A cloud-edge collaborative system, data transmission method and storage medium

Abstract

The present application discloses a cloud-edge collaborative system, a data transmission method and a storage medium, and the method steps are as follows: obtaining a baseline weight for security analysis based on the text in the security management field; obtaining a user demand weight based on the user task requirement, and weighted averaging the two to form a total weight for the information category; constructing a target category entropy value based on a standard analysis case; combining the obtained total weight for the information category with the target category entropy value to generate a target entropy value matrix; calculating a predicted entropy value matrix based on the data output by the edge end; reconstructing or directly sending the data output by the edge end based on the difference between the predicted entropy value matrix and the target entropy value matrix. Beneficial effects of the present application: The present application can accurately balance security standards and user needs through dual weight calculation, avoiding the neglect of key information caused by an imbalance in the weight of information categories; at the same time, the entropy value matrix and the loop optimization mechanism can dynamically calibrate the output of the edge end to ensure that the data input to the large cloud model meets the information entropy threshold requirements.

Description

Technical Field

[0001] The present application relates to the field of data transmission technology, and in particular to a cloud-edge collaborative architecture, a data transmission method and a storage medium. Background Art

[0002] The cloud-edge collaborative architecture tightly integrates cloud computing with edge computing. By rationally allocating cloud computing and edge computing tasks, it enables the downward shift of cloud computing capabilities, extending cloud computing, cloud analytics, and other functions to the edge. The cloud-edge collaborative architecture primarily comprises the cloud, edge, and terminal devices. The cloud is responsible for tasks such as large-scale data processing, model training, global data analysis, and resource management, providing powerful computing power and storage resources. The edge, located close to data sources or terminal devices, is responsible for real-time data collection, preprocessing, local decision-making, and lightweight task processing, enabling rapid response to low-latency requirements. Terminal devices, including sensors, cameras, and IoT devices, are responsible for data generation and initial processing.

[0003] Current cloud-edge collaborative systems generally rely on prompt word templates to format edge outputs, which are then used as input for large cloud-based models. However, during data transmission, there is a lack of necessary information verification mechanisms. This lack of verification neither verifies whether the data meets the information entropy requirements of standard analysis nor confirms whether it meets the information entropy requirements of user task scenarios. This lack of a mechanism can lead to problems such as missing key information and biased analysis conclusions during analysis by large cloud-based models due to insufficient or biased information entropy in the input data. This seriously impacts the decision-making accuracy and task execution reliability of cloud-edge collaborative systems. Summary of the Invention

[0004] One of the objectives of the present application is to provide a data transmission method that can solve at least one of the defects in the above-mentioned background technology.

[0005] Another object of the present application is to provide a storage medium that can solve at least one of the defects in the above-mentioned background technology.

[0006] Another object of the present application is to provide a cloud-edge collaboration system that can solve at least one of the defects in the above-mentioned background technology.

[0007] To achieve at least one of the above purposes, the technical solution adopted in this application is: a data transmission method, comprising the following steps:

[0008] S100: Based on the security management domain text, weights of different information categories are obtained through distance calculation, which are used as baseline weights for security analysis. Based on the user task requirements, user demand weights that match the task scenario are generated through distance calculation. The baseline weights and user demand weights are weighted averaged to form the total weight of the information category.

[0009] S200: Perform internal entropy analysis on the standard analysis case to construct the target category entropy value; combine the obtained total weight of the information category with the target category entropy value to generate the target entropy value matrix required by the cloud-based large model;

[0010] S300: Based on the data output by the edge end, the corresponding category weights and category entropy values ​​are calculated, and then the corresponding prediction entropy matrix is ​​obtained by combining the two.

[0011] S400: Compare the difference between the obtained predicted entropy matrix and the target entropy matrix. If the difference between the two exceeds a set threshold, reconstruct the data output by the edge end and execute step S500; otherwise, pass the data output by the edge end to the cloud-based large model;

[0012] S500: Repeat steps S300 and S400 on the reconstructed data.

