Artificial intelligence-based file flow method

Abstract

The application discloses a file circulation method based on artificial intelligence and relates to the technical field of artificial intelligence, which comprises the following steps: a circulation server receives a semantic abstract metadata package, performs hash operation on file basic attributes and semantic abstract in the semantic abstract metadata package, and generates an intelligent document identifier; a natural language processing model is called to analyze the semantic abstract in the semantic abstract metadata package and extract core semantic tags; routing rules in a business rule library are matched according to the core semantic tags, routing instructions are generated, the intelligent document identifier and the routing instructions are bound, and the bound result is returned to an uploading client; the validity of the intelligent document identifier is verified, a target file storage location is searched, the target file is asynchronously pulled to an adjacent processing node according to the routing instructions, and the state of the target file is updated to distributed. Through calling a generative artificial intelligence model in the uploading client to generate a semantic abstract and integrate file basic attributes, unified expression of file content and structured attributes is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a file transfer method based on artificial intelligence. Background Technology

[0002] With the continuous development of information technology and artificial intelligence, the need for file transfer and processing across different application environments is becoming increasingly frequent. In large-scale distributed architectures and multi-service collaborative scenarios, files not only need to remain intact and secure during transmission, but also need to be quickly understood and classified to support automated processing operations. In recent years, generative artificial intelligence models and natural language processing methods have been gradually applied to the field of file management, making it possible to extract semantics and generate summaries from file content. By embedding semantic parsing capabilities during file transfer, structured semantic information can be added to files, thereby supporting intelligent retrieval, distribution, and scheduling. This trend is driving the evolution of traditional file transfer methods that rely on static rules towards semantically driven intelligent workflows, and laying the foundation for the automated processing of complex business activities.

[0003] Existing file transfer methods, in scenarios involving cross-node and cross-application environments, often emphasize file path and rule matching, while offering relatively limited utilization at the semantic level. This means that under multi-business collaboration requirements, file identification and distribution still rely on manual configuration or fixed rules, making it difficult to reflect the semantic value inherent in the file content. Especially in scenarios requiring automatic generation of transfer instructions based on file semantics, the lack of efficient semantic extraction and dynamic identifier generation methods limits automation and accuracy. To address this issue, existing methods typically improve upon this by adding manual annotations or expanding fixed rule tables. However, with the rapid increase in the number of files and the rise in semantic complexity, these methods suffer from insufficient flexibility and high expansion costs. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based file transfer method to solve the problem of lacking efficient semantic extraction and dynamic identifier generation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a file transfer method based on artificial intelligence, which includes: The uploading client moves the target file to the local HTTP server storage path, calls the generative artificial intelligence model to parse the target file, and generates a semantic summary; integrates the basic file attributes and semantic summary of the target file to form a semantic summary meta data package; The streaming server receives the semantic summary metadata data package, performs hash operations on the basic file attributes and semantic summary in the semantic summary metadata data package to generate a smart document identifier; calls the natural language processing model to parse the semantic summary in the semantic summary metadata data package and extract the core semantic tags; matches the routing rules in the business rule base according to the core semantic tags to generate routing instructions, and returns the smart document identifier and routing instructions to the upload client; The uploading client transmits the smart document identifier to the adjacent processing node. The adjacent processing node extracts the core semantic tag from the smart document identifier. If the core semantic tag matches the preset node processing rule, it is determined that the adjacent processing node needs to process the target file and initiates a download request carrying the smart document identifier to the streaming server. The distribution server receives the download request, verifies the validity of the smart document identifier, retrieves the target file storage location, asynchronously pulls the target file to the adjacent processing node according to the routing instructions, and updates the target file status to "distributed".

