File metadata generation method and device based on large model, equipment and medium

Abstract

The invention provides a file metadata generation method and device based on a large model, equipment and a medium, and relates to the field of artificial intelligence, in particular to the technical fields of NLP, the large model, LLM, deep learning, big data and the like. According to the specific implementation scheme, the method comprises the steps of performing multi-modal analysis on file content of an unstructured file to obtain multi-modal information; performing semantic reasoning on the multi-modal information to obtain semantic metadata; analyzing the multi-modal information by adopting a business rule matched with a target business scene to which the unstructured file belongs, so as to obtain business metadata; and generating file metadata of the unstructured file according to the semantic metadata and the business metadata.

Description

Technical Field

[0001] The present disclosure relates to the field of AI (Artificial Intelligence), specifically to technical fields such as NLP (Natural Language Processing), big models, LLM (Large Language Model), deep learning, big data, and especially to methods, devices, equipment and media for generating file metadata based on big models. Background Art

[0002] With the rapid development of information technology, the amount of data has shown an explosive growth trend. In this huge ocean of data, unstructured data accounts for the vast majority. Due to the huge amount and variety of these unstructured data, traditional data management methods can no longer meet the needs of efficient storage, retrieval and analysis. Therefore, in order to better manage and utilize these valuable resources, it is particularly important to automatically extract the metadata of unstructured files. Summary of the invention

[0003] The present disclosure provides a method, apparatus, device and medium for generating file metadata based on a large model.

[0004] According to one aspect of the present disclosure, a method for generating file metadata based on a large model is provided, comprising:

[0005] Perform multimodal analysis on the file content of unstructured files to obtain multimodal information;

[0006] Performing semantic reasoning on the multimodal information to obtain semantic metadata;

[0007] The multimodal information is parsed using a business rule adapted to a target business scenario to which the unstructured file belongs, so as to obtain business metadata;

[0008] Generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0009] According to another aspect of the present disclosure, there is provided a device for generating file metadata based on a large model, comprising:

[0010] A first parsing module is used to perform multimodal parsing on the file content of the unstructured file to obtain multimodal information;

[0011] A semantic reasoning module, used for performing semantic reasoning on the multimodal information to obtain semantic metadata;

[0012] A second parsing module, configured to parse the multimodal information using a business rule adapted to a target business scenario to which the unstructured file belongs, so as to obtain business metadata;

[0013] A generating module is used to generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0014] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for generating file metadata based on a large model proposed in the above aspect of the present disclosure.

[0018] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium of computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method for generating file metadata based on a large model proposed in the above-mentioned first aspect of the present disclosure.

[0019] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the method for generating file metadata based on a large model proposed in the above-mentioned aspect of the present disclosure.

[0020] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0022] Figure 1 A schematic diagram of a flow chart of a method for generating file metadata based on a large model provided in the first embodiment of the present disclosure;

[0023] Figure 2 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 2 of the present disclosure;

[0024] Figure 3 A schematic diagram of a flow chart of a method for generating file metadata based on a large model provided in Embodiment 3 of the present disclosure;

[0025] Figure 4 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 4 of the present disclosure;

[0026] Figure 5 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 5 of the present disclosure;

[0027] Figure 6 A schematic diagram of the structure of the metadata discovery system based on LLM provided in an embodiment of the present disclosure;

[0028] Figure 7 A schematic diagram of the implementation principle of any embodiment of the present disclosure;

[0029] Figure 8 A schematic diagram of the structure of a device for generating file metadata based on a large model provided in Embodiment 6 of the present disclosure;

[0030] Fig. 9 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0031] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0032] With the rapid development of big model technology, the importance of unstructured data (such as text, images, audio, video, etc.) has become increasingly prominent. These unstructured data account for the vast majority of enterprise and Internet data and are the core data source for big model training and reasoning. However, the inherent characteristics of unstructured data (such as diverse formats, lack of defined structure, and difficulty in parsing) have at least the following problems:

[0033] 1) Unstructured data is difficult to manage and organize:

[0034] Unstructured data is usually stored in distributed file systems, object storage or data lakes in the form of files, lacking a unified structure. Therefore, how to quickly discover, classify and describe these unstructured data is an important challenge for data governance.

[0035] 2) Automatic discovery of metadata is difficult:

[0036] The metadata of unstructured files (such as technical attributes, semantic information, and business context) is usually implicit and requires complex parsing and contextual reasoning. Therefore, how to automatically extract and generate metadata for unstructured files is the key to improving data availability and visibility.

[0037] 3) Problems of tracing and managing training data in large model data engineering:

[0038] During the training and reasoning process of large models, the metadata of unstructured files (such as data source, processing history, semantic tags, etc.) directly affects the transparency and performance of large models. The lack of a complete metadata discovery mechanism will lead to inefficient data governance and model management.

