Text semantic communication method and device, electronic equipment and storage medium

Abstract

The invention provides a text semantic communication method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the semantic information extraction of a target text sent by a sending end, and constructing a semantic subgraph based on the extracted entity information and relation information of the target text; performing semantic matching processing on the semantic sub-graph and the shared knowledge graph, and determining newly added nodes and newly added edges; forming incremental semantic data to be transmitted for an incremental sub-graph formed by the newly-added nodes and the newly-added edges, encoding the incremental semantic data to generate an estimation symbol, and transmitting the estimation symbol to a receiving end through a physical channel; and performing denoising reconstruction processing on the estimation symbol at a receiving end to recover an increment sub-graph, fusing the recovered increment sub-graph and the shared knowledge graph to reconstruct a complete semantic sub-graph, and restoring a target text based on the complete semantic sub-graph. The incremental transmission mechanism avoids repeated transmission of known knowledge, and effectively relieves the transmission pressure of high semantic density information under a low-bandwidth or high-delay channel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, in particular to a text semantic communication method and device, electronic equipment and storage medium. BACKGROUND

[0002] Semantic communication, as a new paradigm for semantic information transmission, has attracted widespread attention in recent years. Traditional communication focuses on the effective transmission of bit streams. Semantic communication seeks to convey the meaning of information, rather than its original representation. Specifically, in semantic communication, the sender extracts semantic information from data. After transmission through the channel, the information is reconstructed at the semantic level at the receiving end. This not only greatly reduces the amount of data transmission, but also improves the intelligent level of the communication system. However, although some progress has been made in semantic communication technology in recent years, the existing semantic communication framework still has deficiencies in the efficiency of semantic transmission and semantic compression. SUMMARY

[0003] Therefore, the purpose of the present application is to provide a text semantic communication method and device, electronic equipment and storage medium, by constructing a semantic subgraph at the sending end and performing semantic matching with a shared knowledge graph shared by both parties, only extracting and transmitting new node and edge information (i.e. incremental subgraph), realizing efficient compression for structured semantics. The incremental transmission mechanism avoids repeated transmission of known knowledge, greatly reducing the amount of data to be transmitted through the channel. Especially in high-frequency co-occurrence context or long-term interaction scenarios, the compression rate is significantly improved, effectively alleviating the transmission pressure of high semantic density information in low bandwidth or high delay channels.

[0004] The text semantic communication method provided by the embodiment of the present application comprises: extracting semantic information from the target text sent by the sending end based on a triple extraction model, and constructing a semantic subgraph of the target text based on the extracted entity information and relationship information of the target text; wherein the semantic subgraph is a directed graph structure of the target text; performing semantic matching processing on the semantic subgraph and a preset shared knowledge graph, and determining new nodes and new edges of the shared knowledge graph; forming an incremental subgraph composed of the new nodes and the new edges into incremental semantic data to be transmitted, encoding the incremental semantic data to generate estimated symbols, and transmitting the estimated symbols to the receiving end through a physical channel; performing denoising reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph, fusing the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph, and restoring the target text based on the complete semantic subgraph.

[0005] In a possible implementation, the semantic information extraction model based on triplets extracts semantic information of the target text sent by the sending end, and the method comprises the following steps: The semantic information extraction model based on triplets extracts semantic information of the target text sent by the sending end, and the method comprises the following steps: The encoder extracts a context-aware vector based on the target text, and determines a latent feature vector of the target text; The decoder performs autoregressive processing on the latent feature vector, and generates a feature sequence of the target text;

[0006] In a possible implementation, the semantic information extraction model based on triplets is determined by the following steps: The sample text is input into the neural network model to extract semantic information, and the predicted triplet information of the sample text is determined; The loss function is used to process the predicted triplet information, and the loss value of the neural network model is determined; The neural network model is iteratively trained based on the loss value, and the semantic information extraction model based on triplets is determined.

[0007] In a possible implementation, the semantic subgraph and the preset shared knowledge graph are matched to determine the new nodes and new edges of the shared knowledge graph, and the method comprises the following steps: The entities and relationships in the semantic subgraph and the shared knowledge graph are mapped to a unified vector space; The cosine similarity between each node and edge in the semantic subgraph and the most similar element in the shared knowledge graph is determined based on a cosine similarity function; If the cosine similarity is less than a preset node threshold, the node is determined to be a new node, and if the cosine similarity is less than a preset edge threshold, the edge is determined to be a new edge.

[0008] In a possible implementation, the incremental semantic data is encoded to generate an estimated symbol, and the method comprises the following steps: The incremental semantic data is mapped to a continuous semantic vector, and noise disturbance is introduced into the semantic vector in the training stage, so that the encoder learns robust semantic features with noise resistance; The encoder based on the Transformer performs context modeling on the semantic vector with noise disturbance to determine a context feature vector, and maps the context feature vector to a low-dimensional channel symbol through a projection matrix; The channel symbol is modulated and input into a physical channel to obtain an observation signal of the receiving end, and the observation signal is equalized to generate an estimated symbol.

