Data credibility traceability method based on blockchain

Abstract

The present application relates to the field of intelligent tracing, and specifically discloses a data credibility tracing method based on blockchain, which uses artificial intelligence technology based on a deep neural network model to obtain product information downloaded from the blockchain and information in the product manual, and uses a semantic encoder to reduce the high-dimensional text data to a low-dimensional vector space, capture the semantic association and meaning between different parts, and then extract the local details and overall structure of the product manual through a multi-scale convolutional network to obtain a classification result for indicating whether the product manual is authentic and credible. In this way, the authenticity of the product manual can be judged to a certain extent, and the credibility of the product information can be improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent tracing, and more specifically, to a data credibility tracing method based on blockchain. Background Art

[0002] Blockchain is a decentralized system with no central authority or intermediary to control and manage data. Once data is written to the blockchain, it is almost impossible to tamper with or delete it. Transactions and data on the blockchain are publicly visible. Every participant can view and verify transaction records on the blockchain, ensuring the transparency and fairness of the data.

[0003] In order to protect the consumer rights of users, the credibility of product information has become increasingly important. The instruction manual is a document or manual that usually contains product features, uses, specifications, etc. to help users use and maintain products or services correctly. The main purpose of the instruction manual is to convey to users the functions, features, operating methods and precautions of products or services to ensure that users can use them safely and effectively.

[0004] However, in the prior art, sometimes the instructions are not authentic. Malicious parties may forge product instructions in order to deceive users or gain improper benefits. It may also be caused by the manufacturer's negligence, imperfect quality control or incorrect translation. It may also be caused by the failure to update the instructions in a timely manner, thus affecting the consumer's use and causing unnecessary impact and losses.

[0005] Therefore, we look forward to an optimized blockchain-based data credibility traceability solution. Summary of the invention

[0006] In order to solve the above technical problems, this application is proposed. The embodiment of the present application provides a data credibility traceability method based on blockchain, which adopts artificial intelligence technology based on deep neural network model to obtain product information downloaded from blockchain and information in product manual, and respectively reduces the high-dimensional text data to low-dimensional vector space through semantic encoders, captures the semantic association and meaning between different parts, and then extracts the local details and overall structure of the product manual through multi-scale convolutional networks to obtain classification results for indicating whether the product manual is authentic and reliable. In this way, the authenticity of the product manual can be judged to a certain extent, and the credibility of product information can be improved.

[0007] According to one aspect of the present application, a data credibility tracing method based on blockchain is provided, which includes:

[0008] Obtain product information downloaded from the blockchain and information in the product manual;

[0009] Passing the product information through a first semantic encoding model including a word embedding layer to obtain a product information feature vector;

[0010] Passing the information in the product manual through a second semantic encoding model including a word embedding layer to obtain a plurality of local feature vectors of the product manual information;

[0011] Arranging the plurality of local feature vectors of product manual information into a product manual information feature matrix, respectively passing through a convolutional neural network model with a first scale and a convolutional neural network model with a second scale to obtain a first-scale manual information feature vector and a second-scale manual information feature vector;

[0012] Cascading the first-scale instruction manual information feature vector and the second-scale instruction manual information feature vector to obtain a global feature vector of product instruction manual information;

[0013] Performing feature engineering matching based on a priori on the product information feature vector and the global feature vector of the product manual information to obtain a classification feature vector;

[0014] The classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the product manual is authentic.

[0015] In the above-mentioned blockchain-based data credibility traceability method, the product information is passed through a first semantic coding model including a word embedding layer to obtain a product information feature vector, including: performing word segmentation processing on the product information to convert the product information into a product information word sequence composed of multiple words; using the embedding layer of the first semantic coding model including the word embedding layer to respectively map each word in the product information word sequence to a word embedding vector to obtain a sequence of product information word embedding vectors; using the converter of the first semantic coding model including the word embedding layer to perform global context semantic encoding based on the converter idea on the sequence of product information word embedding vectors to obtain multiple product information context semantic feature vectors; and, cascading the multiple product information context semantic feature vectors to obtain the product information feature vector.

