiver, and then based on the common knowledge base, the interaction data of the sender and the interaction data of the receiver, the portrait data of the sender and the portrait data of the receiver are generated respectively; further, the target training data corresponding to the sender can be obtained, and the semantic encoding model can be trained by the target training data corresponding to the sender; and after obtaining the target training data corresponding to the receiver, the semantic decoding model can be trained by the target training data corresponding to the receiver; so that the trained semantic encoding model can encode the transmission information with reference to the portrait data of the sender, and the trained semantic decoding model can realize the decoding process with reference to the portrait data of the receiver.

[0120] In addition, the degree of enhancement of the image data to the original training data can be evaluated according to formula (5) and formula (6):

[0121]

[0122] Wherein, W represents the transmission information of the sender; represents the reconstructed information received by the receiving end; S is the symbol stream obtained by encoding the transmission information by the coding model with reference to the image data of the sending end; H(S) is the information entropy of the symbol stream S; Represents the background data learned by the sender based on the original training data; Represents the background data learned by the receiver based on the original training data; is the transmission symbol stream S, and The conditional mutual information between W and The mutual information between them.

[0123] Based on the above formula (3), we can get Then, the enhancement degree I is calculated based on the following formula (4): d :

[0124]

[0125] in, represents the portrait data of the transmitting end obtained based on this exemplary embodiment; This is the portrait data of the receiving end obtained based on this exemplary embodiment.

[0126] I d>0 means that the target training data with portrait data is added, the influence of redundant background data is removed, and the original training data has a data enhancement effect.

[0127] based on Figure 8 The semantic communication system shown has higher semantic accuracy, robustness and adaptability compared to existing semantic communication systems.

[0128] The exemplary embodiment of the present disclosure also provides a data processing device, referring to Fig. 9 As shown, the device is applied to a communication system including at least a data transmission node and a signal processing model, and may include:

[0129] The data acquisition module 910 is configured to acquire the interaction data between the data transmission node and the generative artificial intelligence model, and the characteristic data of the data transmission node;

[0130] The portrait data generation module 920 is configured to input the interaction data and the feature data into the training sample generation model, and generate the portrait data of the data transmission node based on the output of the training sample generation model;

[0131] The training data updating module 930 is configured to update the original training data corresponding to the signal processing model according to the portrait data to obtain the target training data;

[0132] The signal processing model training module 940 is configured to train the signal processing model through target training data so that the trained signal processing model can encode or decode the transmission information with reference to the portrait data when receiving the transmission information from the data transmission node.

[0133] In one embodiment, the training sample generation model includes a first sub-model, a second sub-model and a third sub-model, wherein the second sub-model includes at least a generative artificial intelligence model; inputting the interaction data and the feature data into the training sample generation model, and generating the portrait data of the data transmission node based on the output result of the training sample generation model, may include:

[0134] Generating input data of a second sub-model according to the first sub-model, the feature data, and the interaction data;

[0135] Obtaining response result data of the generative artificial intelligence model to the input data, and using the second sub-model to generate data to be verified based on the response result data;

[0136] The data to be verified is verified by the third sub-model. If the verification fails, correction information output by the third sub-model is obtained, so that the first sub-model updates the input data according to the correction information and regenerates the data to be verified until the verification passes; and if the verification passes, the portrait data is generated based on the data to be verified and the input data corresponding to the data to be verified or the updated input data.

[0137] In one embodiment, the step of generating the input data of the second sub-model according to the first sub-model, the feature data, and the interaction data may include:

[0138] Obtaining question stem data from the interactive data through the first sub-model, and obtaining a set of candidate background data associated with the question stem data based on the interactive data and the feature data; arranging each candidate background data in the set of candidate background data in reverse order according to the degree of association with the question stem data;

[0139] When the first sub-model generates a target background data set associated with the question stem data for the first time, the first N candidate background data in the candidate background data set are determined as the target background data; when the correction information is received, the target background data is determined in the candidate background data set based on the correction information, where N is a positive integer;

[0140] Popping the target background data from the candidate background data set, and adding the popped target background data to the target background data set;

[0141] Combine the question stem data with the target background data in the target background data set to generate prompt words for the generative artificial intelligence model, and determine the input data based on the prompt words.

[0142] In one implementation, determining the target background data in the candidate background data set based on the correction information may include:

[0143] The arrangement order of each candidate background data in the candidate background data set is updated based on the correction information and the question stem data, and the first N candidate background data in the updated candidate background data set are determined as the target background data.