[0009] In one possible implementation, the step of performing denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph includes: The estimated symbols are processed to eliminate channel noise based on a residual denoising network layer consisting of a one-dimensional convolutional layer, a ReLU activation function, and residual connections, thereby determining the first feature; The estimated symbols are decoded using a multi-layer cross-attention decoder to determine the second feature; The first feature and the second feature are added together to recover the incremental subgraph.

[0010] In one possible implementation, after the target text is reconstructed based on the complete semantic subgraph, the text semantic communication method further includes: At the receiving end, new nodes or edges that appear more frequently than a set threshold in historical communications are identified. Under the condition of satisfying storage and latency constraints, the shared knowledge graph is updated based on the new nodes or edges that appear more frequently than a set threshold.

[0011] This application embodiment also provides a text semantic communication device, the text semantic communication device comprising: The semantic subgraph construction module is used to extract semantic information from the target text sent by the sending end based on the triple extraction model, and to construct a semantic subgraph of the target text based on the entity information and relation information of the extracted target text; wherein, the semantic subgraph is a directed graph structure of the target text; The incremental determination module is used to perform semantic matching processing between the semantic subgraph and the preset shared knowledge graph to determine the newly added nodes and newly added edges of the shared knowledge graph. The encoding module is used to form incremental semantic data to be transmitted from the incremental subgraph composed of the newly added node and the newly added edge, encode the incremental semantic data to generate estimated symbols, and transmit them to the receiving end through a physical channel. The reconstruction module is used to perform denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph, fuse the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph, and restore the target text based on the complete semantic subgraph.

[0012] This application also provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the text semantic communication method described above are performed.

[0013] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the text semantic communication method described above.

[0014] This application provides a text semantic communication method, apparatus, electronic device, and storage medium. The text semantic communication method includes: extracting semantic information from target text sent by a sender based on a triplet extraction model; constructing a semantic subgraph of the target text based on the extracted entity information and relational information of the target text; wherein the semantic subgraph is a directed graph structure of the target text; performing semantic matching processing on the semantic subgraph and a preset shared knowledge graph to determine newly added nodes and edges of the shared knowledge graph; forming incremental semantic data to be transmitted from the incremental subgraph composed of the newly added nodes and edges; encoding the incremental semantic data to generate estimated symbols and transmitting them to a receiver through a physical channel; performing denoising and reconstruction processing on the estimated symbols at the receiver to recover the incremental subgraph; fusing the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph; and reconstructing the target text based on the complete semantic subgraph. By utilizing a shared set of semantic triplet vectors and a unique set of semantic triplet vectors, semantic hierarchical transmission is achieved, reducing the transmission of redundant data and realizing efficient semantic extraction and transmission. By constructing a semantic subgraph at the sending end and semantically matching it with a shared knowledge graph shared by both communicating parties, only newly added node and edge information (i.e., the incremental subgraph) is extracted and transmitted, achieving efficient compression for structured semantics. This incremental transmission mechanism avoids the repeated transmission of known knowledge, significantly reducing the amount of data that needs to be transmitted through the channel. Especially in high-frequency co-occurrence contexts or long-term interaction scenarios, the compression rate is significantly improved, effectively alleviating the transmission pressure of high semantic density information under low bandwidth or high-latency channels.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a text semantic communication method provided in an embodiment of this application; Figure 2This is one of the structural schematic diagrams of a text semantic communication device provided in the embodiments of this application; Figure 3 This is a second schematic diagram of the structure of a text semantic communication device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0019] First, the applicable application scenarios of this application will be introduced. This application can be applied to the field of wireless communication technology.

[0020] Research has revealed that semantic communication, as an emerging paradigm for semantic information transmission, has attracted widespread attention in recent years. Traditional communication focuses on the efficient transmission of bit streams. Semantic communication, however, seeks to convey the meaning of information, rather than its raw representation. Specifically, in semantic communication, the sender extracts semantic information from the data. After transmission through the channel, the information is semantically reconstructed at the receiving end. This not only significantly reduces data transmission volume but also enhances the intelligence of the communication system. However, despite the progress made in semantic communication technology in recent years, existing semantic communication frameworks still fall short in terms of the efficiency of semantic transmission.

