A cloud platform-based supply chain information sharing method

Abstract

The application discloses a supply chain information sharing method based on a cloud platform, relates to the technical field of supply chain information processing, and comprises the following steps: S1, generating a character feature channel according to the fusion characteristic value of each character in the key business field of the supply chain text; S2, extracting a semantic embedding vector of the key business field, and combining the character feature channel to generate a semantic enhancement channel; S3, generating a structure-time sequence feature channel according to the feature parameters of each sentence in the supply chain text; S4, generating a final key according to the character feature channel, the semantic enhancement channel and the structure-time sequence feature channel and corresponding weights; and S5, performing encryption processing on the supply chain text by using the final key and uploading the supply chain text to the shared memory of the cloud platform. The multi-channel feature enhances the uniqueness and attack resistance of the key, the shared memory of the cloud platform realizes efficient information sharing, and the safety and privacy of data are ensured.

Description

Technical Field

[0001] This invention relates to the field of supply chain information processing technology, and more specifically to a cloud-based supply chain information sharing method. Background Technology

[0002] With the deepening of global industrial chain restructuring and digital transformation, supply chains have evolved from single-enterprise internal collaboration to ecological collaboration across entities, regions, and industries. Information sharing has become a core support for improving supply chain agility, resilience, and cost control capabilities. However, supply chain text information contains a large amount of core sensitive data (such as pricing, customer privacy, and production processes), and its secure sharing faces multiple technical bottlenecks. Existing technical solutions are insufficient to meet the collaborative requirements of business adaptability, high security, and efficient sharing.

[0003] General text encryption technologies employ common encryption algorithms such as AES and RSA to encrypt all supply chain text, without designing differentiated encryption strategies tailored to the specific characteristics of supply chain business. General semantic embedding models use pre-trained models like BERT to extract semantic features from the text, without fine-tuning for business terminology and scenario logic specific to the supply chain domain, resulting in insufficient accuracy in semantic embedding. Static key generation mechanisms often use fixed keys or generate keys based on single features, making the keys vulnerable to cracking and unable to adapt to the dynamic changes in supply chain business scenarios. Summary of the Invention

[0004] To address the above problems, this invention proposes a supply chain information sharing method based on a cloud platform.

[0005] The technical solution of this invention is: a supply chain information sharing method based on a cloud platform, comprising the following steps:

[0006] S1. Generate character feature channels based on the fusion feature values ​​of each character in the key business fields of the supply chain text;

[0007] S2. Extract the semantic embedding vectors of key business fields and combine them with character feature channels to generate semantic enhancement channels;

[0008] S3. Generate structure-temporal feature channels based on the feature parameters of each sentence in the supply chain text;

[0009] S4. Generate the final key based on the character feature channel, semantic enhancement channel, and structure-temporal feature channel, as well as their corresponding weights;

[0010] S5. Use the final key to encrypt the supply chain text and upload it to the shared memory of the cloud platform.

[0011] Furthermore, S1 includes the following sub-steps:

[0012] S11. Extract several key business fields from the supply chain text and generate character sets for each key business field;

[0013] S12. The weighted character frequencies and positional entropies of each character in the character set are fused to generate fused feature values;

[0014] S13. Arrange the fusion feature values ​​of all characters in the character set order to generate character feature channels.

[0015] The beneficial effects of the above-mentioned further solutions are as follows: In this invention, key business fields include, but are not limited to, product ID, amount, delivery date, and fulfillment threshold, which are core sensitive fields in the supply chain. These fields are automatically identified by the cloud platform based on the business scenario or selected by the user. The character feature channel does not target isolated character frequency and positional entropy, but rather targets key business fields in the supply chain text (such as product ID, amount, and delivery date in orders, and fulfillment threshold and liability for breach of contract in contracts). It performs business-weighted fusion of character frequency and positional entropy to generate statistical feature vectors that are strongly bound to business attributes, avoiding the lack of distinctiveness in general statistical features.

[0016] This invention extracts the character sets of key business fields from supply chain text (such as the character set of product ID in order text being numbers + letters, and the character set of transaction amount being numbers + decimal point). This invention only performs statistics on the character set, rather than the entire text characters.

[0017] Furthermore, in S12, the character Fusion eigenvalues The expression is:

[0018] ;

[0019] ;

[0020] ;

[0021] in, Represents frequency weights. Indicates positional weight. Character Weighted character frequency, Character position entropy, Character In key business fields The number of times it appears, Represents key business fields Total character length, Character Key business fields In terms of the weight of supply chain text, Character In key business fields The The probability of occurrence at each position, Character The business weight, Represents the logarithmic function. Character In key business fields The total number of positions appearing in the text.