[0007] As a preferred embodiment of the AI-based file transfer method of the present invention, the generation of semantic summaries specifically includes: The upload client receives the target file submitted by the user, obtains the storage path and file format information of the target file, moves the target file to the pending folder specified by the local HTTP server, reads the binary data stream of the target file, determines the target file type based on the target file format information, and extracts the text content of the target file. The extracted text content is segmented into fixed lengths to form structured input text; By calling the generative AI model deployed on the client side through the local interface, the structured input text is input into the generative AI model, the structured input text is analyzed, and the summary text is output. The uploading client receives the summary text, verifies the integrity of the summary text, and uses the complete summary text as the semantic summary.

[0008] As a preferred embodiment of the AI-based file transfer method of the present invention, the generative AI model analyzes the structured input text and outputs summary text. This means receiving the structured input text, processing it layer by layer in the encoder layer to extract semantic features between word segments, assigning attention weights to each word segment through a self-attention mechanism to enhance the contextual association in the structured input text, and outputting feature representations; the decoder layer receives the feature representations, predicts the next word segment word by word, and gradually generates summary text.

[0009] As a preferred embodiment of the AI-based file transfer method of the present invention, the generation of intelligent document identifiers specifically includes: The streaming server receives semantic digest metadata data packets sent by the uploading client; The semantic summary metadata package is deserialized to restore the structured key-value pair set, and the file name, file type, file size, check value and semantic summary are extracted from the structured key-value pair set; The fields are concatenated in the order of filename, file type, file size, checksum, and semantic digest to form the original string; The original string is hashed using a secure hash algorithm to obtain a fixed-length hash value, which serves as the smart document identifier.

[0010] As a preferred embodiment of the AI-based file transfer method of the present invention, the step of calling a natural language processing model to parse the semantic summary in the semantic summary metadata data packet and extracting the core semantic tags specifically involves: Load the pre-trained Chinese text classification model, input the semantic summary into the Chinese text classification model, perform forward inference, and obtain the classification probability distribution; The category label with the highest classification probability value is selected from the classification probability distribution as the core semantic label.

[0011] As a preferred embodiment of the AI-based file transfer method described in this invention, the forward inference is a Chinese text classification model that processes the semantic summary of the input based on the BERT model architecture. In the encoder layer, the semantic summary is processed layer by layer to extract semantic features between Chinese word segments. The encoder layer uses a self-attention mechanism to assign attention weights to each Chinese word segment, enhances contextual association, and outputs feature representations. An additional classification layer receives the feature representations, uses the feature representations as input vectors, and passes them to a fully connected layer. The fully connected layer processes the feature representations and applies the softmax function to generate a category probability distribution.

[0012] As a preferred embodiment of the file transfer method based on artificial intelligence described in this invention, the generation of routing instructions specifically includes: The core semantic tags are compared one by one with the rule semantic tags of each rule entry in the business rule base; If the core semantic label is exactly the same as the rule semantic label of rule entry A, then the routing rule of rule entry A will be used as the routing instruction.

[0013] As a preferred embodiment of the file transfer method based on artificial intelligence described in this invention, the business rule base is a set of key-value pairs set based on the routing requirements of contract document management, used to store rule entries. Each rule entry consists of a rule semantic tag and a routing rule, determining the transmission path of the target file.

[0014] In a preferred embodiment of the AI-based file transfer method described in this invention, the transfer server receives a download request and verifies the validity of the intelligent document identifier, specifically: The transfer server receives download requests carrying smart document identifiers initiated by adjacent processing nodes; Extract the smart document identifier from the download request, retrieve the locally maintained hash value record, and if the hash value record contains the extracted smart document identifier, then the smart document identifier is considered valid.

[0015] As a preferred embodiment of the AI-based file transfer method of the present invention, the steps of retrieving the target file storage location, asynchronously fetching the target file to adjacent processing nodes according to routing instructions, and updating the target file status to "distributed" specifically involve: The access path of the target file in the locally maintained metadata package is retrieved by intelligent document identification, and used as the storage location of the target file. Based on the address of the adjacent processing node specified in the routing instruction, an asynchronous transmission task is initiated to pull the target file from the storage location to the adjacent processing node; After the asynchronous transmission task starts, the streaming server updates the status field of the target file to "distributed" in the locally maintained metadata.

