Text semantic communication method based on joint knowledge graph learning

Abstract

The invention relates to a text semantic communication method based on joint knowledge graph learning. The method comprises the steps that alignment of entities, relationships and entity attributes in a knowledge graph library between user sides is completed through a secure hash algorithm; unifying embedded representations of knowledge graph libraries of different user sides by utilizing a generative adversarial network; the sending user side compresses a to-be-transmitted text into semantic triples, and a similar semantic triple set is matched from a local knowledge graph library; encoding the similar semantic triad and then sending the encoded similar semantic triad to a receiving end; the receiving end decodes the received message to obtain a similar semantic triple set; utilizing a character string similarity algorithm to calculate the similarity between the similar semantic triples and a local knowledge graph library, and selecting the most similar triples as restored semantic triples; and converting the recovered semantic triad into a recovered text and recovering the to-be-transmitted text by adopting the fine-tuned semantic recovery module. According to the method and the device, the security and privacy of semantic transmission can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and in particular to a text semantic communication method based on joint knowledge graph learning. Background Art

[0002] Benefiting from the progress of microelectronics and artificial intelligence technology, deep learning (DL) and natural language processing (NLP) technologies have recently emerged and played an important role in the transformation of traditional communication technology to the future. Semantic communication has been proposed as an intelligent communication scheme. Unlike traditional communication, it emphasizes the successful delivery of message semantics rather than the precise bit stream transmission in traditional communication. Semantic communication can further compress data while retaining its meaning by extracting the meaning of data and filtering out useless, irrelevant and unimportant information, processing data in the semantic domain. In addition, semantic communication is expected to be robust to harsh channel environments, i.e., low signal to noise ratio (SNR) regions, which is very suitable for applications requiring high reliability. These factors motivate us to develop intelligent semantic communication systems by considering the semantics behind digital bits to improve the accuracy and efficiency of communication.

[0003] Semantic communication based on knowledge graph is a recently emerging communication paradigm. Knowledge graph is a data structure that uses a graph structure composed of entities and relationships to store human knowledge. Using knowledge graph as a tool to improve communication efficiency and accuracy improves the interpretability of semantic communication under the deep learning architecture. The key role of knowledge graph in communication system is to provide semantic representation of information, so that semantic encoding is not only efficient and accurate but also robust to communication errors. However, the existing semantic communication model cannot fully utilize data from different users and different knowledge domains while protecting the privacy of exchanged data, using the data set collected and organized by users as the background knowledge base. Each user maintains a private knowledge graph library. As the public's requirements for privacy protection increase, most users are reluctant to share their knowledge graph libraries. Under the premise of protecting user privacy, a more loosely coupled and principled way should be considered to share their knowledge graph libraries and benefit multiple parties. Summary of the invention

[0004] In order to solve the problems existing in the background technology and ensure the privacy of users, the present invention provides a text semantic communication method based on joint knowledge graph learning, including:

[0005] S1: Each user terminal uses knowledge graph embedding technology to transform its own knowledge graph library into a low-dimensional vector representation, and calculates the secure hash values ​​of entities, relationships, and entity attributes in the knowledge graph library through a secure hash algorithm;

[0006] S2: The user ends complete the alignment of entities, relationships and entity attributes in the knowledge graph library through the secure hash values ​​of entities, relationships and entity attributes;

[0007] S3: Construct a generative adversarial network system, set the generator and discriminator on two different user ends respectively, and unify the embedding representation of the knowledge graph libraries on different user ends by training the generative adversarial network;

[0008] S4: The sending user terminal inputs the text to be transmitted into the semantic extraction model for semantic information recognition; and constructs semantic triples according to the recognized semantic information; performs similarity matching on the semantic triples according to the embedded representation of the local knowledge graph library to extract a set of similar semantic triples; performs source coding and channel coding on the extracted similar semantic triples, and sends the coded information to the receiving end;

[0009] S5: The receiving user receives the message sent by the sending end, performs channel decoding and source decoding on the received message to obtain a set of similar semantic triples; uses a string similarity algorithm to calculate the similarity between similar semantic triples and the local knowledge graph library, and selects the most similar triple as the restored semantic triple; uses a fine-tuned semantic recovery module to convert the restored semantic triple into a restored text.