[0021] Based on this, embodiments of this application provide a text semantic communication method. By constructing a semantic subgraph at the sending end and semantically matching it with a shared knowledge graph shared by both communicating parties, only newly added node and edge information (i.e., incremental subgraph) is extracted and transmitted, achieving efficient compression for structured semantics. This incremental transmission mechanism avoids the repeated transmission of known knowledge, significantly reducing the amount of data that needs to be transmitted through the channel. Especially in high-frequency co-occurrence contexts or long-term interaction scenarios, the compression rate is significantly improved, effectively alleviating the transmission pressure of high semantic density information under low bandwidth or high latency channels.

[0022] Please see Figure 1 , Figure 1This is a flowchart illustrating a text semantic communication method provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the text semantic communication method includes: S101: Semantic information is extracted from the target text sent by the sending end based on the triple extraction model, and a semantic subgraph of the target text is constructed based on the entity information and relation information of the extracted target text; wherein, the semantic subgraph is a directed graph structure of the target text.

[0023] In this step, the triplet extraction model is used to extract semantic information from the target text sent by the sending end to determine the triplet information set of the target text, and a semantic subgraph is constructed based on the triplet information set of the target text.

[0024] Semantic information extraction is the process of extracting information from natural language text. Extracting core semantic information It is represented in the form of a subject-relationship-object triple, denoted as .in, Represents a set of entities. Represents a set of relations. In a triplet... and These refer to the subject and object in the text, respectively. This describes the semantic relationship between the two.

[0025] It should be noted that the semantic subgraph is a sub-knowledge graph of the target text. ,in A collection of semantic entity (or attribute) nodes. This represents the set of edges representing relationships between entities. The process of extracting entities and relationships from natural text can be formalized as a learnable mapping. Where θ represents the model parameters. To facilitate subsequent matching and incremental extraction, nodes and relations are mapped to the same vector space. Let the embedding functions for nodes and relations be denoted as follows: .in, Let be the node's embedding vector. This is the embedding vector of the relation.

[0026] In one possible implementation, the semantic information extraction of the target text sent by the sender based on the triplet extraction model includes: A: The triplet extraction model extracts context-aware vectors from the target text based on the encoder, thereby determining the latent feature vectors of the target text.

[0027] Here, the latent feature vector of the target text X is determined using the following formula:

[0028] in, For potential feature vectors, It is determined by parameters Defined encoding function.

[0029] B: Based on the decoder, autoregressive processing is performed on the latent feature vectors to generate the feature sequence of the target text.

[0030] Here, the decoder is used to perform autoregressive processing on the latent feature vectors to generate the feature sequence of the target text.

[0031] C: The feature sequence is mapped using a mapping function based on triples to extract entity information and relational information of the target text.

[0032] Here, the mapping function of triples is used to map the feature sequence that maximizes the conditional probability, thereby determining the entity information and relational information of the extracted target text.

[0033] To achieve high-quality semantic extraction, the system employs the Text-to-Text Transfer Transformer (T5) model, modeling the triple extraction task as a sequence-to-sequence generation task. The input text X is encoded into an intermediate representation by the T5 model, and its encoder-decoder structure is used to generate a structured output T. This process can be represented as a conditional probability maximization problem:

[0034] in, Y It is the text representation sequence of the target text. For position t The output sequence, Extracting parameters for the model from triples. P For probability, This represents the number of sequences of the output tokens. X For the target text, It represents the current position in the target sequence. t The previously generated output sequence.

[0035] In one possible implementation, the triplet extraction model is determined through the following steps: (1): Input the sample text into the neural network model to extract semantic information and determine the predicted triplet information of the sample text.

[0036] Here, the process of using a neural network model to determine the predicted triplet information is consistent with the process of using the triplet extraction model to determine the triplet information, and will not be elaborated on here.

[0037] (2): The loss value of the neural network model is determined by processing the predicted triplet information based on the loss function.

[0038] The loss function is defined as follows:

[0039] Where L is the loss value and m is the sample size. Indicates sample i A collection of entities Indicates sample i A set of entity relationships. and These refer to the subject and object in the text, respectively. This describes the semantic relationship between the two. P For probability, For model parameters, O is the latent feature vector of the sample text.

[0040] (3): The neural network model is iteratively trained based on the loss value to determine the triple extraction model.

[0041] Here, if the loss value is greater than or equal to a preset threshold, the network parameters of the neural network model are adjusted, and the adjusted neural network model is trained again. If the loss value is less than the preset threshold, the network parameters of the neural network model are not adjusted, and the neural network model is used as a triple extraction model.

[0042] S102: Perform semantic matching processing between the semantic subgraph and the preset shared knowledge graph to determine the newly added nodes and edges of the shared knowledge graph.

[0043] In this step, the semantic subgraph is semantically matched with the shared knowledge graph pre-defined by both communicating parties to identify newly added nodes and edges that do not exist in the shared knowledge graph.