[0022] This indicates the weight of the business field to which the character belongs. For example, the amount field has a higher weight in the supply chain text, which is 0.8, the product ID has a weight of 0.6, and the remarks have a weight of 0.2. The business weight represents the position of characters. For example, in a monetary field, the first character has a weight of 1, and the last character has a weight of 0.1, which can effectively prevent the amount from being tampered with. The frequency weight and position weight are dynamically adjusted according to the supply chain scenario. For example, the weights of financial text and logistics text should be different.

[0023] Furthermore, S2 includes the following sub-steps:

[0024] S21. Input the key business fields into the encoder to obtain the hidden layer representation of each token;

[0025] S22. Utilize the field pooling layer to perform mean pooling on the hidden layer representation of each token to obtain the semantic embedding vector of the key business fields.

[0026] S23. Use principal component analysis to reduce the dimensionality of the semantic embedding vector;

[0027] S24. Perform a Hadamard product operation on the dimensionality-reduced semantic embedding vector and the character feature channel to obtain the semantic enhancement channel of the key business field.

[0028] The beneficial effects of the above-mentioned further solution are as follows: In this invention, the supply chain text is input into the model composed of the encoder and the hidden layer to obtain the semantic embedding vector of each business field. The business statistical feature vector of the first channel is fused with the semantic embedding vector at the element-wise weighted level to generate an enhanced semantic feature vector, thus linking the semantic features with the business statistical features. The semantic embedding vector adopts the standard output dimension of the general BERT, which is usually a fixed dimension, i.e., 768 dimensions. Therefore, this invention uses principal component analysis (PCA) to reduce the dimensionality of the semantic embedding vector, which can retain 95% of the semantic information, ensuring that the semantic features of the business field can still be accurately reflected after dimensionality reduction, and retaining the core semantic information.

[0029] Furthermore, S3 includes the following sub-steps:

[0030] S31. Perform time-series annotation on each sentence of the supply chain text, extract the action type and corresponding timestamp, and generate a time-series structure sequence;

[0031] S32. Calculate the entropy value of the temporal structure sequence as the structure entropy;

[0032] S33. Calculate the timestamp difference between adjacent sentences as the temporal entropy;

[0033] S34. The structural entropy, temporal difference entropy and 2-gram features after dimensionality reduction of the supply chain text are concatenated to obtain the structural-temporal feature channel.

[0034] The beneficial effects of the above-mentioned further solution are as follows: In this invention, in S31, each sentence in the supply chain text is labeled with two core tags: structure type: what business action this sentence belongs to, i.e., inventory receipt, order creation, and logistics delivery; timestamp: the specific time when this business action occurred, generating an ordered sequence containing business actions and timestamps. For example, if the sentence contains the structure of order creation, it is necessary to determine the timestamp of order creation.

[0035] Structural entropy is a scalar value that calculates the business structure complexity of the entire text by statistically analyzing the probability of occurrence of all structural types in the entire sequence.

[0036] Temporal difference entropy calculates the probability distribution of the time difference between all adjacent sentences in the entire sequence, reflecting the degree of business temporal fluctuation in the entire text, and is also a scalar value.

[0037] 2-gram features are used to count the frequency of 2-grams (two consecutive characters or words) in the entire supply chain text, and are global features of the entire text.

[0038] Furthermore, S4 includes the following sub-steps:

[0039] S41. Calculate the mutual information between the character feature channel and the semantic enhancement channel, calculate the mutual information between the character feature channel and the structure-temporal feature channel, and calculate the mutual information between the semantic enhancement channel and the structure-temporal feature channel;

[0040] S42. Based on the three mutual information, generate channel weights for the character feature channel, semantic enhancement channel, and structure-temporal feature channel, respectively.

[0041] S43. Normalize the character feature channel, semantic enhancement channel, and structure-temporal feature channel, and multiply the normalization result by the corresponding channel weight to obtain the channel normalized feature;

[0042] S44. Concatenate the normalized features of the three channels according to the channel dimension to generate a three-dimensional feature tensor.

[0043] S45. Perform singular value decomposition on the three-dimensional feature tensor to generate three subkey tensors;

[0044] S46. Expand each subkey tensor into a one-dimensional vector, perform zero-padding alignment or truncation alignment, and then perform an XOR operation to generate the final key.