[0010] The present invention has at least the following beneficial effects

[0011] In view of the fact that traditional semantic communication systems cannot fully utilize data from different users and different fields, the present invention provides a text semantic communication method based on joint knowledge graph learning. By using a knowledge extractor as a semantic encoder, the source signal is input into a semantic extraction model for semantic information recognition, knowledge triple information is extracted from the input text, and it is compressed into a triple semantic representation.

[0012] In the joint knowledge graph learning stage, the semantic encoding method introduces a generative adversarial network in order to improve the quality of a single knowledge graph of multiple cross-domain knowledge graphs in a multi-user system, ensure the privacy of each user-side knowledge graph, and make full use of the local data of the user side. The generator and the discriminator are located on different sides of a pair of knowledge graphs. The generator is trained to convert the knowledge data on the generator side into a similar form to the knowledge data on the discriminator side, so that the updated knowledge data has an embedded vector representation similar to the local one and the original data is not exposed.

[0013] By introducing the adversarial mechanism of the adversarial generative network, the generated knowledge is more in line with the requirements of logic and semantics, and can make full use of the user's local data without infringing the privacy of the user's data, thereby improving the quality of the knowledge graph. Through privacy protection technology, all parties can safely share and integrate the data in the knowledge graph to produce a more complete and rich knowledge graph without worrying about the leakage of sensitive data. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of a communication system of the present invention;

[0015] Figure 2 A schematic diagram of the architecture for unifying the embedding representation of different user-side knowledge graph libraries using a generative adversarial network in the present invention;

[0016] Figure 3 A comparison diagram of information transmission bits between the present invention and different communication systems;

[0017] Figure 4 Model performance diagram of BLEU scores for different transmission models at different SNRs. DETAILED DESCRIPTION

[0018] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0019] Example 1. The present invention provides a text semantic communication method based on joint knowledge graph learning, including: training a generative adversarial network to jointly update knowledge graph learning to obtain a richer local knowledge graph library; building and training a semantic extraction model and a semantic recovery module, obtaining the trained text semantic communication input of the text data to be transmitted, and obtaining the recovered text.

[0020] Example 2, please refer to Figure 1 to Figure 4 The present invention provides a text semantic communication method based on joint knowledge graph learning, comprising:

[0021] S1: Each user terminal uses knowledge graph embedding technology to transform its own knowledge graph library into a low-dimensional vector representation, and calculates the secure hash values ​​of entities, relationships, and entity attributes in the knowledge graph library through a secure hash algorithm;

[0022] S2: The user ends complete the alignment of entities, relationships and entity attributes in the knowledge graph library through the secure hash values ​​of entities, relationships and entity attributes;

[0023] S3: Construct a generative adversarial network system, set the generator and discriminator on two different user ends respectively, and unify the embedding representation of the knowledge graph libraries on different user ends by training the generative adversarial network;

[0024] S4: The sending user terminal inputs the text to be transmitted into the semantic extraction model for semantic information recognition; and constructs semantic triples according to the recognized semantic information; performs similarity matching on the semantic triples according to the embedded representation of the local knowledge graph library to extract a set of similar semantic triples; performs source coding and channel coding on the extracted similar semantic triples, and sends the coded information to the receiving end;

[0025] S5: The receiving user receives the message sent by the sending end, performs channel decoding and source decoding on the received message to obtain a set of similar semantic triples; uses a string similarity algorithm to calculate the similarity between similar semantic triples and the local knowledge graph library, and selects the most similar triple as the restored semantic triple; uses a fine-tuned semantic recovery module to convert the restored semantic triple into a restored text.

[0026] In one embodiment, Figure 2 As shown in FIG, it is used to realize end-to-end semantic communication between the sender and the receiver. It includes: a sender, a channel and a receiver. The sender mainly includes functions such as text semantic extraction, semantic encoding and channel encoding; the channel is responsible for information transmission; the receiver mainly includes functions such as channel decoding and text recovery based on a text generation model.