[0016] In the above-mentioned blockchain-based data credibility traceability method, the converter of the first semantic encoding model including the word embedding layer is used to perform global context semantic encoding based on the converter idea on the sequence of the product information word embedding vectors to obtain multiple product information context semantic feature vectors, including: arranging the sequence of the product information word embedding vectors in one dimension to obtain a global feature vector; calculating the product between the global feature vector and the transposed vector of each vector in the sequence of the product information word embedding vector to obtain multiple self-attention association matrices; respectively normalizing each of the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; passing each of the multiple standardized self-attention association matrices through the Softmax classification function to obtain multiple probability values; and, respectively using each of the multiple probability values ​​as a weight to weight each vector in the sequence of the word embedding vectors to obtain the multiple product information context semantic feature vectors.

[0017] In the above-mentioned blockchain-based data credibility traceability method, the information in the product manual is passed through a second semantic coding model including a word embedding layer to obtain multiple local feature vectors of the product manual information, including: performing word segmentation processing on the information in the product manual to convert the product information into a product manual word sequence composed of multiple words; using the embedding layer of the second semantic coding model including the word embedding layer to respectively map each word in the product manual word sequence into a word embedding vector to obtain a sequence of product manual word embedding vectors; and, using the converter of the second semantic coding model including the word embedding layer to perform global context semantic encoding based on the converter idea on the sequence of product manual word embedding vectors to obtain the multiple local feature vectors of the product manual information.

[0018] In the above-mentioned blockchain-based data credibility traceability method, after arranging the multiple local feature vectors of product manual information into a product manual information feature matrix, the product manual information feature vectors are respectively obtained by a convolutional neural network model with a first scale and a convolutional neural network model with a second scale, including: using each layer of the convolutional neural network model with a first scale to perform convolution processing based on a two-dimensional convolution kernel, mean pooling processing and activation processing of each feature matrix along the channel dimension on the input data in the forward pass of the layer, so that the output of the last layer of the convolutional neural network model with the first scale is The first-scale instruction manual information feature vector, wherein the input of the first layer of the convolutional neural network model using the first scale is the product instruction manual information feature matrix; and each layer of the convolutional neural network model using the second scale performs convolution processing based on a two-dimensional convolution kernel, mean pooling processing of each feature matrix along the channel dimension and activation processing on the input data in the forward pass of the layer so that the output of the last layer of the convolutional neural network model using the second scale is the second-scale instruction manual information feature vector, wherein the input of the first layer of the convolutional neural network model using the second scale is the product instruction manual information feature matrix.

[0019] In the above-mentioned blockchain-based data credibility traceability method, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the product manual is authentic and reliable, including: using the fully connected layer of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0020] Compared with the prior art, the data credibility traceability method based on blockchain provided by this application adopts artificial intelligence technology based on deep neural network model to obtain product information downloaded from blockchain and information in product manuals, and respectively reduces the high-dimensional text data into low-dimensional vector space through semantic encoders to capture the semantic association and meaning between different parts, and then extracts the local details and overall structure of the product manual through multi-scale convolutional networks to obtain classification results indicating whether the product manual is authentic and reliable. In this way, the authenticity of the product manual can be judged to a certain extent, and the credibility of product information can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 This is a flowchart of a blockchain-based data credibility traceability method according to an embodiment of the present application.

[0023] Figure 2 Schematic diagram of the architecture of the blockchain-based data credibility traceability method according to an embodiment of the present application.

[0024] Figure 3 This is a flowchart of obtaining a product information feature vector by passing the product information through a first semantic encoding model including a word embedding layer in a blockchain-based data credibility traceability method according to an embodiment of the present application.

[0025] Figure 4 is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0027] Exemplary Methods

[0028] Figure 1 This is a flowchart of a blockchain-based data credibility traceability method according to an embodiment of the present application. Figure 2 Schematic diagram of the architecture of the data credibility traceability method based on blockchain according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the data credibility traceability method based on blockchain includes: S110, obtaining product information downloaded from the blockchain and information in the product manual; S120, passing the product information through a first semantic coding model including a word embedding layer to obtain a product information feature vector; S130, passing the information in the product manual through a second semantic coding model including a word embedding layer to obtain multiple product manual information local feature vectors; S140, arranging the multiple product manual information local feature vectors into a product manual information feature matrix and then passing through a convolutional neural network model with a first scale and a convolutional neural network model with a second scale to obtain a first scale manual information feature vector and a second scale manual information feature vector; S150, cascading the first scale manual information feature vector and the second scale manual information feature vector to obtain a product manual information global feature vector; S160, performing feature engineering matching based on a priori on the product information feature vector and the product manual information global feature vector to obtain a classification feature vector; and, S170, passing the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the product manual is authentic.