[0045] The beneficial effects of the above-mentioned further scheme are as follows: In this invention, channel weights are generated by calculating the mutual information between channels, multi-channel features are normalized and weighted, concatenated into a three-dimensional feature tensor, and then generated into subkey tensors through singular value decomposition. Finally, the total key is generated by XOR fusion. The weights are positively correlated with the mutual information. For any non-empty three-dimensional tensor (i.e., the dimensions of the three normalized feature vectors are all greater than 0, which must be satisfied in the supply chain text encryption scenario because the feature vectors are generated from valid text), high-order singular value decomposition always exists, which can yield three subkey tensors, i.e., factor matrices. The key generated by tensor decomposition has high complexity and uniqueness, improving the security and anti-attack capability of encryption.

[0046] Furthermore, in S46, the final key The expression is:

[0047] ;

[0048] in, This represents the first subkey tensor. This represents the second subkey tensor. This represents the third subkey tensor. This indicates the XOR operation.

[0049] The beneficial effects of the above-mentioned further scheme are as follows: In this invention, the subkey tensor is flattened into a one-dimensional vector, and the final key is generated by fusing them through an XOR operation, ensuring the high complexity and uniqueness of the key. The irreversibility of the XOR operation enhances the security of the key, and the flattening operation compresses the multi-dimensional information of the tensor into one dimension, improving the key generation efficiency and encryption speed.

[0050] The beneficial effects of this invention are as follows: This invention extracts features from three dimensions—character, semantic, and structure-time sequence—combined with tensor decomposition to generate a highly complex key, which is then used to encrypt supply chain text and upload it to a cloud platform's shared memory, achieving secure and controllable information sharing. The multi-channel features of this invention enhance the uniqueness and anti-attack capabilities of the key, while the cloud platform's shared memory enables efficient information sharing, simultaneously ensuring data security and privacy. Attached Figure Description

[0051] Figure 1 This is a flowchart of a cloud-based supply chain information sharing method. Detailed Implementation

[0052] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0053] like Figure 1 As shown, the present invention provides a supply chain information sharing method based on a cloud platform, comprising the following steps:

[0054] S1. Generate character feature channels based on the fusion feature values ​​of each character in the key business fields of the supply chain text;

[0055] S2. Extract the semantic embedding vectors of key business fields and combine them with character feature channels to generate semantic enhancement channels;

[0056] S3. Generate structure-temporal feature channels based on the feature parameters of each sentence in the supply chain text;

[0057] S4. Generate the final key based on the character feature channel, semantic enhancement channel, and structure-temporal feature channel, as well as their corresponding weights;

[0058] S5. Use the final key to encrypt the supply chain text and upload it to the shared memory of the cloud platform.

[0059] In this embodiment of the invention, S1 includes the following sub-steps:

[0060] S11. Extract several key business fields from the supply chain text and generate character sets for each key business field;

[0061] S12. The weighted character frequencies and positional entropies of each character in the character set are fused to generate fused feature values;

[0062] S13. Arrange the fusion feature values ​​of all characters in the character set order to generate character feature channels.

[0063] In this invention, key business fields include, but are not limited to, product ID, amount, delivery date, and fulfillment threshold—core sensitive fields in the supply chain. These fields are automatically identified by the cloud platform based on the business scenario or selected by the user. The character feature channel does not target isolated character frequency and positional entropy, but rather focuses on key business fields in the supply chain text (such as product ID, amount, and delivery date in orders, and fulfillment thresholds and liability for breach of contract in contracts). It weights and fuses character frequency and positional entropy to generate statistical feature vectors that are strongly tied to business attributes, avoiding the lack of distinctiveness in general statistical features.

[0064] This invention extracts the character sets of key business fields from supply chain text (such as the character set of product ID in order text being numbers + letters, and the character set of transaction amount being numbers + decimal point). This invention only performs statistics on the character set, rather than the entire text characters.

[0065] In this embodiment of the invention, in S12, the character Fusion eigenvalues The expression is:

[0066] ;

[0067] ;

[0068] ;

[0069] in, Represents frequency weights. Indicates positional weight. Character Weighted character frequency, Character position entropy, Character In key business fields The number of times it appears, Represents key business fields Total character length, Character Key business fields In terms of the weight of supply chain text, Character In key business fields The The probability of occurrence at each position, Character The business weight, Represents the logarithmic function. Character In key business fields The total number of positions appearing in the text.