[0016] The beneficial effects of this invention are as follows: By calling a generative artificial intelligence model on the upload client to generate semantic summaries and integrating basic file attributes, a unified expression of file content and structured attributes is achieved. This semantic summarization generation and attribute integration method ensures that the file always contains complete semantic information during transmission, enabling subsequent processing nodes to accurately identify and match based on semantic tags, avoiding reliance on manual configuration or fixed rules. Through semantic parsing and unified attribute expression technology, files possess interpretability and searchability during cross-node transmission, thereby improving the automation and flexibility of file flow and achieving efficient distribution and accurate processing of files across different processing nodes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an AI-based file transfer method.

[0019] Figure 2 A flowchart for generating intelligent document identifiers and routing instructions for the streaming server.

[0020] Figure 3 This is a flowchart for transmitting identifiers and initiating requests.

[0021] Figure 4 A flowchart for server-side verification and asynchronous file retrieval. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a file transfer method based on artificial intelligence, including the following steps: S1. The uploading client moves the target file to the local HTTP server storage path, calls the generative artificial intelligence model to parse the target file, and generates a semantic summary; integrates the basic file attributes and semantic summary of the target file to form a semantic summary meta data package.

[0026] The upload client receives the target file submitted by the user, obtains the storage path and file format information of the target file, moves the target file to the pending folder specified by the local HTTP server, reads the binary data stream of the target file, determines the target file type based on the target file format information, and extracts the text content of the target file. Specifically: The upload client compares the file extension in the target file format information with a predefined list of file types. If the file extension is .pdf, the target file type is determined to be PDF; if it's .docx, it's determined to be Word; and if it's .txt, it's determined to be plain text. If the target file type is PDF, the upload client calls a PDF parsing library to read the page structure and embedded font information from the binary data stream, traversing each page node to identify text block positions and encoding methods. Character sequences are extracted from each page node and merged into a continuous string as the text content. If the target file type is Word, the upload client calls a Word document parsing library to load paragraph and table elements from the binary data stream, traversing each paragraph node to obtain character content and formatting marks. Character sequences are collected from each paragraph node, formatting marks are removed, and then merged into a continuous string as the text content. If the target file type is plain text, the upload client directly decodes the character sequences from the binary data stream, converts them byte-by-byte using the UTF-8 encoding standard, and merges them into a continuous string as the text content. The predefined list of file types is based on the standard definition of file extensions. The extracted text content is segmented into fixed length segments (e.g., 512 characters) to form structured input text; The structured input text is input into the generative AI model that has been deployed on the client side through a local interface call; the generative AI model is based on the Transformer model architecture. The training process for generative artificial intelligence models is as follows: We collected a large amount of text corpus as the training dataset. The text corpus included diverse sources such as books, articles, and online documents to ensure coverage of multiple languages ​​and domain knowledge. The text corpus in the training dataset is preprocessed by scanning the text word by word to identify sentence boundaries and word segments. Word sequences are separated from each sentence boundary and converted into a standard encoding format to form a unified input sequence. In the pre-training phase, the Transformer model architecture receives the unified input sequence and passes it layer by layer in the encoder layer to capture inter-word relationships. The encoder layer assigns attention weights to each word segment in the unified input sequence using a self-attention mechanism. These attention weights are generated based on the semantic similarity between word segments and are used to enhance the contextual associations between word segments in the unified input sequence. The encoder layer outputs a feature representation containing contextual relationships. The decoder layer receives the feature representation output from the encoder layer and processes it word by word to predict the next word segment, progressively generating a prediction sequence. During training, some word segments in the unified input sequence are hidden using a masking method. The decoder layer predicts the hidden word segments based on the feature representation output from the encoder layer to learn language patterns. After pre-training, a domain fine-tuning phase is performed. A domain-specific dataset, such as contract document summary samples, is prepared. The input text in the domain-specific dataset is aligned with the corresponding summary to form paired samples. After loading the pre-trained parameters, the Transformer model architecture receives the input text from the paired samples. The encoder layer processes the input text to extract feature representations, and the decoder layer receives the feature representations output by the encoder layer. It processes the feature representations word by word to predict the next word segment, gradually generating the output summary. By comparing the generated summary with the real summary in the paired samples word by word, the Transformer model architecture parameters are adjusted to improve summary accuracy. Finally, the generation performance of the Transformer model architecture on the test dataset is verified to ensure that the output summary text contains core business elements.