[0029] Blockchain is a distributed ledger technology that records and verifies data transactions and information in a decentralized manner. It is a chain consisting of a series of data blocks linked together in chronological order, each of which contains a certain number of transaction records or information. Blockchain technology has now been widely used in other fields, such as supply chain management, digital identity verification, smart contracts, medical record management, etc. It is considered to have revolutionary potential to change traditional business and social models and provide a more secure, transparent and efficient way of data exchange and management. Specifically, the main characteristics of blockchain are as follows. 1. Decentralization: Blockchain has no centralized control agency, but is jointly maintained and managed by multiple nodes in the network. Each node has a complete copy of the ledger and reaches consensus through a consensus algorithm. 2. Tamper-proof: The data in the blockchain is encrypted and verified using cryptographic technology. Once the data is written to the blockchain, it is difficult to tamper with. This is because each data block contains the hash value of the previous data block, and any tampering with the data will cause the hash value to change, which can be detected by other nodes. 3. Transparency: The data in the blockchain is publicly visible, and anyone can view and verify transaction records. This transparency can increase trust and reduce dependence on intermediaries. 4. Security: Blockchain uses cryptography to protect the security of data. Through distributed networks and consensus algorithms, blockchain can resist attacks and fraud. 5. De-trust: Blockchain technology can achieve trustless transactions, that is, without the need for intermediaries, the reliability and security of transactions are ensured through algorithms and protocols.

[0030] Product manuals usually contain product information. A product manual is a document that describes the product features, usage, safety precautions, etc. in detail. It provides detailed information about the product to help users understand and use the product correctly. The authenticity of the product manual is very important. Reliable product manual data can provide information about the correct usage, safety precautions and precautions of the product. This helps users use the product correctly and avoid potential safety risks and accidents. It can also provide information about product specifications, performance and quality certification. Users can use this information to evaluate the quality of the product and match it with their needs and expectations. It can also help consumers understand the characteristics, functions and performance of the product so that they can make wise purchasing decisions. If the data in the product manual is unreliable or false, consumers may be misled and purchase products that do not meet their expectations, which may damage the brand reputation and affect consumers' trust and loyalty to the brand.

[0031] In step S110, the downloaded product information and the information in the product manual are obtained from the blockchain. Considering that the blockchain can be used to store and transmit product-related data, such as product descriptions of products, such as appearance, functions, and features; technical specifications, such as size and weight; production batches, supply chain information, quality inspection results such as ISO certification, CE certification, etc. This information can be used as one of the bases for judging the authenticity of the product manual. By obtaining the product information in the blockchain, we can understand the actual situation and relevant data of the product, and compare and verify it with the product manual. The product manual is a document provided by the manufacturer that describes the product characteristics, uses, specifications, etc. in detail. By analyzing the information in the product manual, we can understand the characteristics and functions of the product. And by comparing with the product information in the blockchain, the consistency and credibility of the product manual can be verified, and by comparing and analyzing them, the actual situation of the product and the content of the manual can be considered comprehensively, so as to judge the authenticity and credibility of the product manual. This helps consumers and relevant stakeholders make more informed decisions when purchasing and using products, and improves the transparency and trust of the product market.

[0032] Typically, manufacturers or related parties upload product information to the corresponding data storage area on the blockchain. This can be achieved by using the functions provided by the blockchain platform or through smart contracts. Then, users or verifiers can obtain product information by querying the blockchain. This can be achieved by using the blockchain's public interface or specific query functions. Users can query and obtain the required information based on the product's identifier, batch number, or other relevant information.

[0033] In step S120, the product information is passed through a first semantic encoding model including a word embedding layer to obtain a product information feature vector. Considering that the semantic encoding model can learn the semantic associations between words, words with similar meanings are mapped to similar vector spaces. In this way, even words that appear in different contexts can obtain similar vector representations, thereby better capturing the semantic information of the text. In addition, after converting the product information into a feature vector, the high-dimensional text data can be reduced to a low-dimensional vector space. Secondly, through the semantic encoding model, product information of different lengths can be converted into a feature vector representation of a fixed length. Doing so can make the model have better generalization capabilities for text inputs of different lengths and be able to process product information of different lengths. Therefore, converting product information into a feature vector through a semantic encoder can extract the semantic information of the text data and convert it into a numerical representation that can be processed by the machine.