[0070] This indicates the weight of the business field to which the character belongs. For example, the amount field has a higher weight in the supply chain text, which is 0.8, the product ID has a weight of 0.6, and the remarks have a weight of 0.2. The business weight represents the position of characters. For example, in a monetary field, the first character has a weight of 1, and the last character has a weight of 0.1, which can effectively prevent the amount from being tampered with. The frequency weight and position weight are dynamically adjusted according to the supply chain scenario. For example, the weights of financial text and logistics text should be different.

[0071] In this embodiment of the invention, S2 includes the following sub-steps:

[0072] S21. Input the key business fields into the encoder to obtain the hidden layer representation of each token;

[0073] S22. Utilize the field pooling layer to perform mean pooling on the hidden layer representation of each token to obtain the semantic embedding vector of the key business fields.

[0074] S23. Use principal component analysis to reduce the dimensionality of the semantic embedding vector;

[0075] S24. Perform a Hadamard product operation on the dimensionality-reduced semantic embedding vector and the character feature channel to obtain the semantic enhancement channel of the key business field.

[0076] In this invention, the supply chain text is input into a model consisting of an encoder and a hidden layer to obtain the semantic embedding vector for each business field. The business statistical feature vector from the first channel is then element-wise weighted and fused with the semantic embedding vector to generate an enhanced semantic feature vector, linking the semantic features with the business statistical features. The semantic embedding vector uses the standard output dimension of the general BERT, typically a fixed dimension of 768. Therefore, this invention uses Principal Component Analysis (PCA) to reduce the dimensionality of the semantic embedding vector, retaining 95% of the semantic information and ensuring that the dimensionality reduction accurately reflects the semantic features of the business fields, preserving the core semantic information.

[0077] In this embodiment of the invention, S3 includes the following sub-steps:

[0078] S31. Perform time-series annotation on each sentence of the supply chain text, extract the action type and corresponding timestamp, and generate a time-series structure sequence;

[0079] S32. Calculate the entropy value of the temporal structure sequence as the structure entropy;

[0080] S33. Calculate the timestamp difference between adjacent sentences as the temporal entropy;

[0081] S34. The structural entropy, temporal difference entropy and 2-gram features after dimensionality reduction of the supply chain text are concatenated to obtain the structural-temporal feature channel.

[0082] In this invention, in step S31, each sentence in the supply chain text is tagged with two core labels: structure type: what business action the sentence belongs to, i.e., inventory receipt, order creation, and logistics delivery; and timestamp: the specific time when this business action occurred, generating an ordered sequence containing the business action and time. For example, if the sentence contains the structure of order creation, the timestamp of order creation needs to be determined.

[0083] Structural entropy is a scalar value that calculates the business structure complexity of the entire text by statistically analyzing the probability of occurrence of all structural types in the entire sequence.

[0084] Temporal difference entropy calculates the probability distribution of the time difference between all adjacent sentences in the entire sequence, reflecting the degree of business temporal fluctuation in the entire text, and is also a scalar value.

[0085] 2-gram features are used to count the frequency of 2-grams (two consecutive characters or words) in the entire supply chain text, and are global features of the entire text.

[0086] In this embodiment of the invention, S4 includes the following sub-steps:

[0087] S41. Calculate the mutual information between the character feature channel and the semantic enhancement channel, calculate the mutual information between the character feature channel and the structure-temporal feature channel, and calculate the mutual information between the semantic enhancement channel and the structure-temporal feature channel;

[0088] S42. Based on the three mutual information, generate channel weights for the character feature channel, semantic enhancement channel, and structure-temporal feature channel, respectively.

[0089] S43. Normalize the character feature channel, semantic enhancement channel, and structure-temporal feature channel, and multiply the normalization result by the corresponding channel weight to obtain the channel normalized feature;

[0090] S44. Concatenate the normalized features of the three channels according to the channel dimension to generate a three-dimensional feature tensor.

[0091] S45. Perform singular value decomposition on the three-dimensional feature tensor to generate three subkey tensors;

[0092] S46. Expand each subkey tensor into a one-dimensional vector, perform zero-padding alignment or truncation alignment, and then perform an XOR operation to generate the final key.

[0093] In this invention, channel weights are generated by calculating mutual information between channels. Multi-channel features are then normalized and weighted, concatenated into a three-dimensional feature tensor, and generated as subkey tensors through singular value decomposition. Finally, these subkey tensors are XORed and fused to generate the total key. The weights are positively correlated with the mutual information. For any non-empty three-dimensional tensor (i.e., all three normalized feature vectors have dimensions greater than 0, which is necessarily satisfied in supply chain text encryption scenarios because the feature vectors are generated from valid text), higher-order singular value decomposition always exists, yielding three subkey tensors, i.e., a factor matrix. The key generated by tensor decomposition has high complexity and uniqueness, improving encryption security and resistance to attacks.