[0144] In one implementation, the above-mentioned generating the data to be verified based on the response result data may include:

[0145] Obtain the question stem data corresponding to the answer result data, and the standard answer data corresponding to the question stem data;

[0146] The stem data is matched with the standard answer data to obtain a first data pair, and the stem data is matched with the answer result data to obtain a second data pair, and the data to be verified is obtained based on the first data pair and the second data pair.

[0147] In one embodiment, the data to be verified includes question stem data, and answer result data and standard answer data corresponding to the question stem data; verifying the data to be verified by the third sub-model may include:

[0148] Based on the data to be verified, obtain standard answer data and answer result data corresponding to the same question stem data;

[0149] Calculate the similarity between the standard answer data and the answer result data corresponding to the same question stem data to verify the data to be verified.

[0150] In one implementation, the generation of the portrait data based on the data to be verified and the input data or updated input data corresponding to the data to be verified may include:

[0151] Obtain the response result data in the data to be verified;

[0152] Associating the input data with the response result data to obtain associated data records;

[0153] Add the associated data record to the data set corresponding to the portrait data to obtain the portrait data.

[0154] The specific details of each part of the above device have been described in detail in the implementation method part, so they will not be repeated here.

[0155] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the exemplary embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0156] The exemplary embodiments of the present disclosure further provide a computer program product, which includes a computer program. When the computer program is executed by a processor, the above data processing method is implemented.

[0157] Computer programs for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The computer program may be executed entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0158] In one embodiment, the computer program product may be an intangible product including a computer program. Exemplarily, the computer program product may be implemented as a virtual digital product, such as a digital file storing an executable file, an installation package, etc. of the computer program.

[0159] The code of the computer program can be written in one or more programming languages. Programming languages ​​such as C language, Java, C++, etc. The program code can be executed entirely on the user computing device, or partially on the user computing device, or as a separate software package, or partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, such as a local area network (LAN), a wide area network (WAN), etc., or can be connected to an external computing device (e.g., an Internet connection provided by an operator).

[0160] Computer programs can be carried or transmitted through electrical, magnetic, optical, electromagnetic, infrared and other signals. Electronic devices can convert signals carrying computer programs into digital signals to run computer programs. When a computer program runs on an electronic device, its code is used to enable the electronic device to execute (more specifically, the processor of the electronic device can execute) the method steps of various exemplary embodiments of the present disclosure, such as the above-mentioned data processing method, which includes the following steps: obtaining the interaction data between the data transmission node and the generative artificial intelligence model, and the feature data of the data transmission node; inputting the interaction data and the feature data into the training sample generation model, and generating the portrait data of the data transmission node based on the output of the training sample generation model; updating the original training data corresponding to the signal processing model according to the portrait data to obtain the target training data; training the signal processing model through the target training data so that the trained signal processing model, when receiving the transmission information of the data transmission node, encodes or decodes the transmission information with reference to the portrait data.

[0161] By executing the above method steps through a computer program, on the one hand, the portrait data of the data transmission node of the communication system is generated, and the training data of the signal processing model is enriched by the portrait data, so that the trained signal processing model can refer to the portrait data of the data transmission node for encoding or decoding; this solution solves the problem that the signal processing model cannot refer to the characteristics of the transmitting end or the receiving end for encoding or decoding to a certain extent, improves the reasoning ability of the signal processing model, enables the encoder to refer to the portrait data and infer the original intention of the transmission information for encoding, effectively improves the accuracy of the encoding, and enables the decoder to refer to the portrait data of the data transmission node for decoding, so that the decoded information is easier to be understood by the data transmission node, thereby improving the reliability of the decoding process, thereby effectively improving the data transmission efficiency during the communication process. On the other hand, the original training data of the signal processing model is updated by the portrait data of the data transmission node to obtain the target training data, and the signal processing model is trained based on the target training data, effectively improving the richness and diversity of the training samples of the signal processing model.

[0162] The exemplary embodiments of the present disclosure also provide an electronic device, which interacts with the above-mentioned communication system to indicate the signal processing model in the above-mentioned communication system, and encodes or decodes the image data of the data transmission node with reference to the image data of the data transmission node. The electronic device may include a processor and a memory. The memory stores executable instructions of the processor, such as a computer program. The processor executes the method steps of various exemplary embodiments of the present disclosure by executing the executable instructions.