[0034] Figure 3 The flowchart is a method for obtaining a product information feature vector by passing the product information through a first semantic encoding model including a word embedding layer in a blockchain-based data credibility traceability method according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3 As shown, the product information is passed through a first semantic encoding model including a word embedding layer to obtain a product information feature vector, including: S210, performing word segmentation processing on the product information to convert the product information into a product information word sequence composed of multiple words; S220, using the embedding layer of the first semantic encoding model including the word embedding layer to respectively map each word in the product information word sequence to a word embedding vector to obtain a sequence of product information word embedding vectors; S230, using the converter of the first semantic encoding model including the word embedding layer to perform global context semantic encoding based on the converter idea on the sequence of product information word embedding vectors to obtain multiple product information context semantic feature vectors; and, S240, cascading the multiple product information context semantic feature vectors to obtain the product information feature vector.

[0035] More specifically, in an embodiment of the present application, a converter of the first semantic encoding model including a word embedding layer is used to perform global context semantic encoding based on the converter idea on the sequence of product information word embedding vectors to obtain multiple product information context semantic feature vectors, including: arranging the sequence of product information word embedding vectors in one dimension to obtain a global feature vector; calculating the product between the global feature vector and the transposed vector of each vector in the sequence of product information word embedding vectors to obtain multiple self-attention association matrices; respectively normalizing each of the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; passing each of the multiple standardized self-attention association matrices through a Softmax classification function to obtain multiple probability values; and, respectively weighting each vector in the sequence of word embedding vectors using each of the multiple probability values ​​as a weight to obtain the multiple product information context semantic feature vectors.

[0036] In step S130, the information in the product manual is passed through a second semantic encoding model including a word embedding layer to obtain multiple local feature vectors of the product manual information. Considering that the product manual usually contains a large amount of text information, which may involve different topics, parts or paragraphs. By using a semantic encoder, each local text fragment can be semantically understood to capture the semantic associations and meanings between different parts. Specifically, different parts in the product manual may involve different features, functions or specifications. By converting each local text fragment into a feature vector, multiple local feature vectors can be obtained, each vector representing the semantic information of a local part. Doing so can provide more feature diversity, allowing the model to better capture the details and characteristics of the product manual.

[0037] In addition, different local text fragments may have different importance for judging the authenticity and credibility of product manuals. By representing each local text fragment as a feature vector, different local information can be processed and focused on independently. This can better discover and analyze the key information in the product manual and improve the model's ability to judge the product manual.

[0038] Specifically, in an embodiment of the present application, the information in the product manual is passed through a second semantic coding model including a word embedding layer to obtain a plurality of local feature vectors of the product manual information, including: performing word segmentation processing on the information in the product manual to convert the product information into a product manual word sequence composed of a plurality of words; using the embedding layer of the second semantic coding model including the word embedding layer to respectively map each word in the product manual word sequence into a word embedding vector to obtain a sequence of product manual word embedding vectors; and, using a converter of the second semantic coding model including the word embedding layer to perform global context semantic encoding based on the converter idea on the sequence of product manual word embedding vectors to obtain the plurality of local feature vectors of the product manual information.

[0039] In step S140, the plurality of local feature vectors of product manual information are arranged into a product manual information feature matrix, and then the first scale feature vector of the product manual information and the second scale feature vector of the product manual information are obtained respectively through a convolutional neural network model with a first scale and a convolutional neural network model with a second scale. Considering that the convolutional neural network model of the first scale can capture local features in a shorter range. By applying a smaller convolution kernel and pooling operation, the model can focus on more local features in the product manual, such as features at the phrase, word group or sentence level. This helps to capture details and local association information in the product manual. The convolutional neural network model of the second scale can capture global features in a longer range. By applying a larger convolution kernel and pooling operation, the model can focus on a larger range of features in the product manual, such as features at the paragraph, chapter or overall document level. This helps to capture the overall structure and contextual information of the product manual. Therefore, by using the convolutional neural network models of the first scale and the second scale respectively, feature extraction of the product manual can be performed at different levels. Doing so can comprehensively consider the local details and overall structure of the product manual, thereby obtaining a more comprehensive and rich feature representation.