[0027] Generative AI models analyze structured input text and output a natural language summary containing core business elements. Specifically, the generative AI model, based on the Transformer model architecture, receives structured input text. In the encoder layer, the structured input text is processed layer by layer to extract semantic features between word segments. The encoder layer assigns attention weights to each word segment through a self-attention mechanism to enhance the contextual relationships in the structured input text and outputs a feature representation containing semantic features. The decoder layer receives the feature representation, predicts the next word segment word by word, and gradually generates a natural language summary containing core business elements. The upload client receives the summary text output by the generative AI model, verifies the integrity of the summary text, and uses the complete summary text as the semantic summary. Specifically: the upload client receives the summary text output by the generative AI model, checks whether the summary text contains a non-empty character sequence to ensure that the summary text is not empty; scans the summary text character by character to verify whether the character sequence conforms to the UTF-8 encoding standard, eliminating garbled characters or missing characters; confirms whether the character length of the summary text is within the predefined summary character length range (based on the typical text length of the target file and the output constraint setting of the generative AI model, such as 10 to 512 characters). If the summary text satisfies the non-empty character sequence, conforms to the UTF-8 encoding standard, and the character length is within the predefined summary character length range, then the summary text is considered complete and is used as the semantic summary; if the summary text is empty, contains garbled characters, has missing characters, or the character length is not within the predefined summary character length range, then the summary text is considered incomplete, and the upload client re-calls the generative AI model to generate a new summary text until the above conditions are met (i.e., the summary text satisfies the non-empty character sequence, conforms to the UTF-8 encoding standard, and the character length is within the predefined summary character length range). Get the filename, file type, file size, and checksum of the target file, and use the filename, file type, file size, and checksum as the basic attributes of the file; Arrange the basic attributes and semantic summary of the file in field order to generate a set of structured key-value pairs; The structured key-value pair set is serialized to generate a continuous byte stream data. Specifically, the uploading client reads the file name, file type, file size, check value, and semantic digest in the structured key-value pair set one by one in the field order, converts the key and value of each field into a JSON format string, concatenates the JSON format strings field by field to form a unified string, and converts the unified string into a continuous byte stream data through UTF-8 encoding. Continuous byte stream data is encapsulated into a unified data format object to form a semantic summary metadata package.

[0028] S2. The streaming server receives the semantic digest metadata data packet, performs hash operations on the basic file attributes and semantic digest in the semantic digest metadata data packet, and generates a smart document identifier; it calls the natural language processing model to parse the semantic digest in the semantic digest metadata data packet and extract the core semantic tags; it matches the routing rules in the business rule base according to the core semantic tags, generates routing instructions, and returns the smart document identifier and routing instructions to the upload client.