[0163] Reference below Fig.10 , the electronic device is exemplarily described in the form of a general-purpose computing device. It should be understood that Fig.10 The electronic device 1000 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0164] like Fig.10 As shown, the electronic device 1000 may include: a processor 1010 , a memory 1020 , a bus 1030 , an I / O (input / output) interface 1040 , and a network adapter 1050 .

[0165] The processor 1010 may include one or more processing units. For example, the processor 1010 may include an AP (Application Processor), a modem processor, a GPU (Graphics Processing Unit), an ISP (Image Signal Processor), a controller, an encoder, a decoder, a DSP (Digital Signal Processor), a baseband processor and / or an NPU (Neural-Network Processing Unit), etc.

[0166] Processor 1010 can be used to execute executable instructions stored in memory 1020, such as the above-mentioned data processing method, which includes the following steps: obtaining interaction data between the data transmission node and the generative artificial intelligence model, as well as feature data of the data transmission node; inputting the interaction data and the feature data into the training sample generation model, and generating portrait data of the data transmission node based on the output of the training sample generation model; updating the original training data corresponding to the signal processing model according to the portrait data to obtain target training data; training the signal processing model with the target training data so that the trained signal processing model, when receiving the transmission information of the data transmission node, encodes or decodes the transmission information with reference to the portrait data.

[0167] By executing the above method steps through a computer program, on the one hand, the portrait data of the data transmission node of the communication system is generated, and the training data of the signal processing model is enriched by the portrait data, so that the trained signal processing model can refer to the portrait data of the data transmission node for encoding or decoding; this solution solves the problem that the signal processing model cannot refer to the characteristics of the transmitting end or the receiving end for encoding or decoding to a certain extent, improves the reasoning ability of the signal processing model, enables the encoder to refer to the portrait data and infer the original intention of the transmission information for encoding, effectively improves the accuracy of the encoding, and enables the decoder to refer to the portrait data of the data transmission node for decoding, so that the decoded information is easier to be understood by the data transmission node, thereby improving the reliability of the decoding process, thereby effectively improving the data transmission efficiency during the communication process. On the other hand, the original training data of the signal processing model is updated by the portrait data of the data transmission node to obtain the target training data, and the signal processing model is trained based on the target training data, effectively improving the richness and diversity of the training samples of the signal processing model.

[0168] The memory 1020 may include a volatile memory, such as a RAM 1021, a cache unit 1022, and may also include a non-volatile memory, such as a ROM 1023. The memory 1020 may also include one or more program modules 1024, such program modules 1024 include but are not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or a combination thereof may include the implementation of a network environment. For example, the program module 1024 may include each module in the above-mentioned apparatus 900.

[0169] The bus 1030 is used to realize the connection between different components of the electronic device 1000, and may include a data bus, an address bus, and a control bus.

[0170] The electronic device 1000 can communicate with one or more external devices 1100 (eg, a keyboard, a mouse, an external controller, etc.) through the I / O interface 1040 .

[0171] The electronic device 1000 can communicate with one or more networks through the network adapter 1050. For example, the network adapter 1050 can provide mobile communication solutions such as 3G / 4G / 5G, or provide wireless communication solutions such as wireless LAN, Bluetooth, near field communication, etc. The network adapter 1050 can communicate with other modules of the electronic device 1000 through the bus 1030.

[0172] although Fig.10Not shown, other hardware and / or software modules may also be provided in the electronic device 1000, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0173] As can be seen from the above, the technical solution of the present disclosure can be implemented as a method, an apparatus, a system, a computer program product, etc. Those skilled in the art can understand that various aspects of the present disclosure can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software, such as being respectively referred to as a "circuit", "module" or "system".

[0174] It should be understood that the present disclosure is not limited to the specific method steps or structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. Those skilled in the art will easily think of other embodiments based on the specific embodiments provided by the present disclosure. Therefore, the specific embodiments provided by the present disclosure are only exemplary, and the scope and spirit of the present disclosure are indicated by the claims, and any variations, uses or adaptive changes of the present disclosure should be covered, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the technical field that are not disclosed in the present disclosure.