[0040] Specifically, in an embodiment of the present application, after arranging the multiple local feature vectors of product manual information into a product manual information feature matrix, the first-scaled feature vector of the product manual information and the second-scaled feature vector of the product manual information are respectively obtained by passing through a convolutional neural network model with a first scale and a convolutional neural network model with a second scale, including: using each layer of the convolutional neural network model with a first scale to perform convolution processing based on a two-dimensional convolution kernel, mean pooling processing of each feature matrix along the channel dimension, and activation processing on the input data in the forward pass of the layer, so that the output of the last layer of the convolutional neural network model with the first scale is the first The invention relates to a method for preparing a product manual information feature vector of a scale, wherein the input of the first layer of the convolutional neural network model using the first scale is the product manual information feature matrix; and each layer of the convolutional neural network model using the second scale performs convolution processing based on a two-dimensional convolution kernel, mean pooling processing of each feature matrix along the channel dimension and activation processing on the input data in the forward pass of the layer so that the output of the last layer of the convolutional neural network model using the second scale is the second-scale manual information feature vector, wherein the input of the first layer of the convolutional neural network model using the second scale is the product manual information feature matrix.

[0041] In step S150, the first scale instruction manual information feature vector and the second scale instruction manual information feature vector are cascaded to obtain a global feature vector of product instruction manual information. Considering that by cascading the feature vectors of two scales, local and global information can be combined together. The first scale instruction manual information feature vector captures the local features of the product instruction manual in a shorter range, while the second scale instruction manual information feature vector captures the global features of the longer range. Cascading these two feature vectors allows the model to take into account both the details and the overall structure of the product instruction manual, thereby obtaining a more global feature representation. In addition, the cascaded feature vector can also provide more information richness. The feature vectors of the first scale and the second scale may capture different semantic information and features. By cascading them, these features of different scales can be organically combined to obtain a more comprehensive and rich global feature vector of product instruction manual information.

[0042] In step S160, the product information feature vector and the product manual information global feature vector are matched based on a priori feature engineering to obtain a classification feature vector.

[0043] In particular, in the technical solution of the present application, it is considered that the product information feature vector and the product manual information global feature vector are obtained from different data sources. Product information usually comes from product databases or information provided by merchants, while product manual information is usually extracted from product manuals through text parsing or natural language processing technology. Due to the difference in data sources, these two feature vectors may reflect different information and features, resulting in a decrease in the alignment between them. At the same time, the product information feature vector and the product manual information global feature vector may use different feature representation methods. For example, they may use different word embedding models, text processing techniques, or feature extraction methods. This inconsistency may lead to semantic differences between feature vectors, thereby reducing the alignment between them. The product information feature vector and the product manual information global feature vector may have different dimensions. If their dimensions do not match, they cannot be directly fused and compared. Dimension mismatch may lead to information loss or feature confusion, thereby affecting the classifier's accurate understanding of the features and the accuracy of the classification results. That is, because they come from different data sources, the feature representations are inconsistent and the feature dimensions do not match, so the alignment between the product information feature vector and the product manual information global feature vector is low.

[0044] In response to the technical problem of the alignment between the product information feature vector and the global feature vector of the product manual information, in the technical solution of the present application, a priori-based feature engineering matching is used to transform the alignment problem between feature vectors into a probabilistic problem, thereby adopting a probabilistic technology to improve the alignment and fusion effect between feature vectors.

[0045] Based on this, in the technical solution of the present application, the product information feature vector and the global feature vector of the product manual information are matched based on a priori feature engineering to obtain a classification feature vector using the following formula: wherein the formula is:

[0046]

[0047] Wherein, V1 represents the product information feature vector, V2 represents the global feature vector of the product manual information, α, β and λ represent predetermined hyperparameters, ||·|| F represents the Frobenius norm of a vector, exp(·) represents the exponential operation of a vector, tanh represents the hyperbolic tangent function, ⊕ represents the positional addition of a vector, represents the positional subtraction of vectors, ⊙ represents the positional dot multiplication of vectors, V f represents the classification feature vector.

[0048] Specifically, firstly, according to the structure and attributes of the product information feature vector and the global feature vector of the product manual information, a feature engineering matching strategy based on a priori is designed to map and transform the feature values ​​of different categories and dimensions according to certain a priori rules, thereby reducing the information loss and error accumulation in the alignment process. Then, a probabilistic technique is used to represent the posterior feature distribution of the implicit feature expression of the parameterized model based on the a priori feature engineering matching vector of the global feature vector of the product manual information relative to the product information feature vector, so as to achieve smoother vector consistency fusion, thereby improving the accuracy of the classification judgment of the classification feature vector.

