Knowledge graph file classification method and system based on intelligent AI

Abstract

The invention relates to the technical field of file management, and discloses a knowledge graph file classification method and system based on intelligent AI, and the method comprises the steps: extracting timestamps and geographic coordinates in files through a clock chip and a Beidou positioning module, recognizing attenuation factors of the files through the timestamps, and extracting metadata of the files; preparing a dual-channel GNN acceleration card of the file; inputting the metadata into a static topology processor to obtain an entity relationship output by the static topology processor; inputting the attenuation factor into an attenuation modulator to obtain a voltage signal output by the attenuation modulator; and according to the entity relationship, the geographic coordinates and the voltage signals, performing intelligent AI classification on the archives by using a feature fusion module and a GraphSAGE accelerator. The method can solve the problem that the classification accuracy of an existing system on long-term stored archives is obviously reduced along with the increase of the age limit of the archives.

Description

Technical Field

[0001] This invention relates to a knowledge graph-based archive classification method and system based on intelligent AI, belonging to the field of document management technology. Background Technology

[0002] With the deep application of artificial intelligence technology in the field of archival management, intelligent AI-driven knowledge graph archival classification methods have become a key means to improve the efficiency of archival value mining.

[0003] Currently, mainstream knowledge graph archival classification methods have significant limitations in handling the time-related decay characteristics of entity relationships. Existing methods rely on fixed entity relationship weights and do not consider the natural decay of archival value over time. For example, the timeliness of position-personnel relationships in personnel appointment and dismissal files from 10 years ago is much lower than that in recent files, but existing technologies still assign them the same weight. Although some studies have attempted to introduce time feature embedding, they only use timestamps as node feature inputs and fail to quantify the nonlinear decay law of entity relationship strength over time. This leads to a systematic decline in the classification accuracy of long-term archives. When the archive contains historical documents spanning decades, static models will incorrectly strengthen invalid entity associations while weakening effective associations.

[0004] Therefore, the accuracy of existing systems in classifying long-term archives shows a significant downward trend as the age of the archives increases. Summary of the Invention

[0005] This invention provides a knowledge graph-based archive classification method and system based on intelligent AI. Its main purpose is to solve the problem that the classification accuracy of existing systems for long-term archives shows a significant downward trend as the archives age.

[0006] To achieve the above objectives, this invention provides a knowledge graph-based archive classification method based on intelligent AI, comprising: The timestamp and geographic coordinates in the archive are extracted using a clock chip and a Beidou positioning module. The attenuation factor of the archive is identified using the timestamp, and the metadata of the archive is extracted. Prepare a dual-channel GNN accelerator card for the file, wherein the dual-channel GNN accelerator card includes a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator; The metadata is input into the static topology processor to obtain the entity relationships output by the static topology processor; The attenuation factor is input into the attenuation modulator to obtain the voltage signal output by the attenuation modulator; Based on the entity relationships, the geographic coordinates, and the voltage signal, the archives are intelligently classified using the feature fusion module and the GraphSAGE accelerator to obtain the archive classification results.

[0007] Optionally, the step of extracting the timestamp and geographic coordinates from the archive using a clock chip and a BeiDou positioning module includes: After connecting the input terminal of the clock chip and the input terminal of the Beidou positioning module to the document scanner, the clock chip receives the time entity information in the document through its input terminal. The clock chip is used to extract the timestamp from the time entity information; In the BeiDou positioning module, spatial entity information in the archive is received through the input terminal of the BeiDou positioning module; Geographic coordinates are extracted from the spatial entity information using the BeiDou positioning module.

[0008] Optionally, identifying the decay factor of the archive using the timestamp includes: After receiving the timestamp via the PCIe bus, the attenuation factor corresponding to the timestamp is calculated in the FPGA using the following formula, based on the industry coefficient stored in the EEPROM chip: ; in, Indicates the attenuation factor. Indicates the initial relation weights. This represents the difference between the current system time and the timestamp. Indicates industry coefficient. This represents the timestamp.