[0027] Specifically, in order to represent the entities and relations in the knowledge graph, the TransE model is adopted. The TransE model is a knowledge graph representation method based on vector embedding, which realizes the structured representation of knowledge by mapping the entities and relations in the knowledge graph to a low-dimensional vector space. In the TransE model, each knowledge graph is represented as a set of facts in the form of triples, namely (h, r, t), where h represents the head entity, r represents the relationship, and t represents the tail entity. The TransE model assumes that for each triple, if h+r≈t, the triple is true. Therefore, TransE learns the low-dimensional vector representation of each entity and relationship by optimizing the loss function, so that the head entity vector h plus the relationship vector r can be as close as possible to the tail entity vector t. The objective function is defined using a distance function. For a given triple (h, r, t), the loss function of the model is:

[0028] L(h,r,t)=||h+rt||

[0029] By minimizing this loss function, TransE is able to learn embedded representations.

[0030] Specifically, Figure 1 As shown, the sending user terminal has its own private knowledge graph library, and in order to share or align entities and relationships, the parties need to jointly update knowledge in a secure and privacy-preserving manner. In the knowledge graph alignment process, a secure hashing method is used, and a secure hash function is used to generate a unique hash value for each entity in the knowledge graph. The SHA-256 algorithm was used in the experiment. After the hash value generation is completed, the two parties compare the hash value lists and find the shared hash values, which can determine that the entities corresponding to these hash values ​​are the same in their respective knowledge graphs, and all matching hash values ​​will be used as aligned entities.

[0031] In order to share and align entities and relationships, all parties need to jointly update knowledge in a secure and privacy-preserving manner. To this end, this paper adopts a GAN-based knowledge graph alignment mechanism, which enables the knowledge graphs of both parties to update and expand knowledge without leaking private data.

[0032] Preferably, unifying the embedding representations of different user-side knowledge graph libraries by training a generative adversarial network includes:

[0033] Use knowledge graph embedding technology to convert the knowledge graph in the local knowledge graph library into a low-dimensional vector representation;

[0034] The generator receives the low-dimensional vector representation from the local knowledge graph as input and outputs the fused embedding representation;

[0035] The discriminator receives the embedding representation generated by the generator or the low-dimensional vector representation from the local knowledge graph as input and outputs a probability value representing the authenticity of the data; the loss function L of the generator is G for:

[0036]

[0037] The loss function L of the discriminator D for:

[0038]

[0039] Among them, G() represents the generator, D() represents the discriminator; n represents the number of training samples; c m represents the sample of the input generator; y k represents the sample of the local knowledge graph library; log represents the natural logarithm;

[0040] After the training of the generative adversarial network is completed, the embedding representation in the local knowledge base of the user side is updated through the embedding representation generated by the generator to unify the embedding representation of different user-side knowledge graph libraries.

[0041] Specifically, given two entities E with alignment i ∩E j and the relation R i ∩R j The knowledge graph library KG(g i , g j ), a generative adversarial network (GAN) is used to unify the embedding representations of aligned entities and relations. The generator and discriminator are located on different sides of a pair of knowledge graphs, that is, on two clients. More specifically, the generator is placed on the g i In the example above, the discriminator is placed in gj. The generator embeds the aligned entities from g i Convert to g j , and the discriminator distinguishes between the synthesized embeddings of the generator and g j After GAN training, the synthesized embedding can learn features from both KGs and can therefore replace g i and g j Medium E i ∩E j and R i ∩R j The original embedding of is used as the refined and unified embedding. One party of the generator is responsible for converting its private knowledge graph data into a form that the other party cannot directly discern. During the generation process, the generator generates corresponding knowledge fragments based on the relationship between the aligned entities and their neighborhoods. The loss function of the generator is defined as:

[0042]

[0043] The discriminator is responsible for determining whether the generated data is real data, that is, whether the generated data matches the knowledge in its own graph. The discriminator gradually improves its ability to distinguish aligned knowledge through adversarial training with the generator. The loss function of the discriminator is defined as:

[0044]

[0045] Among them, G() represents the generator, D() represents the discriminator; n represents the number of training samples; x m represents the sample of the input generator; y k represents the sample of the local knowledge graph library; log represents the natural logarithm.