[0049] In step S170, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the product manual is authentic and reliable. Considering that the classification feature vector is classified by a classifier in order to judge the authenticity and credibility of the product manual. The output result of the classifier represents the classification label of the product manual, which can be a binary classification (authentic / credible or forged / untrustworthy) or a multi-class classification (for example, authentic / credible, forged / untrustworthy, partially authentic / suspicious, etc.). This classification result can be used as a basis for judging whether the product manual is authentic and reliable. Therefore, by using a classifier to classify the product manual, the authenticity and credibility of the product manual can be automatically judged, reducing the subjectivity and uncertainty of manual judgment. The classifier can make decisions based on the information in the feature vector and provide an objective classification result to help users evaluate the credibility of the product manual.

[0050] A classifier is a machine learning model or algorithm that is used to classify input data into different categories or labels. The goal of a classifier is to learn to extract features from the input data and assign the data to predefined categories based on these features. Classifiers can be applied to a variety of fields and tasks, such as text classification, image classification, speech recognition, etc. They are a common method in supervised learning, which requires labeled training data to learn a model, which is then used to classify new unlabeled data.

[0051] The following are some common classifier algorithms. Logistic regression is a widely used classification algorithm for binary classification problems. It predicts the probability that a sample belongs to a certain category by multiplying the input features with weights and applying a logical function (such as the sigmoid function) to map the result to a probability between 0 and 1; the decision tree is a classification algorithm based on a tree structure. It represents the relationship between the condition and category of the feature through a series of decision nodes and leaf nodes. The decision tree gradually divides the data according to the value of the input feature and finally assigns the data to the category of the leaf node; the support vector machine is a binary classification algorithm whose goal is to find an optimal hyperplane to separate samples of different categories. It determines the optimal division boundary by maximizing the interval between the sample and the hyperplane.

[0052] Specifically, in an embodiment of the present application, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the product manual is authentic and reliable, including: using the fully connected layer of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0053] In summary, the data credibility traceability method based on blockchain in the embodiment of the present application is explained, which uses artificial intelligence technology based on a deep neural network model to obtain product information downloaded from the blockchain and information in the product manual, and uses a semantic encoder to reduce the high-dimensional text data to a low-dimensional vector space, capturing the semantic associations and meanings between different parts, and then extracting the local details and overall structure of the product manual through a multi-scale convolutional network to obtain a classification result for indicating whether the product manual is authentic and reliable. In this way, the authenticity of the product manual can be judged to a certain extent, and the credibility of the product information can be improved.

[0054] Exemplary Electronic Devices

[0055] Below, reference Figure 4 To describe an electronic device according to an embodiment of the present application. Figure 4 FIG. 1 is a block diagram of an electronic device according to an embodiment of the present application. Figure 4 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0056] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0057] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the functions of the data credibility traceability method based on blockchain in each embodiment of the present application described above and / or other desired functions. Various contents such as product information downloaded from the blockchain and information in the product manual may also be stored in the computer-readable storage medium.

[0058] In one example, the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0059] The input device 13 may include, for example, a keyboard, a mouse, etc.

[0060] The output device 14 can output various information to the outside, including classification results, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0061] Of course, to simplify, Figure 4 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

Claims

1. A data credibility tracing method based on blockchain, characterized in that: include: Obtain product information downloaded from the blockchain and information in the product manual; Passing the product information through a first semantic encoding model including a word embedding layer to obtain a product information feature vector; Passing the information in the product manual through a second semantic encoding model including a word embedding layer to obtain a plurality of local feature vectors of the product manual information; Arranging the plurality of local feature vectors of product manual information into a product manual information feature matrix, respectively passing through a convolutional neural network model with a first scale and a convolutional neural network model with a second scale to obtain a first-scale manual information feature vector and a second-scale manual information feature vector; Cascading the first-scale instruction manual information feature vector and the second-scale instruction manual information feature vector to obtain a global feature vector of product instruction manual information; Performing feature engineering matching based on a priori on the product information feature vector and the global feature vector of the product manual information to obtain a classification feature vector; Passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the product manual is authentic; The product information feature vector and the product manual information global feature vector are matched based on a priori feature engineering to obtain a classification feature vector, including: The product information feature vector and the product manual information global feature vector are matched based on a priori feature engineering using the following matching formula to obtain a classification feature vector; Wherein, the matching formula is: Wherein, V1 represents the product information feature vector, V2 represents the global feature vector of the product manual information, α, β and λ represent predetermined hyperparameters, ||·|| F represents the Frobenius norm of a vector, exp(·) represents the exponential operation of a vector, tanh represents the hyperbolic tangent function, represents the positional addition of vectors, represents the positional subtraction of vectors, ⊙ represents the positional dot multiplication of vectors, V f represents the classification feature vector.