[0009] Optionally, the dual-channel GNN accelerator card for preparing the file includes: Integrate a static topology processor, attenuation modulator, feature fusion module and GraphSAGE accelerator in an FPGA; The attenuation modulator consists of an analog-to-digital converter and a digital-to-analog converter; the feature fusion module consists of an analog-to-digital converter, a multiplier IP core, a parallel bus, a Normalize IP core, and a block memory; and the GraphSAGE accelerator includes a GraphSAGE IP core and a ReLU hard core. Connect the input of the static topology processor to the output of the metadata extractor. The first input terminal of the feature fusion module is electrically connected to the output terminal of the static topology processor; The second input terminal of the feature fusion module is electrically connected to the output terminal of the attenuation modulator; Electrically connect the third input terminal of the feature fusion module to the output terminal of the Beidou positioning module; Electrically connect the input of the GraphSAGE accelerator to the output of the feature fusion module to determine the dual-channel GNN accelerator card for the archive.

[0010] Optionally, obtaining the entity relationships output by the static topology processor includes: In the static topology processor, a parallel PE array is used to perform sparse adjacency matrix multiplication on the entity-relation triples in the metadata to obtain the adjacency-feature product matrix. The adjacency-feature product matrix is ​​activated using ReLU to obtain the entity relations.

[0011] Optionally, inputting the attenuation factor into the attenuation modulator includes: The attenuation factor is quantized into digital code using an analog-to-digital converter; the analog-to-digital converter includes: ; in, Represents a numeric code. Indicates the attenuation factor. This represents the initial relational weights, and 32 represents the quantization bits of the ADC. Integer function; The digital code is input into the attenuation modulator.

[0012] Optionally, acquiring the voltage signal output by the attenuation modulator includes: The digital code is converted into a voltage signal using a digital-to-analog converter in the attenuation modulator; the digital-to-analog converter includes: ; in, Indicates voltage signal, This represents the numeric code; 32 indicates the quantization bits of the ADC. This represents the reference voltage.

[0013] Optionally, the step of performing intelligent AI classification on the archives based on the entity relationships, the geographic coordinates, and the voltage signal using the feature fusion module and the GraphSAGE accelerator to obtain archive classification results includes: The first, second, and third input terminals of the feature fusion module receive entity relationships, voltage signals, and geographic coordinates, respectively. Acquire the analog-to-digital converter, multiplier IP core, parallel bus, Normalize IP core, and block memory in the feature fusion module; The voltage signal is converted into digital weights using an analog-to-digital converter; The entity relationships are weighted using the digital weights and the multiplier IP core to obtain weighted entity features; The geographic coordinates and the weighted entity features are concatenated into a fused feature vector using a parallel bus; The fused feature vector is normalized using the Normalize IP kernel to obtain normalized features; The normalized features are written into the block memory; After reading the normalized features from the block memory using the GraphSAGE accelerator, the normalized features are then subjected to intelligent AI classification using the GraphSAGE accelerator to obtain the file classification result.

[0014] Optionally, the step of using the GraphSAGE accelerator to perform intelligent AI classification on the normalized features to obtain the archive classification result includes: Obtain the GraphSAGE IP core and ReLU hard core from the GraphSAGE accelerator; The normalized features are subjected to graph convolution operation using the GraphSAGE IP kernel to obtain updated features; The probability distribution corresponding to the updated feature is output using the ReLU hard kernel; The category of the archive corresponding to the highest probability in the probability distribution is taken as the archive classification result.

[0015] To address the aforementioned problems, the present invention also provides a knowledge graph-based archive classification system based on intelligent AI, the system comprising: The data extraction module is used to extract timestamps and geographic coordinates from the archives using a clock chip and a Beidou positioning module, identify the attenuation factor of the archives using the timestamps, and extract the metadata of the archives. The GNN preparation module is used to prepare the dual-channel GNN accelerator card for the file, wherein the dual-channel GNN accelerator card includes a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator. The relation output module is used to input the metadata into the static topology processor to obtain the entity relations output by the static topology processor; A voltage output module is used to input the attenuation factor into the attenuation modulator to obtain the voltage signal output by the attenuation modulator; The archive classification module is used to perform intelligent AI classification of the archives based on the entity relationships, the geographical coordinates, and the voltage signal, using the feature fusion module and the GraphSAGE accelerator to obtain the archive classification results.