[0029] The streaming server receives semantic digest metadata data packets sent by the uploading client; The semantic summary metadata package is deserialized to restore the structured key-value pair set, and the file name, file type, file size, check value and semantic summary are extracted from the structured key-value pair set; The fields are concatenated in the order of filename, file type, file size, checksum, and semantic digest to form the original string; The original string is hashed using a secure hash algorithm to obtain a fixed-length hash value as the smart document identifier. The secure hash algorithm is the SHA-256 algorithm, which belongs to the SHA-2 family of cryptographic hash functions. It is used to convert input data of any length into a fixed-length 256-bit hash value to ensure data integrity and uniqueness. Even if the input data changes slightly, the output hash value will be completely different. The hash operation process is as follows: Starting from preprocessing, the original string is padded to a length conforming to the 512-bit block standard, and a message block is formed by adding padding bits and length information; the hash buffer is initialized using eight predefined 32-bit initial values ​​(based on the SHA-256 algorithm standard definition) as the starting state; each message block is processed, expanding the message block into an extended block of 64 32-bit words, and the working variable and working hash value are updated round by round through 64 rounds of compression operations. Each round of operations involves bit shifting, logical function combination, and expansion of constant extended blocks; after each round, the working hash value is merged with the initial buffer state to form the final 256-bit fixed-length hash value as the smart document identifier.

[0030] Load a pre-trained Chinese text classification model, input semantic summary into the Chinese text classification model, perform forward inference to obtain the classification probability distribution; the Chinese text classification model is an existing model based on the BERT model architecture; The training process for the Chinese text classification model is as follows: A large amount of Chinese text corpus was collected as the training dataset. The Chinese text corpus includes diverse sources such as Chinese books, news articles and online forum content to ensure coverage of multiple fields and language expressions. The Chinese text corpus in the training dataset is preprocessed by scanning the Chinese text corpus character by character to identify sentence boundaries and Chinese word segments. Chinese word sequences are separated from each sentence and converted into a standard encoding format to form a unified input sequence. During the pre-training phase, the BERT model architecture receives a unified input sequence and processes it layer by layer in the encoder layer to capture the semantic relationships between Chinese word segments. The encoder layer uses a self-attention mechanism to assign attention weights to each Chinese word segment in the unified input sequence. The attention weights are generated based on the semantic similarity between Chinese word segments and are used to enhance the contextual associations in the unified input sequence. The encoder layer outputs a feature representation containing contextual relationships. During training, some Chinese word segments in the unified input sequence are hidden by masking. The BERT model architecture predicts the hidden Chinese word segments based on the feature representation output by the encoder layer in order to learn Chinese language patterns. After pre-training, a fine-tuning phase for the classification task is performed. Chinese classification datasets, such as contract document management datasets, are collected from specific domains, including contract document label samples. The input text in these datasets is aligned with the corresponding category labels to form paired samples. After loading the pre-trained parameters, the BERT model architecture receives the input text from the paired samples. The encoder layer processes the input text to extract feature representations. An additional classification layer receives the feature representations output from the encoder layer and passes them as input vectors to a fully connected layer. The fully connected layer processes the feature representations and applies a softmax function to generate a category probability distribution. By comparing the generated category probability distribution with the true category labels in the paired samples, the BERT model architecture parameters are adjusted to improve classification accuracy. The classification performance of the BERT model architecture on the test dataset is verified to ensure that the generated classification probability distribution accurately reflects the core semantic labels. The category label with the highest classification probability value is selected from the classification probability distribution as the core semantic label.