[0013] Preferably, in step S100, the specific process of obtaining the baseline weight and user demand weight based on distance calculation is: encoding the text data through a multi-head attention mechanism to obtain a sentence vector and a category vector; calculating the distance of each sentence to different categories based on the obtained vector data, and accumulating and summing the distance values ​​according to the category; performing probability calculation using the softmax function based on the obtained summation result to obtain the score value of each category and construct a weight vector; wherein, when obtaining the weight vector corresponding to the baseline weight, the text data comes from the text in the field of security management, and when obtaining the weight vector corresponding to the user demand weight, the text data comes from the user task demand.

[0014] Preferably, both the sentence vector and the category vector are represented in the form of a sentence vector. The acquisition of the sentence vector and the category vector includes the following process: converting the text data into a vector form that the BERT model can understand as the model input, including word embedding, paragraph embedding, and position embedding; after obtaining the vector form of the text data, input it into the BERT model, and then encode it through a multi-head attention mechanism; extracting the embedding vector of the CLS token from the output of the BERT model; and enhancing the output of the BERT model by adding a pooling layer to obtain the final sentence vector.

[0015] Preferably, the calculation formula for the cumulative distance value from each statement to the category is as follows:

[0016] ;

[0017] Among them, cos_dist represents the semantic distance from each sentence to the category, a i represents the i-th category, b j Indicates the jth statement, the maximum number of statements is n, li Represents the sum of the semantic distances corresponding to the i-th category.

[0018] Preferably, the calculation formula of the weight vector is as follows:

[0019] ;

[0020] ;

[0021] W0=[w1,w2,……,w k ];

[0022] Among them, s i represents the inverse of the sum of semantic distances corresponding to the i-th category, w i Represents the score value of each category, W0 represents the weight vector, and k represents the maximum number of categories.

[0023] Preferably, in step S200, the acquisition of the target category entropy value includes the following process: encoding the entire document of the standard analysis case into a high-dimensional semantic vector through a pre-trained BERT model; extracting multiple keywords for each category from the standard analysis case; generating corresponding semantic vectors for all extracted keywords through the BERT model, and the semantic vectors are consistent with the dimensions of the high-dimensional semantic vectors; based on the obtained semantic vectors, calculating the two-dimensional distance matrix from each keyword to the corresponding category; and calculating the target category entropy value based on the number of obtained keywords and the two-dimensional distance matrix corresponding to each keyword.

[0024] Preferably, the target category entropy W b The calculation expression is as follows:

[0025] ;

[0026] ;

[0027] W b =[w i ´], i={1, 2,...,k};

[0028] Among them, w i ´ represents the entropy value corresponding to the i-th category, m represents the number of keywords, w i,x represents the entropy value of the xth keyword corresponding to the i-th category, μ represents the average entropy value corresponding to all keywords, and s i,x Represents the inverse of a two-dimensional distance matrix.

[0029] Preferably, in step S400, when the difference between the predicted entropy matrix and the target entropy matrix is ​​greater than a set threshold, reconstruction is completed by increasing the text vocabulary of the edge output data.

[0030] A storage medium stores a computer program, which implements the above-mentioned data transmission method when executed by a processor.

[0031] A cloud-edge collaborative system includes the above-mentioned storage medium.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] This application can accurately balance security standards and user needs through dual weight calculation, avoiding the neglect of key information caused by imbalance in information category weights; at the same time, the entropy matrix and loop optimization mechanism can dynamically calibrate the output of the edge end to ensure that the data input into the large cloud model meets the information entropy threshold requirements, reducing the deviation of the analysis conclusion and information loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a schematic diagram of the overall workflow of this application. DETAILED DESCRIPTION

[0035] Below, the present application is further described in conjunction with specific implementation methods. It should be noted that, in the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like are intended to mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0036] In the description of this application, it should be noted that for directional words, such as the terms "center", "horizontal", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and so on, indicating the orientation and position relationship are based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing this application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and cannot be understood as limiting the specific scope of protection of this application.

[0037] It should be noted that the terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0038] In this application, unless otherwise expressly specified or limited, terms such as "mounted," "connected," "connect," and "fixed" should be understood in a broad sense. For example, they may refer to connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on specific circumstances.

[0039] In this application, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Moreover, a first feature being "above," "above," and "above" a second feature may include the first feature being directly above or obliquely above the second feature, or may simply mean that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may include the first feature being directly below or obliquely below the second feature, or may simply mean that the first feature is lower in level than the second feature.