[0016] Compared to the problems described in the background technology, the embodiments of the present invention precisely quantify the archive generation time through a clock chip, providing a time anchor for attenuation factor calculation. Geographic coordinates are bound to entity spatial attributes to avoid regional correlation distortion. Furthermore, the embodiments of the present invention dynamically quantify the attenuation of entity relationship strength through formulas. The embodiments of the present invention preserve the inherent structure of entities through a static topology processor, and the attenuation modulator converts the attenuation factor into a voltage signal, achieving hardware-level timeliness control. The dual-channel design allows the system to simultaneously process spatial topology and time attenuation, avoiding feature confusion in traditional single-channel GNNs. The embodiments of the present invention achieve irreversible conversion from timeliness to electrical signals through hardware circuitry. The embodiments of the present invention use a feature fusion module to hardware-level concatenate geographic coordinates, entity relationships, and voltage weights, injecting attenuation information into the normalized features, thereby solving the problem that the classification accuracy of existing systems for long-term preserved archives shows a significant downward trend with the increase of archive age. Therefore, the present invention can solve the problem that the classification accuracy of existing systems for long-term preserved archives shows a significant downward trend with the increase of archive age. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a knowledge graph-based archive classification method based on intelligent AI provided in an embodiment of the present invention; Figure 2 A schematic diagram of a dual-channel GNN acceleration card for a knowledge graph archive classification method based on intelligent AI provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the modules for implementing the AI-based knowledge graph archive classification system according to an embodiment of the present invention.

[0018] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] This application provides a knowledge graph archive classification method based on intelligent AI. The executing entity of the intelligent AI-based knowledge graph archive classification method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the intelligent AI-based knowledge graph archive classification method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0021] Reference Figure 1 The diagram shown is a flowchart illustrating a knowledge graph-based archive classification method based on intelligent AI, according to an embodiment of the present invention. In this embodiment, the knowledge graph-based archive classification method based on intelligent AI includes: S1. Use a clock chip and a Beidou positioning module to extract the timestamp and geographic coordinates from the file, use the timestamp to identify the attenuation factor of the file, and extract the metadata of the file.

[0022] This invention uses a clock chip to precisely quantify the archive generation time, providing a time anchor for attenuation factor calculation. Geographic coordinates are bound to entity spatial attributes to avoid regional correlation distortion.

[0023] In one embodiment of the present invention, the step of extracting timestamps and geographic coordinates from an archive using a clock chip and a BeiDou positioning module includes: connecting the input terminal of the clock chip and the input terminal of the BeiDou positioning module to an archive scanner respectively; receiving time entity information from the archive through the input terminal of the clock chip; extracting timestamps from the time entity information using the clock chip; receiving spatial entity information from the archive through the input terminal of the BeiDou positioning module; and extracting geographic coordinates from the spatial entity information using the BeiDou positioning module.

[0024] The clock chip refers to a high-precision real-time clock chip with an internal crystal oscillator and temperature compensation, used to identify and output the timestamp of the archive's creation time from the scanned data. The Beidou positioning module refers to a Beidou satellite navigation receiver module, used to parse and output the geographic coordinates of the entities involved in the archive from the scanned data. The archive scanner refers to a device that converts paper or electronic archives into digital signals and outputs a raw data stream containing time, space, and metadata information. The time entity information refers to the identifiable time description in the archive content, which is used by the clock chip to extract the timestamp. The space entity information refers to the identifiable location description in the archive content, which is used by the Beidou positioning module to extract the geographic coordinates. The timestamp refers to a digital time value in seconds or milliseconds, representing the exact moment the archive was created or the event occurred. The geographic coordinates refer to longitude and latitude values, representing the Earth's surface location of the place involved in the archive.

[0025] Furthermore, embodiments of the present invention dynamically quantify the attenuation of entity relationship strength through formulas.