Claims

1. A data processing method, It is characterized in that The method is applied to a communication system including at least a data transmission node and a signal processing model, and the method includes: Obtaining interaction data between the data transmission node and the generative artificial intelligence model, as well as feature data of the data transmission node; Inputting the interaction data and the feature data into a training sample generation model, and generating the portrait data of the data transmission node based on the output of the training sample generation model; Updating original training data corresponding to the signal processing model according to the portrait data to obtain target training data; The signal processing model is trained using the target training data so that the trained signal processing model, when receiving the transmission information from the data transmission node, encodes or decodes the transmission information with reference to the portrait data.

2. The method according to claim 1, It is characterized in that The training sample generation model includes a first sub-model, a second sub-model and a third sub-model, wherein the second sub-model at least includes the generative artificial intelligence model; the step of inputting the interaction data and the feature data into the training sample generation model, and generating the portrait data of the data transmission node based on the output result of the training sample generation model, includes: generating input data of the second sub-model according to the first sub-model, the feature data, and the interaction data; Obtaining response result data of the generative artificial intelligence model to the input data, and generating data to be verified based on the response result data using the second sub-model; The data to be verified is verified by the third sub-model. If the verification fails, correction information output by the third sub-model is obtained, so that the first sub-model updates the input data according to the correction information and regenerates the data to be verified until the verification passes; and if the verification passes, the portrait data is generated based on the data to be verified and the input data corresponding to the data to be verified or the updated input data.

3. The method according to claim 2, It is characterized in that Generating input data of the second sub-model according to the first sub-model, the feature data, and the interaction data includes: Obtaining question stem data from the interaction data through the first sub-model, and obtaining a set of candidate background data associated with the question stem data based on the interaction data and the feature data; arranging each of the candidate background data in the set of candidate background data in reverse order according to the degree of association with the question stem data; In the case where the first sub-model generates the target background data set associated with the question stem data for the first time, the first N candidate background data in the candidate background data set are determined as the target background data; in the case where the correction information is received, the target background data is determined in the candidate background data set based on the correction information, where N is a positive integer; Popping out the target background data from the candidate background data set, and adding the popped-out target background data to the target background data set; The prompt words of the generative artificial intelligence model are generated by combining the question stem data with the target background data in the target background data set, and the input data is determined based on the prompt words.

4. The method according to claim 3, It is characterized in that The determining target background data in the candidate background data set based on the correction information includes: The arrangement order of each candidate background data in the candidate background data set is updated based on the correction information and the question stem data, and in the updated candidate background data set, the first N candidate background data are determined as the target background data.

5. The method according to claim 2, It is characterized in that The generating the data to be verified based on the response result data comprises: Obtaining the question stem data corresponding to the answer result data, and the standard answer data corresponding to the question stem data; The question stem data is matched with the standard answer data to obtain a first data pair, and the question stem data is matched with the answer result data to obtain a second data pair, and the data to be verified is obtained based on the first data pair and the second data pair.

6. The method according to claim 2, It is characterized in that The data to be verified includes question stem data, and answer result data and standard answer data corresponding to the question stem data; The verifying the data to be verified by using the third sub-model includes: Based on the data to be verified, obtain standard answer data and answer result data corresponding to the same question stem data; The similarity between the standard response data and the response result data corresponding to the same question stem data is calculated to verify the data to be verified.

7. The method according to claim 2, It is characterized in that The generating the portrait data based on the data to be verified and the input data corresponding to the data to be verified or the updated input data includes: Obtaining response result data from the data to be verified; Associating the input data with the response result data to obtain an associated data record; The associated data record is added to the data set corresponding to the portrait data to obtain the portrait data.

8. A data processing device, It is characterized in that The device is applied to a communication system including at least a data transmission node and a signal processing model, and the device includes: A data acquisition module, configured to acquire interaction data between the data transmission node and the generative artificial intelligence model, and feature data of the data transmission node; A portrait data generation module, configured to input the interaction data and the feature data into a training sample generation model, and generate portrait data of the data transmission node based on an output of the training sample generation model; A training data updating module is configured to update the original training data corresponding to the signal processing model according to the portrait data to obtain target training data; The signal processing model training module is configured to train the signal processing model through the target training data so that the trained signal processing model can encode or decode the transmission information with reference to the portrait data when receiving the transmission information of the data transmission node.

9. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device, It is characterized in that include: processor; A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 7 by executing the executable instructions.

Citation Information

Cited By

  • Method for generating realistic features and a module generation system using the same

    TWI924433B