[0044] In one possible implementation, the step of performing semantic matching processing between the semantic subgraph and a preset shared knowledge graph to determine the newly added nodes and edges of the shared knowledge graph includes: The entities and relations in the semantic subgraph and the shared knowledge graph are mapped to a unified vector space; the cosine similarity between each node and edge in the semantic subgraph and the most similar element in the shared knowledge graph is determined based on the cosine similarity function; if the cosine similarity is less than a preset node threshold, the node is determined to be a new node, and if the cosine similarity is less than a preset edge threshold, the edge is determined to be a new edge.

[0045] It's important to note that for a node to be identified in a semantic subgraph, its "most similar element" is an existing entity node in the shared knowledge graph, making their semantic vector representations in the unified embedding space closest. Similarly, for a relation edge to be identified, its "most similar element" is an edge existing in the shared graph, especially one whose relation is semantically closest to its own. This similarity is quantified using cosine similarity and, combined with knowledge graph embedding (KGE) models such as TransE, maps structured triples to a continuous vector space, thus supporting semantic alignment across graphs.

[0046] Here, let the shared knowledge graph of the two communicating parties be... To determine semantic subgraphs Does it already exist? Among them A collection of semantic entity (or attribute) nodes. This is the set of edges representing relationships between entities. First, consider... internal nodes With edge Embedded into the same semantic space, we obtain and Calculate the cosine similarity for both nodes and edges:

[0047]

[0048] For any node in the semantic subgraph With edge The score of the element most similar to the shared knowledge graph is recorded as follows:

[0049]

[0050] in, For semantic subgraphs Cosine similarity with the most similar element For semantic subgraphs The cosine similarity between nodes and their most similar elements is given by the threshold values ​​for determining nodes and edges. Therefore, a new node is added. With the addition of borders The set is defined as:

[0051]

[0052] Here, the incremental subgraph to be transmitted is thus obtained. .

[0053] S103: The incremental semantic data to be transmitted is formed from the incremental subgraph composed of the newly added node and the newly added edge. The incremental semantic data is encoded to generate estimated symbols and transmitted to the receiving end through the physical channel.

[0054] In this step, incremental semantic data to be transmitted is formed from the incremental subgraph composed of newly added nodes and edges. The incremental semantic data is encoded to generate estimated symbols and transmitted to the receiving end through a physical channel.

[0055] Here, the bit-level load of the complete subgraph is recorded to measure the compression gain. for:

[0056] in, The average number of bits used to encode nodes and edges. This includes necessary metadata (position, alignment identifiers, etc.). The corresponding incremental transmission payload. for:

[0057]

[0058] in, For incremental compression ratio, The larger the value, the more fully the compression.

[0059] Here, to ensure that the receiving end can correctly combine the incremental information with the information in the shared graph to reconstruct a subgraph consistent with the sending end, sufficient structural and location information must be transmitted. Therefore, during the transmission of incremental information, this invention additionally transmits its structural location identifier. Specifically, each newly added node carries a unique identifier and its context information in the original subgraph. For each newly added edge, in addition to the embedded representation of the relation type 𝑟, the identifiers of its two endpoints are also transmitted. Based on this, the receiving end can find the corresponding node position in the shared graph and add new relations therein.

[0060] In this application, compared to traditional methods that explicitly encode and transmit triples one by one, this invention performs semantic modeling and incremental extraction on a subgraph structure basis, preserving the topological associations and contextual dependencies between entities, thus enhancing the integrity and structure of the semantic representation. Simultaneously, by introducing structural location identifiers and contextual positioning information, it ensures that the receiving end can accurately reconstruct the topological structure of the subgraph, improving the accuracy and interpretability of semantic reconstruction.

[0061] In one possible implementation, encoding the incremental semantic data to generate estimated symbols includes: a: The incremental semantic data is mapped into continuous semantic vectors, and noise perturbation is introduced into the semantic vectors during the training phase, so that the encoder learns robust semantic features with noise resistance.

[0062] Here, based on the aforementioned matching and incremental extraction mechanism, the incremental subgraph to be transmitted is obtained. Subsequently, the system uses the embedding function of nodes and relationships to... Mapping to continuous semantic vectors :

[0063] in, For the vector representation of the newly added entity, This is a vector representation of the newly added relationship.

[0064] Here, to enhance robustness under channel noise conditions, the encoder introduces a robust feature learning layer before generating symbolic representations. This layer explicitly introduces noise perturbations during training.

[0065] b: The Transformer-based encoder performs context modeling on the semantic vector with introduced noise perturbation to determine the context feature vector, and maps the context feature vector to a low-dimensional channel symbol through the projection matrix.

[0066] Here, through a multi-layer self-attention mechanism and a feedforward network, a context-aware context feature vector is obtained. Next, through the channel projection matrix Dimensional mapping and compression are performed to obtain the channel symbols: in, c This is the channel symbol.