[0046] Through adversarial training between the generator and the discriminator, the generated data gradually approaches the real data that the discriminator cannot distinguish. iThe adjacent entities and connection relationships of the aligned entities in the discriminator are embedded as entities and relationships added to the knowledge graph on the discriminator side. This joint update mechanism achieves the alignment and knowledge sharing of the knowledge graphs of both parties through the continuous iteration of the adversarial generative network.

[0047] Preferably, the similarity matching of semantic triples according to the embedding representation of the local knowledge graph library includes:

[0048] The knowledge graph embedding technology is used to convert semantic triples into low-dimensional vector representations; the Euclidean distance between the low-dimensional vector representation of the semantic triple and the embedded representation of the triple in the local knowledge graph library is calculated, and a distance threshold ε is set. If the distance is less than the threshold, the triple is added to the set of similar semantic triples.

[0049] Preferably, the knowledge graph embedding technology includes: TransE algorithm, whose loss function is as follows:

[0050] L(h, r, t) = ||h+rt||

[0051] Among them, h represents the embedding vector of the head entity, r represents the embedding vector of the relationship, and t represents the embedding vector of the tail entity. The optimization goal is to minimize the loss function until the loss function converges to obtain a low-dimensional vector representation of the triple.

[0052] Preferably, the inputting of the text to be transmitted into a semantic extraction model for semantic information recognition includes: analyzing the input text using natural language processing technology, identifying entities in the input text through a named entity recognition (NER) model, and using a relational classification model to predict the relationship between the identified entities, thereby constructing semantic triples.

[0053] Preferably, the information x encoded by the transmitting end includes:

[0054] x=C(S(m))

[0055] Among them, S represents the source encoder and C represents the channel encoder.

[0056] Preferably, the receiving end receives a message y sent by the sending end, including:

[0057] y=wx+n

[0058] Among them, x is the information encoded by the transmitter, n is the independent and identically distributed vector of circularly symmetric Gaussian distribution, w~N(0,δ 2 I) indicates that w obeys N(0, δ 2 I) Probability distribution, δ 2 represents the average noise power of the channel, and I is the unit matrix.

[0059] Preferably, the performing channel and source decoding on the received information comprises:

[0060]

[0061] in, Represents the semantic triple information after decoding at the receiving end; S -1 represents the source decoder; C -1 In this embodiment, the source encoder adopts Huffman coding; the channel encoder adopts turbo coding; and the decoding method is the corresponding coding method.

[0062] In this embodiment, the sender obtains the text information to be transmitted, and uses the named entity recognition model (semantic extraction model) to obtain the entity-relationship-entity triple set (semantic triple) according to the text information to be transmitted. The identified semantic triple information is represented by an embedded vector according to the knowledge representation model (TransE algorithm) of the local knowledge graph, and the triple vector is matched with the knowledge graph library vector, and the triple vector in the local database is retained and aligned to obtain the semantic expression information that conforms to the local database.

[0063] The use of knowledge graphs to represent semantic information removes redundancy in the transmitted information, thereby improving the effectiveness of the present invention. After the receiving end receives the transmitted knowledge triples, it selects the triples with the highest similarity as semantic information based on the knowledge graph library matching. The knowledge graph text generation model (semantic recovery module) is used to restore text information. The present invention uses the T5 model that has been pre-trained with a large amount of data to realize the knowledge graph text generation function.

[0064] The research scenario of the present invention is in semantic communication. In order to simulate the process of learning and updating the semantic communication system model by joint knowledge graph on a single machine, a local data set needs to be allocated to each user terminal participating in the training. The data set used in this experiment is the WebNLG data set. Among them, the data set covers the triples and sentence information of airports, artists, astronauts, athletes, buildings, celestial bodies, cities, cartoon characters, food, transportation methods, monuments, politicians, sports teams, universities, literary works, movies, scientists and music works.

[0065] The triple information in the WebNLG dataset is evenly divided into 10 subsets. Each subset contains triples from different fields in the WebNLG dataset, ensuring that each subset covers diverse domain knowledge, such as people, places, events, works, etc. This division method ensures the breadth and representativeness of the data in each subset, facilitating subsequent cross-domain joint knowledge updates.