[0026] In one embodiment of the present invention, the step of identifying the attenuation factor of the file using the timestamp includes: receiving the timestamp via the PCIe bus, and calculating the attenuation factor corresponding to the timestamp via the FPGA based on the industry coefficient stored in the EEPROM chip.

[0027] As another embodiment, the step of calculating the attenuation factor corresponding to the timestamp using an FPGA based on the industry coefficients stored in the EEPROM chip includes: calculating the attenuation factor corresponding to the timestamp in the FPGA using the following formula based on the industry coefficients stored in the EEPROM chip: ; in, Indicates the attenuation factor. Indicates the initial relation weights. This represents the difference between the current system time and the timestamp. Indicates industry coefficient. This represents the timestamp.

[0028] Among them, PCIe bus refers to high-speed serial computer expansion bus, used to quickly transmit the timestamp output by the clock chip to the FPGA; EEPROM chip refers to electrically erasable programmable read-only memory, which retains data even when power is off and is used to store industry coefficients; FPGA refers to field-programmable gate array chip, whose internal logic calculates the attenuation factor in real time. This refers to the baseline weight given to the archive at the time of its creation. It's a 100-point scale score assigned to the archive on the day it's entered into the archive. A higher score indicates a closer connection between the archive and a key person / event. This score is determined by pre-set levels, such as the type of archive, which includes classification levels. =0.6*100, Confidentiality level is =0.8*100, Top Secret =1*100, etc. At this point, the file type is only temporary, determined by scanning the file's keywords. Further, a GNN is needed for more detailed file classification. The calculation method is as follows: First, actual measurement , , Calculate the specific values, etc. Afterwards, take a large amount The median of the sample. It is the difference between the latest and earliest times in this large sample, including the median. Substituting into the following formula, we can obtain : ; Furthermore, to better understand the source of the industry coefficients in the above-mentioned knowledge graph archive classification method based on intelligent AI, please refer to Table 1 below, which is a table of industry coefficient values ​​for implementing the knowledge graph archive classification method based on intelligent AI provided in an embodiment of the present invention.

[0029] Table 1 ; As can be clearly seen from Table 1 above, in the knowledge graph archive classification method based on intelligent AI, the intensity decay of entity relationships for different archive types can be dynamically quantified.

[0030] In one embodiment of the present invention, extracting the metadata of the archive includes: extracting the metadata of the archive when the archive conforms to a preset archive standard metadata format.

[0031] The aforementioned archival standard metadata format refers to the electronic document archiving and electronic archive management standard, such as GB / T 18894-2016. For example, if the archive mentions that user A attended a meeting on August 2, 2025, the system will identify the following entity-relation triples: (user A, attended, meeting); (user A, date, August 2, 2025). Metadata mainly exists in the form of entity-relation triples.

[0032] S2. Prepare a dual-channel GNN accelerator card for the file, wherein the dual-channel GNN accelerator card includes a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator.

[0033] The embodiments of the present invention preserve the inherent structure of the entity through a static topology processor, and convert the attenuation modulator into a voltage signal to achieve hardware-level timeliness control. The dual-channel design enables the system to process spatial topology and time attenuation at the same time, avoiding feature confusion in traditional single-channel GNNs.

[0034] In one embodiment of the present invention, the dual-channel GNN accelerator card for preparing the archive includes: integrating a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator in an FPGA; wherein the attenuation modulator consists of an analog-to-digital converter and a digital-to-analog converter, the feature fusion module consists of an analog-to-digital converter, a multiplier IP core, a parallel bus, a Normalize IP core, and a block memory, and the GraphSAGE accelerator includes a GraphSAGE IP core and a ReLU hard core; electrically connecting the input terminal of the static topology processor to the output terminal of the metadata extractor; electrically connecting the first input terminal of the feature fusion module to the output terminal of the static topology processor; electrically connecting the second input terminal of the feature fusion module to the output terminal of the attenuation modulator; electrically connecting the third input terminal of the feature fusion module to the output terminal of the BeiDou positioning module; and electrically connecting the input terminal of the GraphSAGE accelerator to the output terminal of the feature fusion module, thereby determining the dual-channel GNN accelerator card for the archive.