[0067] c: After modulating the channel symbols and inputting them into the physical channel, the observed signal at the receiver is obtained. The observed signal is then subjected to equalization processing to generate estimated symbols.

[0068] Here, the channel symbol After modulation, it is input into the physical channel. Taking the AWGN channel as an example, the signal observed at the receiver is... in Indicates channel noise. This refers to noise power. Under fading channel conditions, ,in This is the channel gain matrix. The receiver first obtains the estimated symbol using an equalizer. .

[0069] S104: At the receiving end, the estimated symbols are denoised and reconstructed to recover the incremental subgraph. The recovered incremental subgraph is fused with the shared knowledge graph to reconstruct a complete semantic subgraph. The target text is then restored based on the complete semantic subgraph.

[0070] In this step, the estimated symbols are denoised and reconstructed at the receiving end to recover the incremental subgraph. The recovered incremental subgraph is then fused with the shared knowledge graph to reconstruct the complete semantic subgraph. The target text is then restored based on the complete semantic subgraph.

[0071] This invention proposes a hierarchical communication mechanism of "shared large knowledge graph - incremental transmission of local subgraphs". The sending end only extracts and transmits newly added nodes and edge information in the sub-knowledge graph corresponding to the text that do not exist in the shared graph. The receiving end completes semantic reconstruction by fusing incremental information. This mechanism avoids redundant transmission of repetitive semantic content, significantly compresses the amount of data to be transmitted, and achieves extremely high compression rates, especially in high-frequency co-occurrence semantic scenarios, effectively improving the spectrum utilization and energy efficiency of the communication system.

[0072] In one possible implementation, the step of performing denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph includes: The estimated symbols are processed to eliminate channel noise using a residual denoising network layer consisting of a one-dimensional convolutional layer, a ReLU activation function, and residual connections to determine a first feature; the estimated symbols are then decoded using a multi-layer cross-attention decoder to determine a second feature; the first feature and the second feature are added together to recover the incremental subgraph.

[0073] Here, the decoder structure introduces a residual denoising network layer on top of the traditional Transformer. This residual denoising network layer eliminates channel noise by learning the difference between the input signal and clean semantic features, and is ultimately added to the output of the Transformer decoder to obtain a robust semantic representation. Its input-output relationship can be expressed as:

[0074] in, It is a multi-layer cross-attention decoder. This is a residual denoising network, consisting of a one-dimensional convolutional layer (Conv1D) + residual connections + ReLU activation. It can effectively remove channel noise and preserve semantic features during the decoding process. This is the recovered incremental subgraph.

[0075] Here, to achieve complete semantic reconstruction, the receiving end merges the recovered incremental subgraph with the shared knowledge spectrum to obtain the expanded complete semantic subgraph: And with the help of the text reconstruction module Restore it to natural language.

[0076] Among them, the text reconstruction module A decoder is employed to adapt semantic communication scenarios through prefix-tuning. Given a complete semantic subgraph... The process of generating natural language text can be formalized as follows:

[0077] in, This represents the generated text sequence. The generation model not only focuses on the semantic relationships within the triples but also ensures the fluency and readability of the reconstructed text. The goal of text reconstruction is to maximize the similarity between the generated text and the original text. To ensure the quality of the generated text, the cross-entropy loss function is used to effectively improve text reconstruction performance.

[0078] In this application, a dual denoising enhancement mechanism is introduced into the Transformer-based end-to-end semantic communication framework: a robust semantic feature learning module is designed in the encoding stage, and the model learns interference-resistant semantic representations through noise injection training; a residual denoising module is introduced in the decoding stage, which combines a one-dimensional convolutional residual network to perform noise correction on the channel output symbols. This collaborative mechanism effectively suppresses semantic distortion caused by complex channels such as AWGN and Rayleigh fading, and significantly improves the semantic recovery performance of the system under adverse channel conditions.

[0079] In one possible implementation, after the target text is reconstructed based on the complete semantic subgraph, the text semantic communication method further includes: At the receiving end, new nodes or edges that appear more frequently than a set threshold in historical communications are identified. Under the condition of satisfying storage and latency constraints, the shared knowledge graph is updated based on the new nodes or edges that appear more frequently than a set threshold.

[0080] This application employs a frequency-driven dynamic knowledge graph update strategy, which statistically analyzes frequently occurring incremental semantic elements at the receiving end and incorporates them into the shared knowledge graph while satisfying resource constraints. As the communication process continues, the coverage of the shared graph gradually expands, and the matching hit rate continuously improves, forming a positive feedback loop of "increased compression rate → reduced transmission burden → enhanced system performance," demonstrating good adaptive evolution capabilities.