[0039] In related technologies, metadata management is mainly carried out for structured data, for example:

[0040] 1. Database metadata management: Use metadata management tools to generate technical metadata and business metadata for objects such as tables, fields, indexes, etc. in relational databases and data warehouses.

[0041] 2. Metadata management in the data lake: Use open source tools to extract metadata from semi-structured files in formats such as JSON (JavaScript Object Notation, an open standard file format and data exchange format), and support automatic crawling of technical metadata such as file format and field name.

[0042] 3. Data asset catalog: Use open source tools (such as Amundsen) and commercial platforms (such as Informatica) to classify, index and retrieve data assets.

[0043] However, the implementation solutions for metadata discovery of unstructured files are relatively limited, and only a few technologies have attempted to extract metadata from unstructured files. For example:

[0044] 4. Technical metadata extraction based on file attributes: Based on file system scanning technology (such as Hadoop HDFS (Hadoop Distributed File System) or S3 (Simple Storage Service) metadata API), extract basic metadata information such as file name, size, creation time, etc.

[0045] However, this method can only extract technical metadata and cannot parse file content or generate semantic information.

[0046] 5. Metadata extraction based on content parsing: Use tools (such as Apache Tika) to parse the file content of unstructured files to extract basic semantic information (such as text in PDF (Portable Document Format) documents and EXIF ​​(Exchangeable Image File Format) metadata in image files).

[0047] However, this method cannot perform deep semantic analysis on complex file structures.

[0048] In summary, the limitations of relevant technologies are still significant, mainly in the following aspects:

[0049] (1) The ability to extract technical metadata is limited and the range of file attributes is narrow: Related technologies mostly rely on file system APIs to extract basic attributes (such as file name, size, and timestamp), which makes it difficult to meet the extraction requirements of technical metadata in complex scenarios (such as the nested relationship between objects in the file and the encoding method).

[0050] (2) Lack of contextual understanding: Traditional tools cannot generate semantic metadata by combining file content and context. For example, a document containing “quarterly financial report” may have semantic metadata including “financial report”, “time range (quarter)”, and “department (finance department)”, but related technologies cannot automatically generate this semantic information.

[0051] (3) Limited ability to extract technical metadata and insufficient file format support: Existing tools have limited support for multimodal files (such as PDF or video files containing text, images, and audio) and are unable to fully parse technical details.

[0052] (4) Insufficient understanding of semantic hierarchy: Static parsing methods can only extract the surface logic of data flow and have difficulty understanding complex business logic or implicit relationships between fields. For example, some programs perform fuzzy matching on fields or dynamically generate field names, which cannot be effectively parsed by existing methods.

[0053] Therefore, in response to at least one of the above problems, the present disclosure proposes a method, apparatus, device and medium for generating file metadata based on a large model.

[0054] The following describes the method, device, apparatus and medium for generating file metadata based on a large model according to an embodiment of the present disclosure with reference to the accompanying drawings. Before describing the embodiment of the present disclosure in detail, for ease of understanding, common technical terms are first introduced:

[0055] A large model refers to a machine learning model with large-scale parameters and complex computational structures. It is usually built from deep neural networks and has billions or even hundreds of billions of parameters. The purpose of designing a large model is to improve the model's expressiveness and predictive performance, and to be able to handle more complex tasks and data. Large models are widely used in various fields, including natural language processing, computer vision, speech recognition, and recommendation systems.

[0056] The LLM in the big model is a type of natural language processing model based on deep learning. Its main features are huge model parameters and complex neural network structure, strong language understanding, context perception and language generation capabilities. It can automatically learn useful feature representations from input data and generate relevant text.

[0057] An intelligent agent refers to a computer program based on LLM that has planning, memory, and the ability to use tool functions and can autonomously complete a given task.

[0058] Technical metadata is information that describes the structure, storage, and processing of data from the perspective of technical implementation. Technical metadata is the basis for data operation and management, and provides developers and operation and maintenance personnel with the physical implementation details of data. In data governance, the clarity and accuracy of technical metadata can improve the maintainability of the system and the transparency of data flow.

[0059] Business metadata describes the meaning, purpose, and context of data from a business perspective. It is oriented to business users and helps non-technical personnel understand the value and usage specifications of data. Business metadata is a bridge between business and technology, helping business personnel better understand and use data. It provides a unified business language for data governance and reduces the deviation in data understanding between departments.

[0060] Semantic metadata is a knowledge representation of file content that goes beyond simple structural description or business definition and goes deep into the meaning and relationship level of file content. Semantic metadata provides explanatory information about file content, enabling computer systems and users to better understand and use file data.

[0061] Figure 1 A flowchart of a method for generating file metadata based on a large model provided in the first embodiment of the present disclosure.

[0062] The embodiment of the present disclosure takes the example of the method for generating file metadata based on the big model being configured in the device for generating file metadata based on the big model. The device for generating file metadata based on the big model can be applied to any electronic device so that the electronic device can perform the function of generating file metadata.