[0040] The terms "comprises" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units expressly listed, but may include other steps or units not expressly listed or inherent to such process, method, product or apparatus.

[0041] One aspect of the present application provides a data transmission method, such as Figure 1 As shown, one of the preferred embodiments includes the following steps:

[0042] S100: Obtain weights of different information categories through distance calculation based on security management domain texts as baseline weights for security analysis; generate user demand weights that match task scenarios through distance calculation based on user task requirements; and perform weighted averaging of the baseline weights and user demand weights to form the total weight of the information category.

[0043] S200: Perform internal entropy analysis on the standard analysis case to construct the target category entropy value; combine the obtained total weight of the information category with the target category entropy value to generate the target entropy value matrix required by the cloud-based large model.

[0044] S300: Based on the data output by the edge end, the corresponding category weights and category entropy values ​​are calculated, and then the corresponding prediction entropy value matrix is ​​obtained by combining the two.

[0045] S400: Compare the difference between the obtained predicted entropy matrix and the target entropy matrix. If the difference between the two exceeds the set threshold, reconstruct the data output by the edge end and execute step S500; otherwise, pass the data output by the edge end to the cloud large model.

[0046] S500: Repeat steps S300 and S400 on the reconstructed data.

[0047] It should be known that the technical solution of this application is faced with multiple types of edge output data, which can be data output by a visual model or data output by a sensor model, etc. The specific type can be selected according to actual needs.

[0048] It can be understood that the technical solution of the present application forms a closed-loop optimization through a dual weight system and an entropy verification mechanism. First, the weights of information categories are calculated separately based on security field standards and user needs. After weighted averaging, the importance of information in different dimensions can be balanced to avoid weight deviations caused by a single standard. Secondly, a two-dimensional entropy matrix is ​​constructed to convert abstract information entropy requirements into a quantifiable target matrix, providing a clear standard for subsequent large cloud model inputs. The entropy evaluation and distance calculation mechanism can quantify the information entropy deviation output by the edge in real time, and trigger cyclic optimization when the entropy of a certain type of information is insufficient, forming an "evaluation-comparison-optimization" feedback closed loop. This mechanism can not only ensure that the input data meets the dual entropy standards, but also make up for the lack of information through dynamic optimization, fundamentally solving the problem of analysis deviation caused by information entropy imbalance.

[0049] Compared to traditional methods, this application significantly improves the data quality and analysis accuracy of cloud-edge collaborative systems through a systematic information entropy verification mechanism. In practical applications, this solution can increase the information accuracy of cloud-edge collaborative decision-making by over 30%, reducing the risk of misjudgment due to insufficient data entropy. Especially in security management scenarios, it can effectively enhance the system's ability to identify abnormal events and respond reliably, providing more solid data support for intelligent decision-making.

[0050] In this embodiment, when performing step S100, in the security management scenario, security standards can be constructed by analyzing domain texts. Specifically, the requirements for security standards come from the analysis requirements of the cloud-based big model for the scenario and the customer's task requirements, and are specifically presented in the form of proportions. For example, the requirements for the security management field are: the proportion of scenario description is 0.1, the proportion of statistical information is 0.2, the proportion of normal information is 0.2, the proportion of abnormal information is 0.3, and the proportion of analytical information is 0.2; if the user requires more analysis, then: the proportion of scenario description is 0.1, the proportion of statistical information is 0.2, the proportion of normal information is 0.2, the proportion of abnormal information is 0.2, and the proportion of analytical information is 0.3. Based on different proportion requirements, the baseline weight of security analysis and the user demand weight are constructed respectively. The specific construction process is as follows.

[0051] In this embodiment, the process of obtaining a baseline weight based on distance calculation in step S100 includes the following: encoding the text data using a multi-head attention mechanism to obtain a sentence vector and a category vector. Based on the obtained vector data, the distance between each sentence and different categories is calculated, and the distance values ​​are cumulatively summed according to the category. Based on the obtained summation results, a probability calculation is performed using the softmax function to obtain a score for each category and construct a weight vector, which is the baseline weight.