[0035] It should be noted that there is only one analog-to-digital converter (ADC). The attenuation modulator and feature fusion module both need to call this ADC during operation. The metadata extractor refers to the component used to extract the metadata of the aforementioned file. The various components in the dual-channel GNN accelerator card are connected via a bus. The multiplier IP core refers to the hardware multiplication unit inside the FPGA, used to perform multiplication operations on two digital signals or data. The parallel bus refers to the data transmission channel, used to transmit data in parallel between multiple modules or devices. The normalize IP core refers to the normalization processing unit inside the FPGA, used to normalize the input data, such as standardization or batch normalization. The block memory refers to the block random access memory. The GraphSAGE IP core refers to the GraphSAGE algorithm acceleration unit implemented inside the FPGA, specifically used to perform graph convolution operations in graph neural networks, accelerating graph data processing and feature learning. The ReLU hard core refers to the ReLU activation function hardware unit implemented inside the FPGA, used to introduce nonlinearity.

[0036] See Figure 2 The diagram shown is a schematic of a dual-channel GNN accelerator card for a knowledge graph-based archive classification method based on intelligent AI, provided in an embodiment of the present invention. Figure 2 In this process, the output data of the static topology processor, attenuation modulator, and BeiDou positioning module are simultaneously input into the feature fusion module for feature fusion, and then classified by the GraphSAGE accelerator.

[0037] S3. Input the metadata into the static topology processor to obtain the entity relationships output by the static topology processor.

[0038] In one embodiment of the present invention, obtaining the entity relation output by the static topology processor includes: performing sparse adjacency matrix multiplication on the entity-relation triples in the metadata using a parallel PE array in the static topology processor to obtain an adjacency-feature product matrix; and performing ReLU activation processing on the adjacency-feature product matrix to obtain the entity relation.

[0039] The static topology processor refers to a hardware module specifically designed for processing graph data, capable of performing static analysis and processing tasks of graph structures, such as calculating the topology of a graph. The parallel PE array refers to an array composed of multiple parallel processing units, each of which can independently execute computational tasks and jointly complete large-scale parallel processing. The sparse adjacency matrix multiplication operation is used to update the node features in a graph neural network. Considering the sparsity of the graph, only non-zero elements are calculated to improve computational efficiency. The adjacency-feature product matrix can be obtained through the multiplication operation of the adjacency matrix and the feature matrix. The adjacency matrix represents the relationship between each entity; if a relationship exists, it is 1, otherwise it is 0. The elements in the feature matrix are the feature vectors of each entity.

[0040] S4. Input the attenuation factor into the attenuation modulator to obtain the voltage signal output by the attenuation modulator.

[0041] The embodiments of the present invention realize the irreversible conversion from time-sensitive signals to electrical signals through hardware circuits.

[0042] In one embodiment of the present invention, inputting the attenuation factor into the attenuation modulator includes: quantizing the attenuation factor into a digital code using an analog-to-digital converter; and inputting the digital code into the attenuation modulator.

[0043] As another embodiment, quantizing the attenuation factor into digital code using an analog-to-digital converter includes: the analog-to-digital converter comprising: ; in, Represents a numeric code. Indicates the attenuation factor. This represents the initial relational weights, and 32 represents the quantization bits of the ADC. Integer function.

[0044] In one embodiment of the present invention, obtaining the voltage signal output by the attenuation modulator includes: converting digital code into a voltage signal using a digital-to-analog converter in the attenuation modulator.

[0045] As another embodiment, the conversion of digital code into a voltage signal using a digital-to-analog converter in the attenuation modulator includes: the digital-to-analog converter comprising: converting digital code into a voltage signal using the following formula: ; in, Indicates voltage signal, This represents the numeric code; 32 indicates the quantization bits of the ADC. This represents the reference voltage.

[0046] S5. Based on the entity relationship, the geographic coordinates, and the voltage signal, the archive is intelligently classified using the feature fusion module and the GraphSAGE accelerator to obtain the archive classification result.