[0031] The core semantic tags are compared one by one with the semantic tags of each rule entry in the business rule base. The business rule base stores multiple rule entries, and each rule entry consists of a semantic tag and a routing rule in sequence (the semantic tag comes first, followed by the routing rule). The process of constructing the business rule base is as follows: collect requirements from specific domains, such as routing requirements in contract document management; define rule semantic tags based on routing requirements, such as "confidentiality clause" and "financial related"; associate a routing rule with each rule semantic tag, for example, "confidentiality clause" corresponds to "legal department node"; assemble rule semantic tags and routing rules into rule entries in sequence; and store all rule entries as a set of key-value pairs to form the business rule base. The core semantic tag is compared one by one with the rule semantic tags of each rule entry in the business rule base. Specifically, the routing server reads each rule entry from the business rule base one by one and extracts the rule semantic tag from the rule entry. The core semantic tag is compared with the rule semantic tag in the current rule entry to check if they are completely identical. If the core semantic tag is completely identical with the rule semantic tag in the current rule entry, the routing rule in that rule entry is extracted. If they are not identical, the next rule entry is read and the comparison is repeated until a completely identical rule semantic tag is found or all rule entries are traversed. If no rule semantic tag completely identical to the core semantic tag is found after traversing all rule entries, the routing server sets the routing rule as the default routing rule, such as the default storage node. If the core semantic tag is exactly the same as the rule semantic tag of rule entry A, then the routing rule of rule entry A is used as the routing instruction. For example, if the core semantic tag is "confidentiality clause", the routing server reads rule entry A from the business rule library. The rule semantic tag of rule entry A is "confidentiality clause", and the corresponding routing rule is "legal department node". The routing server compares the core semantic tag "confidentiality clause" with the rule semantic tag "confidentiality clause" of rule entry A. If they are found to be exactly the same, then the routing rule "legal department node" of rule entry A is used as the routing instruction. It should be noted that the fact that the core semantic tag is the same as the rule semantic tag indicates that the semantic content of the target file matches the semantic requirements of rule item A in the business rule base. This means that the target file needs to be processed according to the routing rules of rule item A. Therefore, the routing server uses the routing rules of rule item A as routing instructions to ensure that the target file is correctly distributed to the specified node. For example, "confidentiality clause" matches "legal department node" to meet the business needs of contract document management. Bind smart document identifiers and routing instructions, and return them to the upload client.

[0032] S3. The uploading client transmits the smart document identifier to the adjacent processing node. The adjacent processing node extracts the core semantic tag from the smart document identifier. If the core semantic tag matches the preset node processing rule, it is determined that the adjacent processing node needs to process the target file and initiates a download request carrying the smart document identifier to the streaming server.

[0033] The uploading client transmits the smart document identifier to the adjacent processing node; Adjacent processing nodes receive intelligent document identifiers and extract core semantic tags from them. The core semantic tag is compared one by one with the node semantic tags in the preset node processing rules. Specifically, adjacent processing nodes read each node processing rule one by one from the preset node processing rules and extract the node semantic tags in the node processing rules. The core semantic tag is compared with the node semantic tags in the current node processing rule to check if they are completely identical. If the core semantic tag is completely identical with the node semantic tags in the current node processing rule, the comparison stops. If they are not identical, the next node processing rule is read and the comparison is repeated until a completely identical node semantic tag is found or all node processing rules have been traversed. If no node semantic tag completely identical with the core semantic tag is found after traversing all node processing rules, the adjacent processing node determines that the target file does not need to be processed and does not initiate a download request. The preset node processing rules are based on specific domains, such as the processing requirements of contract document management, and include node semantic tags and corresponding processing conditions. For example, the "financial related" tag corresponds to "requires financial department review". When the core semantic tag is exactly the same as the node semantic tag in a certain node processing rule, it is determined that the adjacent processing node needs to process the target file and initiates a download request carrying the smart document identifier to the streaming server.

[0034] S4. The streaming server receives the download request, verifies the validity of the smart document identifier, retrieves the target file storage location, asynchronously pulls the target file to the adjacent processing node according to the routing instructions, and updates the target file status to "distributed".