[0063] Among them, the electronic device can be any device with computing capabilities, such as a personal computer, a mobile terminal, a server, etc. The mobile terminal can be, for example, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, and other hardware devices with various operating systems, touch screens and / or display screens.

[0064] like Figure 1 As shown, the method for generating file metadata based on the large model may include the following steps S101 to S104:

[0065] Step S101, performing multimodal analysis on the file content of the unstructured file to obtain multimodal information.

[0066] There is no restriction on the method of obtaining unstructured files. For example, unstructured files may be files provided by users, or unstructured files may be files collected online, or unstructured files may be files collected from data sources (such as distributed file systems, object storage, data lakes, etc.), and so on.

[0067] There is no restriction on the business scenario to which the unstructured file belongs (referred to as the target business scenario in this disclosure). For example, the target business scenario includes but is not limited to: finance, medical care, energy and other scenarios, wherein each business scenario can also be finely divided into multiple sub-business scenarios. Taking the financial scenario as an example, the financial scenario can also include contract scenarios, transaction record scenarios, account management scenarios, etc.

[0068] Among them, multimodality is a way of transmitting and receiving information using multiple information channels or media, including but not limited to: documents (such as PDF), text, images, audio, and video.

[0069] In the disclosed embodiment, after the unstructured file is acquired, the unstructured file may be subjected to multimodal analysis to obtain multimodal information (or multimodal content).

[0070] Step S102: perform semantic reasoning on the multimodal information to obtain semantic metadata.

[0071] The semantic metadata includes multiple key fields and contextual relationships between the multiple key fields.

[0072] Among them, key fields include but are not limited to: keywords, topics, named entities, semantic tags, etc.

[0073] Among them, contextual relationships include but are not limited to: direct association relationships, implicit relationships, dependency relationships, logical order (such as time order, cause-effect order, etc.) relationships, subject consistency relationships, etc.

[0074] As an example, semantic extraction technology or semantic reasoning technology may be used to perform semantic reasoning on multimodal information to obtain semantic metadata.

[0075] As another example, the target business scenario to which the unstructured file belongs can also be determined, and based on the target business scenario, semantic reasoning can be performed on the multimodal information to obtain semantic metadata under the target business scenario, wherein the semantic metadata includes multiple key fields under the target business scenario and the contextual relationship between the multiple key fields.

[0076] Step S103 , using business rules adapted to the target business scenario to which the unstructured file belongs, the multimodal information is parsed to obtain business metadata.

[0077] Among them, business metadata includes but is not limited to: sensitive fields, business purposes (such as procurement contracts), compliance labels (such as GDPR (General Data Protection Regulation)), etc. Among them, sensitive fields include but are not limited to: bank account number, ID number, etc.

[0078] The business rules adapted to different business scenarios may be the same or different, and the embodiments of the present disclosure do not limit this. For example, corresponding business rules may be configured in advance for each business scenario.

[0079] Among them, business rules include but are not limited to: identification rules for sensitive fields, determination rules for business purposes, etc.

[0080] In the disclosed embodiment, the business rules adapted to the target business scenario to which the unstructured file belongs may be queried, and the multimodal information may be parsed based on the business rules to obtain the business metadata.

[0081] Optionally, in order to improve the completeness and richness of the extracted business metadata, LLM may be used to extract business metadata from the multimodal information, and the business metadata is integrated with the business metadata extracted in step S103 to obtain complete business metadata.

[0082] As an example, a prompt template (Prompt Templates) adapted to the target business scenario can be obtained, which is recorded as the first prompt template in the present disclosure; wherein the first prompt template is used to instruct the large language model to perform the business metadata acquisition task, as well as the business information to be extracted in the target business scenario. Afterwards, the first prompt information can be generated according to the multimodal information and the first prompt template, and the large language model can be called to process the first prompt information to obtain the business metadata.

[0083] Step S104: Generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0084] In the embodiment of the present disclosure, file metadata of an unstructured file may be generated according to semantic metadata and business metadata, that is, the file metadata includes semantic metadata and business metadata.

[0085] The method for generating file metadata based on a large model of the disclosed embodiment can capture more comprehensive, rich and accurate multimodal information by performing multimodal analysis on the file content of unstructured files, and then perform semantic reasoning on the complete and accurate multimodal information, which can deeply mine implicit and deep-level semantic metadata, thereby significantly improving the accuracy and practicality of semantic metadata. In addition, the use of business rules that are closely adapted to the business scenarios to which unstructured files belong to parse multimodal information can ensure that the generated business metadata is highly consistent with the actual business needs in the target business scenario, thereby optimizing and improving the efficiency and accuracy of business processing. Furthermore, the integration of semantic metadata and business metadata to generate file metadata for unstructured files can improve the rationality, richness and reliability of business metadata generation, thereby providing an accurate and reliable data source for unstructured data management, and providing strong support for enterprise data governance, business decision-making and intelligent applications.