[0052] Specifically, the text data encoded by the multi-head attention mechanism comes from text in the security management field. Both sentence vectors and category vectors are represented by sentence vectors. The specific acquisition process includes the following: converting the text data into a vector form that the BERT model can understand as the model input, including word embedding (Token Embedding), paragraph embedding (Segment Embedding), and position embedding (Position Embedding). This can be specifically expressed by the following expression.

[0053] .

[0054] It's understandable that the multi-head attention mechanism is the core mechanism of the BERT model, which uses it to achieve deep bidirectional language representation learning. In the BERT model, token embedding maps the words in each input sentence to a high-dimensional vector space; segment embedding distinguishes different sentences, that is, judges the similarity between different sentences; and position embedding encodes the position information of the text. The Transformer architecture used by the BERT model itself lacks a recurrent structure and cannot capture the position information of the sequence, so the additional position embedding is required.

[0055] After obtaining the vector form of text data, it is input into the BERT model and then encoded through the multi-head attention mechanism. It can be specifically expressed by the following expression:

[0056] .

[0057] .

[0058] in, Represents the final output of the multi-head attention mechanism, which is obtained by concatenating the outputs of multiple attention heads and then transforming them through a linear transformation matrix W o Get the final output; Q, K, V represent query, key, and value matrices respectively; head i represents the i-th attention head, and the number of attention heads is h; Attention represents scaled dot product attention; , , Represents the weight matrix corresponding to each head; d k Indicates the dimensions along which the keys are adjacent, Used for scaling to prevent gradient disappearance.

[0059] After encoding the text data and obtaining the output, the embedding vector of CLStoken can be extracted from the output of the BERT model. ; Enhance the output of the BERT model by adding a pooling layer to obtain the final sentence vector Embedding vector and sentence vectors The expression is as follows:

[0060] .

[0061] .

[0062] Where W represents the learnable weight matrix and tanh represents the hyperbolic tangent activation function.

[0063] It should be known that both sentence vectors and category vectors are represented in the form of sentence vectors, that is, the BERT model based on the multi-head attention mechanism can output a sentence vector including a sentence vector and a category vector based on the input text data.

[0064] In this embodiment, after the sentence vector and category vector are constructed, the distance from each sentence to the category can be calculated. The specific calculation formula is as follows:

[0065] .

[0066] Among them, cos_dist represents the semantic distance from each sentence to the category, a i represents the i-th category, b j Indicates the jth statement, a i and b j The values ​​​​are all derived from the sentence vector ; The maximum number of statements is n, l i The specific number of categories can be set according to the actual needs of those skilled in the art.

[0067] After obtaining the sum of the semantic distances corresponding to each category, the information category weights corresponding to different categories can be obtained based on the softmax probability calculation, that is, the required benchmark weight W w The specific calculation formula is as follows:

[0068] ; ;W0=[w1,w2,……,w k ]=W w .

[0069] Among them, s i represents the inverse of the sum of semantic distances corresponding to the i-th category, w i Represents the score value of each category, W0 represents the weight vector, and k represents the maximum number of categories.

[0070] It should be known that when the user demand weight W is calculated based on the user task requirements, u The method used in the calculation of the benchmark weight W w The calculation method is the same as that of , the difference is that the text data comes from the user task requirements; the user task requirements are decomposed into sentences, the sentences are vectorized, the distance with the category is calculated, and finally the user demand weight W is obtained. u Finally, the statistical benchmark weight W w and user demand weight W u The total weight W of the information category is obtained by weighted averaging a .

[0071] In this embodiment, the acquisition of the target category entropy value in step S200 includes the following process: the entire document of the standard analysis case is encoded into a high-dimensional semantic vector through the pre-trained BERT model. Multiple keywords are extracted for each category from the standard analysis case; corresponding semantic vectors are generated for all extracted keywords through the BERT model, and the semantic vectors have the same dimension as the high-dimensional semantic vectors. Based on the obtained semantic vectors, a two-dimensional distance matrix is ​​calculated from each keyword to the corresponding category; and the target category entropy value is calculated based on the number of obtained keywords and the two-dimensional distance matrix corresponding to each keyword.