[0047] This invention addresses the problem that the classification accuracy of existing systems for long-term archives shows a significant downward trend as the archives age, by using a feature fusion module to hardware-level splice geographic coordinates, entity relationships, and voltage weights, and injecting attenuation information into the normalized features.

[0048] In one embodiment of the present invention, the step of performing intelligent AI classification on the archives based on the entity relationships, the geographic coordinates, and the voltage signal using the feature fusion module and the GraphSAGE accelerator to obtain archive classification results includes: receiving entity relationships, voltage signals, and geographic coordinates through the first, second, and third input terminals of the feature fusion module; acquiring the analog-to-digital converter, multiplier IP core, parallel bus, Normalize IP core, and block memory in the feature fusion module; converting the voltage signal into digital weights using the analog-to-digital converter; weighting the entity relationships using the digital weights and the multiplier IP core to obtain weighted entity features; concatenating the geographic coordinates and the weighted entity features using the parallel bus to form a fused feature vector; normalizing the fused feature vector using the Normalize IP core to obtain normalized features; writing the normalized features into the block memory; and reading the normalized features from the block memory using the GraphSAGE accelerator and then performing intelligent AI classification on the normalized features using the GraphSAGE accelerator to obtain archive classification results.

[0049] In another embodiment of the present invention, the step of using the GraphSAGE accelerator to perform intelligent AI classification on the normalized features to obtain the archive classification result includes: obtaining the GraphSAGE IP core and ReLU hard core in the GraphSAGE accelerator; performing graph convolution operation on the normalized features using the GraphSAGE IP core to obtain updated features; outputting the probability distribution corresponding to the updated features using the ReLU hard core; and taking the archive category corresponding to the highest probability in the probability distribution as the archive classification result.

[0050] The categories of archives include, but are not limited to: confidential archives, classified archives, top secret archives, scientific research archives, administrative archives, financial archives, personnel archives, project archives, legal archives, and educational archives.

[0051] Optionally, the process of using the GraphSAGE IP kernel to perform graph convolution operation on the normalized features to obtain updated features refers to: the graph convolution operation uses the weighted entity features output by the feature fusion module and the normalized features in the block memory to perform mean aggregation within the GraphSAGE IP kernel. The aggregation operation involves calculating the average value of the neighbor features of each node, and the aggregation result is used as the updated feature output.

[0052] Compared to the problems described in the background technology, the embodiments of the present invention precisely quantify the archive generation time through a clock chip, providing a time anchor for attenuation factor calculation. Geographic coordinates are bound to entity spatial attributes to avoid regional correlation distortion. Furthermore, the embodiments of the present invention dynamically quantify the attenuation of entity relationship strength through formulas. The embodiments of the present invention preserve the inherent structure of entities through a static topology processor, and the attenuation modulator converts the attenuation factor into a voltage signal, achieving hardware-level timeliness control. The dual-channel design allows the system to simultaneously process spatial topology and time attenuation, avoiding feature confusion in traditional single-channel GNNs. The embodiments of the present invention achieve irreversible conversion from timeliness to electrical signals through hardware circuitry. The embodiments of the present invention use a feature fusion module to hardware-level concatenate geographic coordinates, entity relationships, and voltage weights, injecting attenuation information into the normalized features, thereby solving the problem that the classification accuracy of existing systems for long-term preserved archives shows a significant downward trend with the increase of archive age. Therefore, the present invention can solve the problem that the classification accuracy of existing systems for long-term preserved archives shows a significant downward trend with the increase of archive age.

[0053] like Figure 3 The diagram shown is a functional module diagram of a knowledge graph archive classification system based on intelligent AI according to the present invention.