2. The data credibility tracing method based on blockchain according to claim 1 is characterized in that: Passing the product information through a first semantic encoding model including a word embedding layer to obtain a product information feature vector includes: Performing word segmentation processing on the product information to convert the product information into a product information word sequence consisting of multiple words; Using the embedding layer of the first semantic encoding model including the word embedding layer, each word in the product information word sequence is mapped into a word embedding vector to obtain a sequence of product information word embedding vectors; Using the converter of the first semantic encoding model including the word embedding layer, performing global context semantic encoding based on the converter idea on the sequence of product information word embedding vectors to obtain multiple product information context semantic feature vectors; The multiple product information context semantic feature vectors are cascaded to obtain the product information feature vector.

3. The data credibility tracing method based on blockchain according to claim 2 is characterized in that: Using the converter of the first semantic encoding model including the word embedding layer, the sequence of product information word embedding vectors is subjected to global context semantic encoding based on the converter idea to obtain multiple product information context semantic feature vectors, including: Arrange the sequence of product information word embedding vectors in one dimension to obtain a global feature vector; Calculating the product between the global feature vector and the transposed vector of each vector in the sequence of product information word embedding vectors to obtain multiple self-attention association matrices; Normalizing each of the plurality of self-attention association matrices to obtain a plurality of standardized self-attention association matrices; Passing each of the plurality of standardized self-attention association matrices through a Softmax classification function to obtain a plurality of probability values; Each vector in the sequence of word embedding vectors is weighted using each probability value in the multiple probability values ​​as a weight to obtain the multiple product information context semantic feature vectors.

4. The data credibility tracing method based on blockchain according to claim 3 is characterized in that: The information in the product manual is passed through a second semantic encoding model including a word embedding layer to obtain a plurality of local feature vectors of the product manual information, including: Performing word segmentation processing on the information in the product manual to convert the product information into a product manual word sequence consisting of multiple words; Using the embedding layer of the second semantic encoding model including the word embedding layer, each word in the product manual word sequence is mapped into a word embedding vector to obtain a sequence of product manual word embedding vectors; The converter of the second semantic encoding model including the word embedding layer is used to perform global context semantic encoding based on the converter idea on the sequence of the product manual word embedding vectors to obtain the multiple product manual information local feature vectors.

5. The data credibility tracing method based on blockchain according to claim 4 is characterized in that: Arranging the plurality of local feature vectors of product manual information into a product manual information feature matrix and respectively passing through a convolutional neural network model with a first scale and a convolutional neural network model with a second scale to obtain a first-scale manual information feature vector and a second-scale manual information feature vector, including: Using each layer of the convolutional neural network model with a first scale, in the forward pass of the layer, respectively performs convolution processing based on a two-dimensional convolution kernel, mean pooling processing of each feature matrix along the channel dimension, and activation processing on the input data, so that the output of the last layer of the convolutional neural network model with a first scale is the first-scale instruction manual information feature vector, wherein the input of the first layer of the convolutional neural network model with a first scale is the product instruction manual information feature matrix; Each layer of the convolutional neural network model using the second scale performs convolution processing based on a two-dimensional convolution kernel, mean pooling processing of each feature matrix along the channel dimension and activation processing on the input data in the forward pass of the layer, so that the output of the last layer of the convolutional neural network model using the second scale is the second-scale instruction manual information feature vector, wherein the input of the first layer of the convolutional neural network model using the second scale is the product instruction manual information feature matrix.

6. The data credibility tracing method based on blockchain according to claim 5 is characterized in that: The classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate whether the product manual is authentic and reliable, including: Performing full connection encoding on the classification feature vector using the fully connected layer of the classifier to obtain an encoded classification feature vector; The encoded classification feature vector is passed through the Softmax classification function of the classifier to obtain the classification result.

Images