[0035] The transfer server receives download requests carrying smart document identifiers initiated by adjacent processing nodes; Extract the smart document identifier from the download request, retrieve the locally maintained hash value record, and if the hash value record contains the extracted smart document identifier, then the smart document identifier is deemed valid. Specifically, the streaming server parses the smart document identifier from the download request, reads the locally maintained hash value record, which stores all generated smart document identifiers; it compares the extracted smart document identifier with each smart document identifier in the hash value record one by one to check if they are completely identical; if the extracted smart document identifier is completely identical to any smart document identifier in the hash value record, the extracted smart document identifier is deemed valid; if no completely identical smart document identifier is found after traversing the hash value record, the extracted smart document identifier is deemed invalid, and the streaming server returns an error response to the adjacent processing node. The access path of the target file in the locally maintained metadata is retrieved using the smart document identifier, and used as the storage location of the target file. Specifically: the streaming server reads the locally maintained metadata set, which stores all semantic digest metadata sets. Each semantic digest metadata set contains a smart document identifier and the access path of the target file. Each semantic digest metadata set in the metadata set is traversed one by one, and the smart document identifier is extracted from it. The extracted smart document identifier is compared with the input smart document identifier to check if they are completely identical. If the extracted smart document identifier is completely identical to the input smart document identifier, the access path of the target file is obtained from the corresponding semantic digest metadata set and used as the storage location of the target file. If no completely identical smart document identifier is found after traversing the metadata set, an empty path is returned, indicating that the target file storage location is unavailable. Based on the adjacent processing node addresses specified in the routing instructions, an asynchronous transmission task is initiated to retrieve the target file from its storage location to the adjacent processing node. Specifically: the streaming server extracts the IP address and port number of the adjacent processing node from the routing instructions; it initializes the connection using an HTTP client library, establishing a TCP connection based on the IP address and port number of the adjacent processing node; it sends an HTTP GET request through the TCP connection, which includes the access path to the target file's storage location; the local HTTP server at the target file's storage location responds to the request, and the streaming server receives the binary data stream of the target file in a streaming mode; the received binary data stream is written to the specified storage path of the adjacent processing node via an asynchronous thread; the asynchronous thread runs in the background, allowing the streaming server to continue processing other tasks without waiting for the writing to complete, until the entire binary data stream of the target file has been transmitted to the adjacent processing node. After the asynchronous transmission task starts, the streaming server updates the status field of the target file to "distributed" in the locally maintained metadata.

[0036] This embodiment also provides a computer device applicable to the file transfer method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the file transfer method based on artificial intelligence as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the file transfer method based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] In summary, this invention achieves a unified expression of file content and structured attributes by using a generative artificial intelligence model on the upload client to generate semantic summaries and integrate basic file attributes. This semantic summarization and attribute integration method ensures that the file always contains complete semantic information during transmission, enabling subsequent processing nodes to accurately identify and match based on semantic tags, avoiding reliance on manual configuration or fixed rules. Through semantic parsing and unified attribute expression technology, files possess interpretability and searchability during cross-node transmission, thereby improving the automation and flexibility of file flow and achieving efficient distribution and accurate processing of files across different processing nodes.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A file transfer method based on artificial intelligence, characterized in that: include, The uploading client moves the target file to the local HTTP server storage path, calls the generative artificial intelligence model to parse the target file, and generates a semantic summary; it then integrates the basic file attributes and semantic summary of the target file to form a semantic summary metadata package. The streaming server receives the semantic digest metadata data package, performs hash operations on the basic file attributes and semantic digest in the semantic digest metadata data package, and generates a smart document identifier; The system calls a natural language processing model to parse the semantic summary in the semantic summary metadata data package and extract the core semantic tags; it then matches the routing rules in the business rule base according to the core semantic tags, generates routing instructions, and returns the intelligent document identifier and routing instructions to the upload client. The uploading client transmits the smart document identifier to the adjacent processing node. The adjacent processing node extracts the core semantic tag from the smart document identifier. If the core semantic tag matches the preset node processing rule, it is determined that the adjacent processing node needs to process the target file and initiates a download request carrying the smart document identifier to the streaming server. The distribution server receives the download request, verifies the validity of the smart document identifier, retrieves the target file storage location, asynchronously pulls the target file to the adjacent processing node according to the routing instructions, and updates the target file status to "distributed".

2. The file transfer method based on artificial intelligence as described in claim 1, characterized in that: The generation of semantic summaries specifically involves, The upload client receives the target file submitted by the user, obtains the storage path and file format information of the target file, moves the target file to the pending folder specified by the local HTTP server, reads the binary data stream of the target file, determines the target file type based on the target file format information, and extracts the text content of the target file. The extracted text content is segmented into fixed lengths to form structured input text; By calling the generative AI model deployed on the client side through the local interface, the structured input text is input into the generative AI model, the structured input text is analyzed, and the summary text is output. The uploading client receives the summary text, verifies the integrity of the summary text, and uses the complete summary text as the semantic summary.