[0081] In a specific embodiment, the sending end first performs semantic parsing on the input text, extracting entities and relationships to construct a corresponding semantic subgraph. This subgraph not only depicts the semantic content carried by the text but also preserves the semantic dependencies and associations between entities in a structured graphical form. Since both communicating parties pre-share a large-scale knowledge graph, after the sending end completes the subgraph construction, it compares it with the shared knowledge graph, retaining and encoding only the newly added node and edge information. The receiving end can reconstruct a complete subgraph consistent with the sending end by merging the received newly added nodes and edges into the existing shared knowledge graph. Compared with traditional transmission methods, this incremental update mechanism significantly reduces the amount of redundant data transmitted, improving communication efficiency while ensuring semantic integrity. After extracting the incremental information of the subgraph, the transmission module is responsible for mapping it to the physical channel for transmission and reconstruction. This invention introduces a denoising enhancement mechanism during the encoding and decoding process to address the semantic distortion problem under low signal-to-noise ratio conditions. At the transmitting end, the encoder not only performs embedding and contextual representation modeling of the subgraph incremental information, but also applies noise modeling to the input semantic vector through a robust semantic feature learning layer. This explicitly simulates channel disturbances during the training phase, making the learned representation noise-resistant. Subsequently, these robustly optimized features are mapped to channel symbols, modulated, and transmitted through physical channels such as AWGN and Rayleigh. At the receiving end, after signal equalization and symbol recovery, a residual denoising module is introduced to correct the decoded input. This module effectively eliminates interference from channel noise by combining residual mapping with a multi-layer denoising network, enabling the subsequent Transformer decoder to reconstruct accurate subgraph incremental information in a relatively clean feature space. Finally, the receiving end merges the newly added nodes and edges obtained from the decoding with the shared knowledge graph to obtain a complete subgraph consistent with that of the transmitting end, and then uses a graph-to-text generation model to recover it into natural language text.

[0082] In this application, the natural language text is semantically parsed at the sending end to construct a semantic subgraph corresponding to the text's semantics. This subgraph is then matched and aligned with a large-scale knowledge graph shared by both communicating parties, with only newly added nodes and edges encoded and transmitted. The receiving end uses this subgraph to reconstruct the subgraph and merge it with the local large knowledge graph, thus ensuring complete semantic reconstruction with minimal overhead. Unlike traditional explicit transmission of triples one by one, this scheme utilizes the holistic representation of the subgraph structure and an incremental extraction mechanism to significantly reduce redundant data while maintaining readability and reasonability.

[0083] This application provides a text semantic communication method, which includes: extracting semantic information from target text sent by a sender based on a triplet extraction model; constructing a semantic subgraph of the target text based on the extracted entity information and relational information; wherein the semantic subgraph is a directed graph structure of the target text; performing semantic matching processing on the semantic subgraph and a preset shared knowledge graph to determine newly added nodes and edges of the shared knowledge graph; forming incremental semantic data to be transmitted from the incremental subgraph composed of the newly added nodes and edges; encoding the incremental semantic data to generate estimated symbols and transmitting them to a receiver through a physical channel; performing denoising and reconstruction processing on the estimated symbols at the receiver to recover the incremental subgraph; fusing the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph; and reconstructing the target text based on the complete semantic subgraph. By utilizing a shared set of semantic triplet vectors and a unique set of semantic triplet vectors, semantic hierarchical transmission is achieved, reducing the transmission of redundant data and realizing efficient semantic extraction and transmission. By constructing a semantic subgraph at the sending end and semantically matching it with a shared knowledge graph shared by both communicating parties, only newly added node and edge information (i.e., the incremental subgraph) is extracted and transmitted, achieving efficient compression for structured semantics. This incremental transmission mechanism avoids the repeated transmission of known knowledge, significantly reducing the amount of data that needs to be transmitted through the channel. Especially in high-frequency co-occurrence contexts or long-term interaction scenarios, the compression rate is significantly improved, effectively alleviating the transmission pressure of high semantic density information under low bandwidth or high-latency channels.

[0084] Please see Figure 2 , Figure 3 , Figure 2 This is one of the structural schematic diagrams of a text semantic communication device provided in the embodiments of this application; Figure 3 This is a second schematic diagram of the structure of a text semantic communication device provided in an embodiment of this application. Figure 2 As shown, the text semantic communication device 200 includes: The semantic subgraph construction module 210 is used to extract semantic information from the target text sent by the sending end based on the triple extraction model, and to construct a semantic subgraph of the target text based on the entity information and relation information of the extracted target text; wherein, the semantic subgraph is a directed graph structure of the target text; The incremental determination module 220 is used to perform semantic matching processing between the semantic subgraph and the preset shared knowledge graph to determine the newly added nodes and newly added edges of the shared knowledge graph. The encoding module 230 is used to form incremental semantic data to be transmitted from the incremental subgraph composed of the newly added node and the newly added edge, encode the incremental semantic data to generate estimated symbols, and transmit them to the receiving end through a physical channel. The reconstruction module 240 is used to perform denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph, fuse the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph, and restore the target text based on the complete semantic subgraph.