[0094] The ratio of the correlation strength of each channel to the total correlation strength is used as the weight.

[0095] In this embodiment of the invention, in S46, the final key The expression is:

[0096] ;

[0097] in, This represents the first subkey tensor. This represents the second subkey tensor. This represents the third subkey tensor. This indicates the XOR operation.

[0098] In this invention, the subkey tensor is flattened into a one-dimensional vector, and the final key is generated by fusing them through an XOR operation, ensuring the high complexity and uniqueness of the key. The irreversibility of the XOR operation enhances the security of the key, and the flattening operation compresses the multidimensional information of the tensor into one dimension, improving the efficiency of key generation and encryption speed.

[0099] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A supply chain information sharing method based on a cloud platform, characterized in that, Includes the following steps: S1. Generate character feature channels based on the fusion feature values ​​of each character in the key business fields of the supply chain text; S2. Extract the semantic embedding vectors of key business fields and combine them with character feature channels to generate semantic enhancement channels; S3. Generate structure-temporal feature channels based on the feature parameters of each sentence in the supply chain text; S4. Generate the final key based on the character feature channel, semantic enhancement channel, and structure-temporal feature channel, as well as their corresponding weights; S5. Use the final key to encrypt the supply chain text and upload it to the shared memory of the cloud platform; S1 includes the following sub-steps: S11. Extract several key business fields from the supply chain text and generate character sets for each key business field; S12. The weighted character frequencies and positional entropies of each character in the character set are fused to generate fused feature values; S13. Arrange the fusion feature values ​​of all characters in the character set order to generate character feature channels; In S12, the character Fusion eigenvalues The expression is: ； ； ； in, Represents frequency weights. Indicates positional weight. Character Weighted character frequency, Character position entropy, Character In key business fields The number of times it appears, Represents key business fields Total character length, Character Key business fields In terms of the weight of supply chain text, Character In key business fields The The probability of occurrence at each position, Character The business weight, Represents the logarithmic function. Character In key business fields The total number of positions appearing in; S2 includes the following sub-steps: S21. Input the key business fields into the encoder to obtain the hidden layer representation of each token; S22. Utilize the field pooling layer to perform mean pooling on the hidden layer representation of each token to obtain the semantic embedding vector of the key business fields. S23. Use principal component analysis to reduce the dimensionality of the semantic embedding vector; S24. Perform a Hadamard product operation on the dimension-reduced semantic embedding vector and the character feature channel to obtain the semantic enhancement channel of the key business field. S3 includes the following sub-steps: S31. Perform time-series annotation on each sentence of the supply chain text, extract the action type and corresponding timestamp, and generate a time-series structure sequence; S32. Calculate the entropy value of the temporal structure sequence as the structure entropy; S33. Calculate the timestamp difference between adjacent sentences as the temporal entropy; S34. The structural entropy, temporal difference entropy and 2-gram features after dimensionality reduction of the supply chain text are concatenated to obtain the structural-temporal feature channel.

2. The supply chain information sharing method based on a cloud platform according to claim 1, characterized in that, S4 includes the following sub-steps: S41. Calculate the mutual information between the character feature channel and the semantic enhancement channel, calculate the mutual information between the character feature channel and the structure-temporal feature channel, and calculate the mutual information between the semantic enhancement channel and the structure-temporal feature channel; S42. Based on the three mutual information, generate channel weights for the character feature channel, semantic enhancement channel, and structure-temporal feature channel, respectively. S43. Normalize the character feature channel, semantic enhancement channel, and structure-temporal feature channel, and multiply the normalization result by the corresponding channel weight to obtain the channel normalized feature; S44. Concatenate the normalized features of the three channels according to the channel dimension to generate a three-dimensional feature tensor. S45. Perform singular value decomposition on the three-dimensional feature tensor to generate three subkey tensors; S46. Expand each subkey tensor into a one-dimensional vector, perform zero-padding alignment or truncation alignment, and then perform an XOR operation to generate the final key.

3. The supply chain information sharing method based on a cloud platform according to claim 2, characterized in that, In S46, the final key The expression is: ； in, This represents the first subkey tensor. This represents the second subkey tensor. This represents the third subkey tensor. This indicates the XOR operation.

Images