3. The file transfer method based on artificial intelligence as described in claim 2, characterized in that: The generative artificial intelligence model analyzes the structured input text and outputs a summary text. This means that the structured input text is received and processed layer by layer in the encoder layer to extract semantic features between word segments. The encoder layer assigns attention weights to each word segment through a self-attention mechanism to enhance the contextual association in the structured input text and outputs feature representations. The decoder layer receives feature representations, predicts the next word segment word by word, and gradually generates summary text.

4. The file transfer method based on artificial intelligence as described in claim 1, characterized in that: The generation of intelligent document identifiers specifically refers to, The streaming server receives semantic digest metadata data packets sent by the uploading client; The semantic summary metadata package is deserialized to restore the structured key-value pair set, and the file name, file type, file size, check value and semantic summary are extracted from the structured key-value pair set; The fields are concatenated in the order of filename, file type, file size, checksum, and semantic digest to form the original string; The original string is hashed using a secure hash algorithm to obtain a fixed-length hash value, which serves as the smart document identifier.

5. The file transfer method based on artificial intelligence as described in claim 1, characterized in that: The process involves calling a natural language processing model to parse the semantic summary in the semantic summary metadata data package and extracting the core semantic tags. Specifically, this includes... Load the pre-trained Chinese text classification model, input the semantic summary into the Chinese text classification model, perform forward inference, and obtain the classification probability distribution; The category label with the highest classification probability value is selected from the classification probability distribution as the core semantic label.

6. The file transfer method based on artificial intelligence as described in claim 5, characterized in that: The forward inference described is a process in which the Chinese text classification model processes the semantic summary of the input based on the BERT model architecture. In the encoder layer, the semantic summary is processed layer by layer to extract semantic features between Chinese word segments. The encoder layer uses a self-attention mechanism to assign attention weights to each Chinese word segment, enhances contextual association, and outputs feature representations. The additional classification layer receives the feature representations, uses the feature representations as input vectors, and passes them to the fully connected layer. The fully connected layer processes the feature representations and applies the softmax function to generate the class probability distribution.

7. The file transfer method based on artificial intelligence as described in claim 1, characterized in that: The generated routing instructions are specifically as follows: The core semantic tags are compared one by one with the rule semantic tags of each rule entry in the business rule base; If the core semantic label is exactly the same as the rule semantic label of rule entry A, then the routing rule of rule entry A will be used as the routing instruction.

8. The file transfer method based on artificial intelligence as described in claim 7, characterized in that: The business rule base is a set of key-value pairs set based on the routing requirements of contract document management. It is used to store rule entries. Each rule entry consists of a rule semantic tag and a routing rule, which determines the transmission path of the target file.

9. The file transfer method based on artificial intelligence as described in claim 1, characterized in that: The transfer server receives the download request and verifies the validity of the smart document identifier, specifically by: The transfer server receives download requests carrying smart document identifiers initiated by adjacent processing nodes; Extract the smart document identifier from the download request, retrieve the locally maintained hash value record, and if the hash value record contains the extracted smart document identifier, then the smart document identifier is considered valid.

10. The file transfer method based on artificial intelligence as described in claim 1, characterized in that: The process of retrieving the target file's storage location involves asynchronously fetching the target file to adjacent processing nodes based on routing instructions, and updating the target file's status to "distributed." Specifically, this involves... The access path of the target file in the locally maintained metadata package is retrieved by intelligent document identification, and used as the storage location of the target file. Based on the address of the adjacent processing node specified in the routing instruction, an asynchronous transmission task is initiated to pull the target file from the storage location to the adjacent processing node; After the asynchronous transmission task starts, the streaming server updates the status field of the target file to "distributed" in the locally maintained metadata.

Images