[0054] The AI-based knowledge graph archive classification system 300 described in this invention can be installed in an electronic device. Depending on the functions implemented, the AI-based knowledge graph archive classification system may include a data extraction module 301, a GNN preparation module 302, a relation output module 303, a voltage output module 304, and an archive classification module 305. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0055] In this embodiment of the invention, the functions of each module / unit are as follows: The data extraction module 301 is used to extract timestamps and geographic coordinates from the archive using a clock chip and a Beidou positioning module, identify the attenuation factor of the archive using the timestamps, and extract the metadata of the archive. The GNN preparation module 302 is used to prepare the dual-channel GNN acceleration card for the file, wherein the dual-channel GNN acceleration card includes a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator. The relationship output module 303 is used to input the metadata into the static topology processor to obtain the entity relationship output by the static topology processor; The voltage output module 304 is used to input the attenuation factor into the attenuation modulator to obtain the voltage signal output by the attenuation modulator; The archive classification module 305 is used to perform intelligent AI classification of the archives based on the entity relationships, the geographical coordinates, and the voltage signal, using the feature fusion module and the GraphSAGE accelerator, to obtain the archive classification result.

[0056] In detail, the modules in the AI-based knowledge graph archive classification system 300 described in this embodiment of the invention employ the same methods as described above. Figure 1 The method uses the same technical means as the knowledge graph archive classification method based on intelligent AI described in the article, and can produce the same technical effect, so it will not be elaborated here.

[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0058] Finally, it should be noted that in the above embodiments, each embodiment can be combined with each other or independent. Deleting any one of them will not affect the technical implementation of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A knowledge graph-based archive classification method based on intelligent AI, characterized in that, The method includes: The timestamp and geographic coordinates in the archive are extracted using a clock chip and a Beidou positioning module. The attenuation factor of the archive is identified using the timestamp, and the metadata of the archive is extracted. Prepare a dual-channel GNN accelerator card for the file, wherein the dual-channel GNN accelerator card includes a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator; The metadata is input into the static topology processor to obtain the entity relationships output by the static topology processor; The attenuation factor is input into the attenuation modulator to obtain the voltage signal output by the attenuation modulator; Based on the entity relationships, the geographic coordinates, and the voltage signal, the archives are intelligently classified using the feature fusion module and the GraphSAGE accelerator to obtain the archive classification results.

2. The knowledge graph-based archive classification method based on intelligent AI as described in claim 1, characterized in that, The process of extracting timestamps and geographic coordinates from files using a clock chip and a BeiDou positioning module includes: After connecting the input terminal of the clock chip and the input terminal of the Beidou positioning module to the document scanner, the clock chip receives the time entity information in the document through its input terminal. The clock chip is used to extract the timestamp from the time entity information; In the BeiDou positioning module, spatial entity information in the archive is received through the input terminal of the BeiDou positioning module; Geographic coordinates are extracted from the spatial entity information using the BeiDou positioning module.

3. The knowledge graph-based archive classification method based on intelligent AI as described in claim 1, characterized in that, The step of identifying the decay factor of the archive using the timestamp includes: After receiving the timestamp via the PCIe bus, the attenuation factor corresponding to the timestamp is calculated in the FPGA using the following formula, based on the industry coefficient stored in the EEPROM chip: ； in, Indicates the attenuation factor. Indicates the initial relation weights. This represents the difference between the current system time and the timestamp. Indicates industry coefficient. This represents the timestamp.

4. The knowledge graph-based archive classification method based on intelligent AI as described in claim 1, characterized in that, The dual-channel GNN accelerator card for preparing the file includes: Integrate a static topology processor, attenuation modulator, feature fusion module and GraphSAGE accelerator in an FPGA; The attenuation modulator consists of an analog-to-digital converter and a digital-to-analog converter; the feature fusion module consists of an analog-to-digital converter, a multiplier IP core, a parallel bus, a Normalize IP core, and a block memory; and the GraphSAGE accelerator includes a GraphSAGE IP core and a ReLU hard core. Connect the input of the static topology processor to the output of the metadata extractor. The first input terminal of the feature fusion module is electrically connected to the output terminal of the static topology processor; The second input terminal of the feature fusion module is electrically connected to the output terminal of the attenuation modulator; Electrically connect the third input terminal of the feature fusion module to the output terminal of the Beidou positioning module; Electrically connect the input of the GraphSAGE accelerator to the output of the feature fusion module to determine the dual-channel GNN accelerator card for the archive.