[0072] It can be understood that the two-dimensional distance matrix from each keyword to the corresponding category is the information entropy matrix of each category. The category information entropy matrix is ​​a two-dimensional rectangle, with dimension one being the information category dimension (scene information + complete statistical information + normal information + abnormal information + analysis information), and dimension two being the keyword information entropy dimension. The total weight of the information category obtained in the aforementioned content is a one-dimensional matrix; in order to facilitate the subsequent calculation of the target entropy matrix, the two-dimensional distance matrix needs to be one-dimensionalized to obtain a one-dimensional target category entropy value. For ease of understanding, the specific process of obtaining the target category entropy value will be described in detail below.

[0073] Specifically, the text data comes from a standard analysis case, and the calculation of the category information entropy value matrix corresponding to each keyword in the text includes the following steps:

[0074] First, we need to extract keywords, and extract multiple keywords from each category; the specific expression is as follows:

[0075] D = BERT(Document)∈R d ;w z =BERT(w z )∈R d (z=1, 2,…, n).

[0076] That is, the entire document is encoded into a high-dimensional semantic vector D through the pre-trained BERT model, where d represents the embedding dimension; then for each keyword w in the document z The semantic vector is generated by the BERT model, which is consistent with the dimension of the document vector D.

[0077] After obtaining the semantic vectors for each keyword, we can calculate the distance between the keyword and each category to form the required two-dimensional distance matrix, namely the category information entropy matrix. The specific calculation formula can be referred to the distance calculation formula for each sentence to the category mentioned above, so it will not be repeated here.

[0078] In this embodiment, the target category entropy value W can be calculated by softmax probability. b The specific calculation formula is as follows:

[0079] .

[0080] .

[0081] W b =[w i ´], i={1, 2,...,k}.

[0082] Among them, w i´ represents the entropy value corresponding to the i-th category, m represents the number of keywords, w i,x represents the entropy value of the xth keyword corresponding to the i-th category, μ represents the average entropy value corresponding to all keywords, and s i,x Represents the inverse of a two-dimensional distance matrix.

[0083] It is understandable that after obtaining the target category entropy value W b And the total weight of information category W a After that, the required target entropy matrix target can be calculated, target=W a ×W b .

[0084] In this embodiment, in step S300, the process of predicting the entropy value matrix based on the data output by the edge end is consistent with the calculation process of the target entropy value matrix. It is only necessary to replace the text data in the above-mentioned target entropy value matrix calculation process with the data output by the edge end. Therefore, the calculation process of the predicted entropy value matrix will not be repeated here.

[0085] In this embodiment, when performing step S400, when the difference between the predicted entropy matrix and the target entropy matrix is ​​greater than the set threshold, reconstruction is completed by increasing the text vocabulary of the edge output data; when the text vocabulary increases, the corresponding vocabulary richness TTR and information entropy H values ​​will increase, that is, the vocabulary diversity and distribution uniformity are improved.

[0086] It should be known that the calculation formula for vocabulary richness TTR is: TTR=|v| / N.

[0087] Among them, v represents the vocabulary size, and N represents the total number of words in the text; the higher the vocabulary richness TTR, the greater the vocabulary diversity.

[0088] The calculation formula of information entropy H is: .

[0089] Among them, p(w y ) represents word w y The probability of appearing in the text, the higher the value of information entropy H, the more uniform the vocabulary distribution.

[0090] It is understandable that the output data of the edge end may be reconstructed multiple times, and the text vocabulary increased each time may be a fixed ratio, such as 10% of the total number of words in the text. The text vocabulary increased each time may also not be a fixed ratio, such as gradually decreasing the text vocabulary each time, with the first increase in text vocabulary being relatively large. The specific selection can be made according to the actual needs of those skilled in the art.

[0091] Another aspect of the present application provides a storage medium on which a computer program is stored. A preferred embodiment thereof is that the computer program is used to implement the above-mentioned data transmission method when executed by a processor.

[0092] Another aspect of the present application provides a cloud-edge collaborative system, a preferred embodiment of which includes the above-mentioned storage medium; also includes a cloud-side large model, an edge end and a terminal device; and optimizes the data sent from the terminal device to the edge end by executing a computer program in the storage medium, and the optimization method adopts the closed-loop optimization formed by the above-mentioned dual weight system and entropy value verification mechanism.