[0066] For each knowledge graph, we calculate the number of relations, entities, and triplets, and divide the triplets into training, validation, and test sets at a ratio of 90:5:5 according to the settings in the TransE model, and conduct triple classification test experiments to train knowledge graph embedding representations. The default dimension d of the embedding vector is set to d=100, the number of training rounds is set to 300, the learning rate is set to 0.01, and the batch size is set to 100. In the triple classification experiment, negative samples are generated by destroying the head or tail entity, and the ratio of negative samples to positive samples is set to 1:1.

[0067] For the 10 client knowledge graphs, they are jointly updated according to the fully connected joint update strategy. Two knowledge graphs are selected for joint update in each round. All knowledge graphs are updated in sequence. The goal is to ensure that each knowledge graph g i Can be with other g j All are jointly updated once. The training rounds of the adversarial generative network are set to 300, the batch size is set to 32, the learning rate is set to 0.01, and the momentum is set to 0.9. Among them, the generator receives the embedding vectors of the aligned entities of the two knowledge graphs, performs a linear transformation on the input entity embedding vector E, and uses matrix multiplication W×X to transform the embedding space. G=W·E, where W is the weight matrix to be learned and E is the input aligned entity embedding vector. The generator outputs the transformed vector as the input of the discriminator.

[0068] The task of the discriminator is to distinguish between embeddings from real knowledge graphs and embeddings generated by the generator. The discriminator is a multi-layer perceptron, which includes three fully connected layers and a normalization layer. After the activation function sigmoid, it outputs a binary classification probability value, which indicates the probability that the input embedding vector is a real embedding, and determines whether the input vector is generated or comes from real data.

[0069] The semantic recovery at the receiving end uses the T5 model that fine-tunes a large amount of data pre-training to implement the knowledge graph text generation method. Based on the WebNLG dataset, the training dataset and verification dataset of the text generation model are obtained, and the parameters of the triple text generation model are fine-tuned. The learning rate of the training model is set to 2e-5, the training batch size is 16, and the model is trained for a total of 40 epochs. Finally, the model outputs the training parameters, and the system uses the training parameters to realize triple recovery sentence information.

[0070] This experiment uses the T5 model that has been pre-trained with a large amount of data to realize the knowledge graph text generation function. Figure 3 The relationship between the bit length required to be sent and the sentence length is shown. Especially for long sentences, the method in this experiment achieves a high compression rate in the system. In order to explore the impact of different encoding and decoding methods on performance, such as Figure 4As shown in the figure, the relationship between SNR and BLEU score under AWGN channel is shown. The line marked with circular symbols represents the semantic communication system model, the line marked with triangle symbols represents the semantic communication system model without joint knowledge graph learning, and the line marked with square symbols represents the communication system combining Huffman source coding method with Turbo channel coding method. Based on image comparison analysis, it can be concluded that the proposed model performs better at low SNR, and its performance in high SNR area is not as good as the traditional communication system. This is because the proposed semantic recovery method attempts to recover the same semantics rather than the same sentence structure.

[0071] Based on the image comparison analysis, it can be concluded that, first, the proposed model performs better at low SNR and performs worse than the traditional communication system in the high SNR region. This is because the proposed semantic restoration method attempts to restore the same semantics rather than the same sentence structure. Secondly, the semantic communication system model learned by joint knowledge graphs improves the restoration quality in semantic restoration by integrating and aligning multiple knowledge graphs to extract entity relationships from a wider range of data sources. In contrast, the semantic communication system model that is not learned by joint knowledge graphs has poor text restoration effects due to the lack of some triples due to the inability to fully cover entity relationships. The experimental results show that the performance of the jointly updated knowledge graph in the text restoration task has significant advantages, which not only improves the extraction quality of knowledge triples, but also shows better semantic accuracy in the restored text.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A text semantic communication method based on joint knowledge graph learning, characterized in that: include: S1: Each user terminal uses knowledge graph embedding technology to transform its own knowledge graph library into a low-dimensional vector representation, and calculates the secure hash values ​​of entities, relationships, and entity attributes in the knowledge graph library through a secure hash algorithm; S2: The user ends complete the alignment of entities, relationships and entity attributes in the knowledge graph library through the secure hash values ​​of entities, relationships and entity attributes; S3: Construct a generative adversarial network system, set the generator and discriminator on two different user ends respectively, and unify the embedding representation of the knowledge graph libraries on different user ends by training the generative adversarial network; S4: The sending user terminal inputs the text to be transmitted into the semantic extraction model for semantic information recognition; and constructs semantic triples according to the recognized semantic information; performs similarity matching on the semantic triples according to the embedded representation of the local knowledge graph library to extract a set of similar semantic triples; performs source coding and channel coding on the extracted similar semantic triples, and sends the coded information to the receiving end; S5: The receiving user receives the message sent by the sending end, performs channel decoding and source decoding on the received message to obtain a set of similar semantic triples; uses a string similarity algorithm to calculate the similarity between similar semantic triples and the local knowledge graph library, and selects the most similar triple as the restored semantic triple; uses a fine-tuned semantic recovery module to convert the restored semantic triple into a restored text.