5. The knowledge graph archive classification method based on intelligent AI as described in claim 1, characterized in that, The step of obtaining the entity relationships output by the static topology processor includes: In the static topology processor, a parallel PE array is used to perform sparse adjacency matrix multiplication on the entity-relation triples in the metadata to obtain the adjacency-feature product matrix. The adjacency-feature product matrix is ​​activated using ReLU to obtain the entity relations.

6. The knowledge graph-based archive classification method based on intelligent AI as described in claim 1, characterized in that, The step of inputting the attenuation factor into the attenuation modulator includes: The attenuation factor is quantized into digital code using an analog-to-digital converter; the analog-to-digital converter includes: quantizing the attenuation factor into digital code using the following formula: ； in, Represents a numeric code. Indicates the attenuation factor. This represents the initial relational weights, and 32 represents the quantization bits of the ADC. Integer function; The digital code is input into the attenuation modulator.

7. The knowledge graph archive classification method based on intelligent AI as described in claim 1, characterized in that, The step of acquiring the voltage signal output by the attenuation modulator includes: The digital code is converted into a voltage signal using a digital-to-analog converter in the attenuation modulator; the digital-to-analog converter includes converting the digital code into a voltage signal using the following formula: ； in, Indicates voltage signal, This represents the numeric code; 32 indicates the quantization bits of the ADC. This represents the reference voltage.

8. The knowledge graph-based archive classification method based on intelligent AI as described in claim 1, characterized in that, The process involves using the feature fusion module and the GraphSAGE accelerator to perform intelligent AI classification on the archives based on the entity relationships, geographical coordinates, and voltage signals, to obtain archive classification results, including: The first, second, and third input terminals of the feature fusion module receive entity relationships, voltage signals, and geographic coordinates, respectively. Acquire the analog-to-digital converter, multiplier IP core, parallel bus, Normalize IP core, and block memory in the feature fusion module; The voltage signal is converted into digital weights using an analog-to-digital converter; The entity relationships are weighted using the digital weights and the multiplier IP core to obtain weighted entity features; The geographic coordinates and the weighted entity features are concatenated into a fused feature vector using a parallel bus; The fused feature vector is normalized using the Normalize IP kernel to obtain normalized features; The normalized features are written into the block memory; After reading the normalized features from the block memory using the GraphSAGE accelerator, the normalized features are then subjected to intelligent AI classification using the GraphSAGE accelerator to obtain the file classification result.

9. The knowledge graph-based archive classification method based on intelligent AI as described in claim 8, characterized in that, The process of using the GraphSAGE accelerator to perform intelligent AI classification on the normalized features to obtain the archive classification results includes: Obtain the GraphSAGE IP core and ReLU hard core from the GraphSAGE accelerator; The normalized features are subjected to graph convolution operation using the GraphSAGE IP kernel to obtain updated features; The probability distribution corresponding to the updated feature is output using the ReLU hard kernel; The category of the archive corresponding to the highest probability in the probability distribution is taken as the archive classification result.

10. A knowledge graph-based archive classification system based on intelligent AI, characterized in that, The system includes: The data extraction module is used to extract timestamps and geographic coordinates from the archives using a clock chip and a Beidou positioning module, identify the attenuation factor of the archives using the timestamps, and extract the metadata of the archives. The GNN preparation module is used to prepare the dual-channel GNN accelerator card for the file, wherein the dual-channel GNN accelerator card includes a static topology processor, an attenuation modulator, a feature fusion module, and a GraphSAGE accelerator. The relation output module is used to input the metadata into the static topology processor to obtain the entity relations output by the static topology processor; A voltage output module is used to input the attenuation factor into the attenuation modulator to obtain the voltage signal output by the attenuation modulator; The archive classification module is used to perform intelligent AI classification of the archives based on the entity relationships, the geographical coordinates, and the voltage signal, using the feature fusion module and the GraphSAGE accelerator to obtain the archive classification results.

Images