[0085] Furthermore, the semantic subgraph construction module 210 is used to extract semantic information from the target text sent by the sending end based on the triplet extraction model: The triplet extraction model extracts context-aware vectors from the target text based on the encoder, thereby determining the latent feature vectors of the target text. Based on the decoder, the potential feature vectors are subjected to autoregressive processing to generate the feature sequence of the target text; The feature sequence is mapped using a triplet mapping function to extract entity information and relational information of the target text.

[0086] Furthermore, such as Figure 3 As shown, the text semantic communication device 200 also includes a model training module 250, which determines the triple extraction model through the following steps: The sample text is input into a neural network model to extract semantic information and determine the predicted triplet information of the sample text. The loss value of the neural network model is determined by processing the predicted triplet information based on the loss function. The neural network model is iteratively trained based on the loss value to determine the triple extraction model.

[0087] Furthermore, the incremental determination module 220 is used to perform semantic matching processing between the semantic subgraph and the preset shared knowledge graph to determine the newly added nodes and newly added edges of the shared knowledge graph: The entities and relations in the semantic subgraph and the shared knowledge graph are mapped to a unified vector space; The cosine similarity between each node and edge in the semantic subgraph and the most similar element in the shared knowledge graph is determined based on the cosine similarity function. If the cosine similarity is less than a preset node threshold, the node is determined to be a new node; if the cosine similarity is less than a preset edge threshold, the edge is determined to be a new edge.

[0088] Furthermore, the encoding module 230 is used to encode the incremental semantic data to generate estimated symbols: The incremental semantic data is mapped to continuous semantic vectors, and noise perturbation is introduced into the semantic vectors during the training phase, so that the encoder learns robust semantic features with noise resistance. The Transformer-based encoder performs context modeling on the semantic vector with introduced noise perturbation to determine the context feature vector, and maps the context feature vector to a low-dimensional channel symbol through a projection matrix. The channel symbols are modulated and input into the physical channel to obtain the observation signal at the receiver. The observation signal is then equalized to generate an estimated symbol.

[0089] Furthermore, the reconstruction module 240 is used to perform denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph: The estimated symbols are processed to eliminate channel noise based on a residual denoising network layer consisting of a one-dimensional convolutional layer, a ReLU activation function, and residual connections, thereby determining the first feature; The estimated symbols are decoded using a multi-layer cross-attention decoder to determine the second feature; The first feature and the second feature are added together to recover the incremental subgraph.

[0090] Furthermore, such as Figure 3 As shown, the text semantic communication device 200 further includes an update module 260, which is used for: At the receiving end, new nodes or edges that appear more frequently than a set threshold in historical communications are identified. Under the condition of satisfying storage and latency constraints, the shared knowledge graph is updated based on the new nodes or edges that appear more frequently than a set threshold.

[0091] This application provides a text semantic communication device, comprising: a semantic subgraph construction module, used to extract semantic information from target text sent by a sender based on a triplet extraction model, and construct a semantic subgraph of the target text based on the extracted entity information and relation information of the target text; wherein the semantic subgraph is a directed graph structure of the target text; an incremental determination module, used to perform semantic matching processing on the semantic subgraph and a preset shared knowledge graph to determine the newly added nodes and newly added edges of the shared knowledge graph; an encoding module, used to form incremental semantic data to be transmitted from the incremental subgraph composed of the newly added nodes and newly added edges, encode the incremental semantic data to generate estimated symbols, and transmit them to a receiver through a physical channel; and a reconstruction module, used to perform denoising and reconstruction processing on the estimated symbols at the receiver to recover the incremental subgraph, fuse the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph, and restore the target text based on the complete semantic subgraph. By constructing a semantic subgraph at the sending end and semantically matching it with a shared knowledge graph shared by both communicating parties, only newly added node and edge information (i.e., the incremental subgraph) is extracted and transmitted, achieving efficient compression for structured semantics. This incremental transmission mechanism avoids the repeated transmission of known knowledge, significantly reducing the amount of data that needs to be transmitted through the channel. Especially in high-frequency co-occurrence contexts or long-term interaction scenarios, the compression rate is significantly improved, effectively alleviating the transmission pressure of high semantic density information under low bandwidth or high-latency channels.