2. According to claim 1, a text semantic communication method based on joint knowledge graph learning is characterized in that: The method of unifying the embedding representations of different user-side knowledge graph libraries by training a generative adversarial network includes: Use knowledge graph embedding technology to convert the knowledge graph in the local knowledge graph library into a low-dimensional vector representation; The generator receives the low-dimensional vector representation from the local knowledge graph as input and outputs the fused embedding representation; The discriminator receives the embedding representation generated by the generator or the low-dimensional vector representation from the local knowledge graph as input and outputs a probability value representing the authenticity of the data; the loss function L of the generator is G for: The loss function L of the discriminator D for: Among them, G() represents the generator, D() represents the discriminator; n represents the number of training samples; x m represents the sample of the input generator; y k represents the sample of the local knowledge graph library; log represents the natural logarithm; After the training of the generative adversarial network is completed, the embedding representation in the local knowledge base of the user side is updated through the embedding representation generated by the generator to unify the embedding representation of different user-side knowledge graph libraries.

3. According to claim 2, a text semantic communication method based on joint knowledge graph learning is characterized in that: The similarity matching of semantic triples according to the embedding representation of the local knowledge graph library includes: The knowledge graph embedding technology is used to convert semantic triples into low-dimensional vector representations; the Euclidean distance between the low-dimensional vector representation of the semantic triple and the embedded representation of the triple in the local knowledge graph library is calculated, and a distance threshold ε is set. If the distance is less than the threshold, the triple is added to the set of similar semantic triples.

4. A text semantic communication method based on joint knowledge graph learning according to claim 3, characterized in that: The knowledge graph embedding technology includes: TransE algorithm, whose loss function is as follows: L(h,r,t)=‖h+rt‖ Among them, h represents the embedding vector of the head entity, r represents the embedding vector of the relationship, and t represents the embedding vector of the tail entity. The optimization goal is to minimize the loss function until the loss function converges to obtain a low-dimensional vector representation of the triple.

5. A text semantic communication method based on joint knowledge graph learning according to claim 2, characterized in that: The inputting of the text to be transmitted into the semantic extraction model for semantic information recognition includes: analyzing the input text using natural language processing technology, identifying entities in the input text through a named entity recognition (NER) model, predicting the relationship between the identified entities using a relational classification model, and then constructing semantic triples.

6. A text semantic communication method based on joint knowledge graph learning according to claim 1, characterized in that: The information x encoded by the transmitting end includes: x=C(S(m)) Among them, S represents the source encoder and C represents the channel encoder.

7. A text semantic communication method based on joint knowledge graph learning according to claim 1, characterized in that: The message y sent by the sending end and received by the receiving end includes: y=wx+h Among them, x is the information encoded by the transmitter, n is the independent and identically distributed vector of circularly symmetric Gaussian distribution, w~N(0,δ 2 I) indicates that w obeys N(0,δ 2 I) Probability distribution, δ 2 represents the average noise power of the channel, and I is the unit matrix.

8. The text semantic communication method based on joint knowledge graph learning according to claim 1, characterized in that: The channel and source decoding of the received information comprises: in, Represents the semantic triple information after decoding at the receiving end; S -1 represents the source decoder; C -1 Represents a channel decoder.

Images