[0086] It should be noted that in the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out with the user's consent, and are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0087] In order to clearly explain how to generate file metadata of unstructured files based on semantic metadata and business metadata in any embodiment of the present disclosure, the present disclosure also proposes a method for generating file metadata based on a large model.

[0088] Figure 2 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 2 of the present disclosure.

[0089] like Figure 2 As shown, the method for generating file metadata based on the large model may include the following steps S201 to S206:

[0090] Step S201, performing multimodal analysis on the file content of the unstructured file to obtain multimodal information.

[0091] Step S202: perform semantic reasoning on the multimodal information to obtain semantic metadata.

[0092] Step S203 , using business rules adapted to the target business scenario to which the unstructured file belongs, the multimodal information is parsed to obtain business metadata.

[0093] It should be noted that the explanation of steps S201 to S203 can be found in the relevant description in any embodiment of the present disclosure, and will not be repeated here.

[0094] Step S204: obtaining basic attributes and complex attributes of the unstructured file.

[0095] The basic attributes include, but are not limited to: file name, file type, file size, creation time, modification time, etc.

[0096] Among them, complex attributes include but are not limited to: the nested relationship between objects in the unstructured file, which is recorded as the first nested relationship in the present disclosure, and the first nested relationship is used to describe the hierarchical structure between the parent data object and the child data object; the encoding method of the unstructured file; the content quality of the unstructured file; the file traceability information of the unstructured file, etc.

[0097] In any embodiment of the present disclosure, the basic attributes of the unstructured file may be extracted using a file system API.

[0098] In any embodiment of the present disclosure, the complex attributes of the unstructured file can be obtained by following steps A to C:

[0099] Step A: Obtain the first nested relationship between objects in the unstructured file.

[0100] As an example, LLM may be called to process the multimodal information of the unstructured file to obtain the first nested relationship output by LLM.

[0101] Step B: Detect the file content of the unstructured file to obtain the encoding method and content quality of the unstructured file.

[0102] As an example, LLM may be called to detect the file content of an unstructured file to obtain an encoding method of the unstructured file.

[0103] As an example, LLM may be called to perform quality assessment on the accuracy, completeness, consistency, readability, etc. of the file content of the unstructured file to obtain the content quality of the unstructured file.

[0104] As another example, a content quality assessment model or a content quality assessment algorithm may be used to perform quality assessment on the accuracy, completeness, consistency, readability, etc. of the file content of the unstructured file to obtain the content quality of the unstructured file.

[0105] Step C: Obtaining complex attributes of the unstructured file according to at least one of the first nested relationship, the encoding method, and the content quality.

[0106] That is, the complex attributes of the unstructured file include at least one of the first nested relationship, the encoding method, and the content quality.

[0107] Therefore, by integrating the nested relationship between objects in unstructured files, the encoding method of unstructured files and the content quality, the complex attributes of unstructured files can be generated, which can improve the precision of data management, enhance the flexibility of data processing, promote the mining and utilization of data value, and optimize data storage and retrieval. This helps enterprises better manage and utilize unstructured data resources and improve data governance and business decision-making capabilities.

[0108] As a possible implementation method, in order to improve the richness and comprehensiveness of complex attribute acquisition, thereby improving the accuracy of technical metadata generation, the following steps C_1 to C_3 can also be used to obtain complex attributes of unstructured files:

[0109] Step C_1: Obtain file traceability information of unstructured files.

[0110] Among them, file traceability information includes detailed records of the file source, creation, modification, transmission and storage of unstructured files, which is crucial to ensuring the security, integrity and compliance of unstructured files.

[0111] Exemplarily, the file traceability information includes, but is not limited to: the creator, modifier, modification time, transmission path, storage location, etc. of the unstructured file.

[0112] Step C_2: Obtain a nested relationship graph associated with the unstructured file; wherein the nested relationship graph is used to indicate a second nested relationship of the unstructured file in the storage system.

[0113] The second nested relationship is used to indicate the hierarchical structure of unstructured files in the storage system, that is, how unstructured files are organized in folders (or directories). This nested relationship is crucial for tasks such as file classification, archiving, searching, and access control.

[0114] Step C_3: Obtain the complex attributes of the unstructured file according to at least one of the first nested relationship, encoding method, content quality, file traceability information and the nested relationship diagram.

[0115] That is, the complex attributes of the unstructured file include at least one of the first nested relationship, the encoding method, the content quality, the file traceability information, and the nested relationship graph.

[0116] As a result, the richness and completeness of complex attribute generation of unstructured files can be improved, thereby further improving the sophistication of data management, enhancing the flexibility of data processing, promoting the mining and utilization of data value, and optimizing data storage and retrieval.

[0117] Step S205: Generate technical metadata of the unstructured file according to the basic attributes and the complex attributes.