[0093] The above describes the basic principles, main features, and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-described embodiments. The above-described embodiments and the specification merely illustrate the principles of the present application. Various changes and improvements may be made to the present application without departing from the spirit and scope of the present application. These changes and improvements fall within the scope of the present application for which protection is sought. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A data transmission method, characterized in that: The steps include: S100: Obtain weights for different information categories through distance calculation based on security management domain texts as baseline weights for security analysis; Generate user demand weights that match the task scenario through distance calculation based on user task requirements; Take the weighted average of the benchmark weight and the user demand weight to form the total weight of the information category; S200: Perform internal entropy analysis on the standard analysis case and construct the target category entropy value; Combine the total weight of the information categories obtained with the entropy of the target categories to generate the target entropy matrix required by the cloud-based large model; S300: Based on the data output by the edge end, the corresponding category weights and category entropy values ​​are calculated, and then the corresponding prediction entropy matrix is ​​obtained by combining the two. S400: Compare the difference between the obtained predicted entropy matrix and the target entropy matrix. If the difference between the two exceeds a set threshold, reconstruct the data output by the edge end and execute step S500; otherwise, pass the data output by the edge end to the cloud-based large model; S500: Repeat steps S300 and S400 on the reconstructed data; In step S100, the specific process of obtaining the reference weight and the user demand weight based on the distance calculation is as follows: Use the multi-head attention mechanism to encode text data and obtain sentence vectors and category vectors; Based on the obtained vector data, the distance between each sentence and different categories is calculated, and the cumulative sum of the distance values ​​is performed according to the category; Based on the obtained summation results, the softmax function is used to calculate the probability, obtain the score value of each category and construct the weight vector; Among them, when obtaining the weight vector corresponding to the baseline weight, the text data comes from the security management field text, and when obtaining the weight vector corresponding to the user demand weight, the text data comes from the user task demand; Both sentence vectors and category vectors are represented in the form of sentence vectors. The acquisition of sentence vectors and category vectors includes the following process: Convert text data into a vector form that the BERT model can understand as model input, including word embedding, paragraph embedding, and position embedding; After obtaining the vector form of text data, it is input into the BERT model and then encoded through the multi-head attention mechanism; Extract the embedding vector of the CLS token from the output of the BERT model and enhance the output of the BERT model by adding a pooling layer to obtain the final sentence vector; The calculation formula for the cumulative distance value from each statement to the category is as follows: ； The calculation formula of the weight vector is as follows: ； ； W0=[w1,w2,……,w k ]; In step S200, obtaining the target category entropy value includes the following process: The entire document of the standard analysis case is encoded into a high-dimensional semantic vector through the pre-trained BERT model; Extract multiple keywords from each category using the standard analysis case. Generate corresponding semantic vectors for all extracted keywords using the BERT model. The semantic vectors have the same dimensions as the high-dimensional semantic vectors. Based on the obtained semantic vector, calculate the two-dimensional distance matrix between each keyword and the corresponding category; Calculate the target category entropy value based on the number of keywords obtained and the two-dimensional distance matrix corresponding to each keyword; Target category entropy W b The calculation expression is as follows: ； ； W b =[w i ´],i={1,2,……,k}; Among them, cos_dist represents the semantic distance from each sentence to the category, a i represents the i-th category, b j Indicates the jth statement, the maximum number of statements is n, l i represents the sum of the semantic distances corresponding to the i-th category, s i represents the inverse of the sum of semantic distances corresponding to the i-th category, w i Represents the score value of each category, W0 represents the weight vector, k represents the maximum number of categories, w i ´ represents the entropy value corresponding to the i-th category, m represents the number of keywords, w i,x represents the entropy value of the xth keyword corresponding to the i-th category, μ represents the average entropy value corresponding to all keywords, and s i,x Represents the inverse of a two-dimensional distance matrix.

2. The data transmission method according to claim 1, wherein: In step S400, when the difference between the predicted entropy matrix and the target entropy matrix is ​​greater than a set threshold, reconstruction is completed by increasing the text vocabulary of the edge output data.

3. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the data transmission method according to claim 1 or 2 is implemented.

4. A cloud-edge collaborative system, characterized in that: Including the storage medium according to claim 3.

Images