[0092] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0093] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, they can perform the operations described above. Figure 1 The steps of the text semantic communication method in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0094] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of the text semantic communication method in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0095] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0096] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0097] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0099] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0100] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A text semantic communication method, characterized in that, The text semantic communication method includes: Semantic information is extracted from the target text sent by the sending end based on the triple extraction model. A semantic subgraph of the target text is constructed based on the entity information and relation information of the extracted target text. The semantic subgraph is a directed graph structure of the target text. The semantic subgraph is semantically matched with a preset shared knowledge graph to determine the newly added nodes and edges of the shared knowledge graph. Incremental semantic data to be transmitted is formed from the incremental subgraph composed of the newly added nodes and the newly added edges. The incremental semantic data is encoded to generate estimated symbols and transmitted to the receiving end through a physical channel. At the receiving end, the estimated symbols are denoised and reconstructed to recover the incremental subgraph. The recovered incremental subgraph is then fused with the shared knowledge graph to reconstruct a complete semantic subgraph. Based on the complete semantic subgraph, the target text is restored.

2. The text semantic communication method according to claim 1, characterized in that, The semantic information extraction of the target text sent by the sender based on the triplet extraction model includes: The triplet extraction model extracts context-aware vectors from the target text based on the encoder, thereby determining the latent feature vectors of the target text. Based on the decoder, the potential feature vectors are subjected to autoregressive processing to generate the feature sequence of the target text; The feature sequence is mapped using a triplet mapping function to extract entity information and relational information of the target text.

3. The text semantic communication method according to claim 1, characterized in that, The triplet extraction model was determined through the following steps: The sample text is input into a neural network model to extract semantic information and determine the predicted triplet information of the sample text. The loss value of the neural network model is determined by processing the predicted triplet information based on the loss function. The neural network model is iteratively trained based on the loss value to determine the triple extraction model.

4. The text semantic communication method according to claim 1, characterized in that, The step of performing semantic matching processing between the semantic subgraph and the preset shared knowledge graph to determine the newly added nodes and edges of the shared knowledge graph includes: The entities and relations in the semantic subgraph and the shared knowledge graph are mapped to a unified vector space; The cosine similarity between each node and edge in the semantic subgraph and the most similar element in the shared knowledge graph is determined based on the cosine similarity function. If the cosine similarity is less than a preset node threshold, the node is determined to be a new node; if the cosine similarity is less than a preset edge threshold, the edge is determined to be a new edge.

5. The text semantic communication method according to claim 1, characterized in that, The process of encoding the incremental semantic data to generate estimated symbols includes: The incremental semantic data is mapped to continuous semantic vectors, and noise perturbation is introduced into the semantic vectors during the training phase, so that the encoder learns robust semantic features with noise resistance. The Transformer-based encoder performs context modeling on the semantic vector with introduced noise perturbation to determine the context feature vector, and maps the context feature vector to a low-dimensional channel symbol through a projection matrix. The channel symbols are modulated and input into the physical channel to obtain the observation signal at the receiver. The observation signal is then equalized to generate an estimated symbol.

6. The text semantic communication method according to claim 1, characterized in that, The step of performing denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph includes: The estimated symbols are processed to eliminate channel noise based on a residual denoising network layer consisting of a one-dimensional convolutional layer, a ReLU activation function, and residual connections, thereby determining the first feature; The estimated symbols are decoded using a multi-layer cross-attention decoder to determine the second feature; The first feature and the second feature are added together to recover the incremental subgraph.

7. The text semantic communication method according to claim 1, characterized in that, After reconstructing the target text based on the complete semantic subgraph, the text semantic communication method further includes: At the receiving end, new nodes or edges that appear more frequently than a set threshold in historical communications are identified. Under the condition of satisfying storage and latency constraints, the shared knowledge graph is updated based on the new nodes or edges that appear more frequently than a set threshold.

8. A text semantic communication device, characterized in that, The text semantic communication device includes: The semantic subgraph construction module is used to extract semantic information from the target text sent by the sending end based on the triple extraction model, and to construct a semantic subgraph of the target text based on the entity information and relation information of the extracted target text; wherein, the semantic subgraph is a directed graph structure of the target text; The incremental determination module is used to perform semantic matching processing between the semantic subgraph and the preset shared knowledge graph to determine the newly added nodes and newly added edges of the shared knowledge graph. The encoding module is used to form incremental semantic data to be transmitted from the incremental subgraph composed of the newly added node and the newly added edge, encode the incremental semantic data to generate estimated symbols, and transmit them to the receiving end through a physical channel. The reconstruction module is used to perform denoising and reconstruction processing on the estimated symbols at the receiving end to recover the incremental subgraph, fuse the recovered incremental subgraph with the shared knowledge graph to reconstruct a complete semantic subgraph, and restore the target text based on the complete semantic subgraph.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the text semantic communication method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the text semantic communication method as described in any one of claims 1 to 7.

Images