[0118] That is, the technical metadata of the unstructured file includes the above-mentioned basic attributes and complex attributes.

[0119] Step S206: Generate file metadata of the unstructured file according to the semantic metadata, business metadata and technical metadata.

[0120] In the disclosed embodiment, the above-mentioned semantic metadata, business metadata and technical metadata may be integrated to obtain file metadata of the unstructured file.

[0121] The method for generating file metadata based on a large model in the disclosed embodiment integrates the semantic metadata, business metadata and technical metadata of unstructured files to generate file metadata for unstructured files, which can improve the richness, comprehensiveness and accuracy of file metadata generation, thereby providing an accurate and reliable data source for unstructured data management, and providing strong support for enterprise data governance, business decision-making and intelligent applications.

[0122] In order to clearly explain how semantic reasoning is performed on multimodal information in the above embodiments to obtain semantic metadata of unstructured files, the present disclosure also proposes a method for generating file metadata based on a large model.

[0123] Figure 3 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 3 of the present disclosure.

[0124] like Figure 3 As shown, the method for generating file metadata based on the large model may include the following steps S301 to S307:

[0125] Step S301, performing multimodal analysis on the file content of the unstructured file to obtain multimodal information.

[0126] It should be noted that the explanation of step S301 can refer to the relevant description in any embodiment of the present disclosure, and will not be repeated here.

[0127] In any embodiment of the present disclosure, the multimodal information includes but is not limited to at least one of the following:

[0128] The first item is the text content in the file content of the unstructured file.

[0129] The text content includes the plain text content and document content in the file content, for example, the text content extracted from the PDF document.

[0130] The second item is, in response to the file content including an image, image embedding content obtained by performing character recognition on the image.

[0131] For example, the OCR (Optical Character Recognition) technology may be used to perform character recognition on the image to obtain the image embedded content.

[0132] Taking the target business scenario of the unstructured file as a contract scenario as an example, the image embedded content includes but is not limited to: company name, signature, seal information, etc.

[0133] The third item is, in response to the file content including audio, performing speech recognition on the audio to obtain audio embedded content.

[0134] For example, ASR (Automatic Speech Recognition) technology may be used to perform speech recognition on the audio to obtain audio embedded content.

[0135] Taking the target business scenario to which the unstructured file belongs as a contract scenario for exemplary description, the audio embedded content includes but is not limited to: contract negotiation content.

[0136] A fourth item is, in response to the file content including a video, a text description of the video obtained by visually analyzing the video.

[0137] The video text description includes semantic information of the video and character information obtained by performing OCR recognition on the video frame. Exemplarily, the video text description may include a conference record segment.

[0138] Exemplarily, deep learning technology (such as a visual model) can be used to perform visual analysis on the video to obtain a text description of the video.

[0139] As an example, the video text description can be obtained by following steps D to F:

[0140] Step D: In response to the file content including a video, a key frame recognition technology is used to extract key frames from the video to obtain key video frames.

[0141] Step E: Perform visual analysis on key video frames to obtain frame text descriptions of key video frames.

[0142] Exemplarily, deep learning techniques (such as visual models) can be used to perform visual analysis on key video frames to obtain frame text descriptions.

[0143] Step F, combining the frame text descriptions of the key video frames according to the arrangement order of the key video frames in the video to obtain a video text description.

[0144] Therefore, by extracting key video frames from the video, the amount of data that needs to be processed can be greatly reduced while retaining important information in the video, which makes the visual analysis process of the video more efficient and reduces unnecessary consumption of computing resources. Visual analysis of key video frames can quickly identify key elements and scenes in the video. This method is more efficient than frame-by-frame analysis and can quickly generate a preliminary understanding of the video content. In addition, through visual analysis, the visual content of key video frames can be converted into frame text descriptions, which can clearly express the key information and details in the video. According to the order of arrangement of key video frames in the video, the frame text descriptions can be combined to generate a coherent video text description, which reflects the overall structure and flow of the video content and helps to better understand the video content.

[0145] In summary, the richness, completeness and accuracy of multimodal information extraction can be improved, thereby improving the reliability and accuracy of subsequent file metadata generation.

[0146] Step S302: determine the target business scenario to which the unstructured file belongs.

[0147] The target business scenario may be a manually specified business scenario, or may be a business scenario obtained by automatically identifying unstructured files using relevant algorithms, and the embodiments of the present disclosure do not limit this.

[0148] As an example, semantic and grammatical analysis may be performed on the file content or multimedia information of the unstructured file to obtain the target business scenario to which the unstructured file belongs.

[0149] As another example, since the file content in different business scenarios has different language features, such as professional terminology, expressions, sentence structures, etc., in the present disclosure, the target business scenario to which the unstructured file belongs can also be identified by analyzing the file content of the unstructured file or the professional terminology, keywords and sentence structures in the multimedia information.

[0150] As another example, the target business scenario to which the unstructured file belongs can be identified based on NLP technology and text classification algorithms.

[0151] As another example, a special automated tool or software may be used to classify and identify unstructured files to obtain a target business scenario to which the unstructured file belongs.

[0152] It should be noted that the above-mentioned target business scenario identification method is only an exemplary description, but the present disclosure is not limited thereto. In actual application, other algorithms may also be used to identify the target business scenario to which the unstructured file belongs.

[0153] Step S303: perform semantic reasoning on the multimodal information to obtain multiple key fields in the target business scenario.

[0154] Among them, key fields include but are not limited to: keywords, topics, named entities, semantic tags, etc. Taking the target business scenario of the unstructured file as an example, the core subject of the contract can be "purchase agreement", and the named entities can include key information such as Party A, Party B, amount, time, etc. in the contract.

[0155] In the embodiments of the present disclosure, semantic extraction technology or semantic reasoning technology may be used to perform semantic reasoning on multimodal information to obtain multiple key fields in a target business scenario.

[0156] As an example, LLM can be used to perform semantic reasoning on multimodal information to obtain multiple key fields in the target business scenario.

[0157] Step S304: Generate contextual relationships between the multiple key fields based on the contextual information of the multiple key fields in the multimodal information.

[0158] Among them, contextual relationships include but are not limited to: direct association relationships, implicit relationships, dependency relationships, logical order (such as time order, cause-effect order, etc.) relationships, subject consistency relationships, etc.

[0159] In the embodiments of the present disclosure, NLP technology may be used to generate contextual relationships between multiple key fields based on context information of the multiple key fields in multimodal information.

[0160] As an example, LLM may be used to generate contextual relationships between multiple key fields based on context information of the multiple key fields in multimodal information.

[0161] Step S305: Generate semantic metadata according to the multiple key fields and the contextual relationships between the multiple key fields.

[0162] That is, the semantic metadata includes multiple key fields in the target business scenario and the contextual relationships between the multiple key fields.

[0163] Step S306 , using business rules adapted to the target business scenario to which the unstructured file belongs, the multimodal information is parsed to obtain business metadata.

[0164] Step S307: Generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0165] It should be noted that the explanation of steps S306 to S307 can be found in the relevant description in any embodiment of the present disclosure and will not be repeated here.

[0166] The method for generating file metadata based on a large model in the disclosed embodiment includes not only multiple key fields in the target business scenario to which the unstructured file belongs, but also the contextual relationship between multiple key fields, which can reveal the intrinsic connection and mutual influence between data, which helps to understand the data more comprehensively and avoid viewing the data in a one-sided or isolated manner, that is, to improve the depth and breadth of data understanding. In addition, it can also optimize the efficiency and accuracy of data processing, support more advanced data applications, and improve the level of data governance, which helps to better manage and utilize data resources and promote data-driven business development.

[0167] In order to clearly illustrate how any embodiment of the present disclosure performs semantic reasoning on multimodal information to obtain semantic metadata of unstructured files, the present disclosure also proposes a method for generating file metadata based on a large model.

[0168] Figure 4 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 4 of the present disclosure.

[0169] like Figure 4 As shown, the method for generating file metadata based on the large model may include the following steps S401 to S406:

[0170] Step S401, performing multimodal analysis on the file content of the unstructured file to obtain multimodal information, and determining the target business scenario to which the unstructured file belongs.

[0171] It should be noted that the explanation of step S401 can refer to the relevant description in any embodiment of the present disclosure and will not be repeated here.

[0172] Step S402, obtaining a prompt template adapted to the target business scenario; wherein the prompt template is used to instruct the large language model to perform semantic reasoning tasks and semantic information to be extracted in the target business scenario.

[0173] In an embodiment of the present disclosure, the correspondence or adaptation relationship between different business scenarios and prompt templates can be pre-configured. Therefore, in the present disclosure, the above correspondence can be queried based on the target business scenario to determine the prompt template adapted to the target business scenario, which is recorded as the second prompt template in the present disclosure.

[0174] Among them, the second prompt template adapted to the target business scenario is used to instruct the LLM to perform semantic reasoning tasks and semantic information to be extracted in the target business scenario.

[0175] Exemplarily, the second prompt template may also be used to prompt the LLM for output format requirements, such as prompting the LLM to output semantic metadata in a structured format.

[0176] It should be noted that the prompt template adapted to different business scenarios may be dynamically updated according to actual application requirements, and the embodiments of the present disclosure do not limit this.

[0177] In any embodiment of the present disclosure, the second prompt template may further include a reference example, wherein the reference example may include a reference text in a target business scenario and semantic information in the reference text.

[0178] As an example, the target business scenario to which the unstructured file belongs is taken as a contract scenario for exemplary description, and the reference example may be:

[0179] {

[0180] Please extract key information from this article based on the following contract:

[0181] Contract text:

[0182] Contract No.: 12345

[0183] Party A: ABC Company

[0184] Party B: XYZ Company

[0185] Contract amount: 500,000 yuan

[0186] Signing date: January 1, 2024

[0187] Output format:

[0188] {

[0189] "Contract Number": "12345",

[0190] “Party A”: “ABC Company”,

[0191] “Party B”: “XYZ Company”,

[0192] "Contract Amount": "ABC Company",

[0193] "Signature Date":"January 1, 2024"

[0194] }

[0195] }

[0196] It should be noted that the above example only takes the semantic information in the reference text including only key fields as an example. In actual application, the semantic information in the reference text may also include contextual relationships between multiple key fields.

[0197] Therefore, by supplementing the large language model with reference examples, the reference examples can be used as prior information to guide the learning of the large language model, thereby improving the prediction accuracy of the large language model, that is, improving the accuracy of the semantic reasoning of the large language model.

[0198] Step S403: Generate prompt information according to the multimodal information and the prompt template.

[0199] In an embodiment of the present disclosure, it can be determined whether the number of characters contained in the multimodal information exceeds the target number, where the target number is determined based on the input requirements of the large language model. If it does not exceed, the multimodal information is directly used to fill in the second prompt template to obtain prompt information, which is recorded as the second prompt information in the present disclosure.

[0200] If it exceeds, the multimodal information is segmented or split to obtain multiple text segments, and the multiple text segments are used to fill the second prompt template respectively to obtain multiple second prompt information.

[0201] Step S404: calling a large language model to perform semantic reasoning on the prompt information to obtain semantic metadata.

[0202] In the disclosed embodiment, a large language model may be called to perform semantic reasoning on the second prompt information to obtain semantic metadata output by the large language model.

[0203] Step S405 , using business rules adapted to the target business scenario to which the unstructured file belongs, the multimodal information is parsed to obtain business metadata.

[0204] Step S406: Generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0205] It should be noted that the explanation of steps S405 to S406 can be found in the relevant description in any embodiment of the present disclosure, and will not be repeated here.

[0206] The method for generating file metadata based on a large model in the disclosed embodiment extracts semantic metadata of unstructured files by adopting a prompt template adapted to the business scenario to which the unstructured files belong. The prompt template can be used as task information or prior information to indicate the semantic information to be extracted by the large language model, thereby improving the prediction accuracy of the large language model, that is, improving the accuracy of the semantic reasoning of the large language model, and meeting the personalized semantic reasoning needs in different business scenarios.

[0207] In order to clearly illustrate any embodiment of the present disclosure, the present disclosure also proposes a method for generating file metadata based on a large model.

[0208] Figure 5 A flowchart of a method for generating file metadata based on a large model provided in Embodiment 5 of the present disclosure.

[0209] like Figure 5 As shown, the method for generating file metadata based on the large model may include the following steps S501 to S506:

[0210] Step S501, performing multimodal analysis on the file content of the unstructured file to obtain multimodal information.

[0211] Step S502: perform semantic reasoning on the multimodal information to obtain semantic metadata.

[0212] Step S503 , using business rules adapted to the target business scenario to which the unstructured file belongs, the multimodal information is parsed to obtain business metadata.

[0213] Step S504: Generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0214] For explanations of steps S501 to S504, reference may be made to the relevant descriptions in any embodiment of the present disclosure, and will not be repeated here.

[0215] Step S505, vectorizing the file metadata to obtain an embedded vector of the file metadata.

[0216] As an example, a statistical method (such as a one-hot model, BOW (Bag of Words Model), TF-IDF (Term Frequency-Inverse Document Frequency), etc.) can be used to vectorize (Embedding) the file metadata and obtain an embedded vector of the file metadata.

[0217] As another example, a deep learning-based method (such as RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), etc.) can be used to vectorize the file metadata and obtain an embedded vector of the file metadata.

[0218] As another example, LLM may be used to vectorize the file metadata to obtain an embedded vector of the file metadata.

[0219] It should be noted that the above-mentioned vectorized representation method is only an exemplary description, but the present disclosure is not limited to this. In actual application, other algorithms can also be used to vectorize the file metadata, and the embodiments of the present disclosure are not limited to this.

[0220] Step S506, storing the embedded vector in a vector database; wherein the vector database is used for semantic retrieval.

[0221] In an embodiment of the present disclosure, the embedding vector of the unstructured file may be stored in a vector database, wherein the vector database is used for semantic retrieval.

[0222] In any embodiment of the present disclosure, the semantic search process may include the following steps a to d:

[0223] Step a: In response to receiving a natural language query text, vectorize the natural language query text to obtain a query vector.

[0224] The natural language query text is input by the user, and the input method of the natural language query text includes but is not limited to touch input (such as sliding, clicking, etc.), keyboard input, voice input, etc.

[0225] It should be noted that the vectorized representation of natural language query text is similar to the vectorized representation of file metadata, which will not be elaborated here.

[0226] Step b: From each embedding vector in the vector database, retrieve the target embedding vector that is the nearest neighbor to the query vector.

[0227] In the embodiments of the present disclosure, based on semantic retrieval technology, a target embedding vector that is the nearest neighbor to the query vector can be retrieved from each embedding vector in the vector database.

[0228] Step c: Obtain file metadata corresponding to the target embedding vector, which is referred to as target metadata in this disclosure, and obtain a target file associated with the target metadata.

[0229] The target file may be an unstructured text or a structured file, which is not limited in the embodiments of the present disclosure.

[0230] The target embedding vector is an embedding vector obtained by vectorizing the target metadata.

[0231] The target metadata is the file metadata of the target file.

[0232] Step d: Send the target file and target metadata.

[0233] Exemplarily, the target file and target metadata may be sent to a client or electronic device where a user who sends the natural language query text is located.

[0234] Therefore, the natural language query text entered by the user is vectorized, and the target embedding vector and its associated target files and target metadata are retrieved from the vector database, which can significantly improve the retrieval efficiency and user experience. That is, the vector database uses efficient index structures and algorithms to quickly find the target embedding vector that is most similar to the query vector, thereby greatly shortening the retrieval time and improving the system's response speed; by calculating the similarity between vectors, semantic-based precise matching can be achieved, which means that the system can more accurately understand the user's query intent and return the most relevant results, improving the user's retrieval experience.

[0235] The method for generating file metadata based on a large model of the disclosed embodiment vectorizes the file metadata of unstructured files, obtains the embedded vector of the file metadata, and stores it in a vector database for semantic retrieval. Since the vector database provides efficient storage and query capabilities for the unique structure of vector embedding, it uses advanced indexing algorithms to quickly retrieve relevant vectors in large-scale data sets, thereby significantly improving retrieval efficiency. This search method is more flexible and accurate than traditional keyword-based search methods and can better meet user query needs. In addition, after the file metadata is vectorized, the characteristics of the vector database can be used to organize and classify the data, which helps to simplify the data management process and improve the accessibility and manageability of the data.

[0236] In any one of the embodiments of the present disclosure, the present disclosure proposes an LLM semantic reasoning solution to comprehensively extract technical metadata, semantic metadata, and business metadata of unstructured files, and integrates data governance functions to incorporate unstructured files into a unified metadata management system to improve the comprehensiveness of data asset management.

[0237] The solution supports multimodal content analysis, semantic reasoning, business rule inference, and metadata storage. The specific implementation principle is as follows: Figure 6, mainly includes the following modules:

[0238] 1. File collection module and preliminary multimodal content analysis module.

[0239] Document Loaders based on LangChain (a development framework for applications driven by large language models) scan data sources (such as distributed file systems, data lakes, etc.), load file contents of unstructured files, and extract preliminary technical metadata.

[0240] 2. Multimodal content parsing module.

[0241] Combined with LangChain's toolset, multimodal analysis of text, images, audio and video content is performed to extract multimodal information.

[0242] As an example, the principle of extracting multimodal information can be as follows Figure 7 shown.

[0243] Among them, the toolset includes tools and agents.

[0244] 3. Semantic reasoning and context generation module.

[0245] Through LangChain's large language model interface (LLM Interface), a large language model is used to perform semantic reasoning on multimodal information and generate semantic metadata (such as keywords, topics, named entities, semantic tags, and contextual relationships) for unstructured files.

[0246] 4. Business rule inference module.

[0247] Based on business needs, business metadata (such as sensitive field tags and business purposes) are generated through LangChain's prompt templates and custom rule engines.

[0248] 5. Metadata storage and management module.

[0249] Utilize LangChain's vector storage to provide embedded vectorized storage and support efficient retrieval and metadata management.

[0250] In summary, the ability to discover metadata of unstructured files can be enhanced and a data source can be provided for unstructured data management.

[0251] With the above Figures 1 to 7Corresponding to the method for generating file metadata based on a large model provided in the embodiment, the present disclosure also provides a device for generating file metadata based on a large model. Figures 1 to 7 The embodiment provides a method for generating file metadata based on a big model, so the implementation method of the method for generating file metadata based on a big model is also applicable to the device for generating file metadata based on a big model provided in the embodiment of the present disclosure, and will not be described in detail in the embodiment of the present disclosure.

[0252] Figure 8 This is a schematic diagram of the structure of a device for generating file metadata based on a large model provided in Embodiment 6 of the present disclosure.

[0253] like Figure 8 As shown, the large model-based file metadata generation device 800 may include: a first parsing module 810, a semantic reasoning module 820, a second parsing module 830 and a generation module 840.

[0254] The first parsing module 810 is used to perform multimodal parsing on the file content of the unstructured file to obtain multimodal information;

[0255] A semantic reasoning module 820, for performing semantic reasoning on multimodal information to obtain semantic metadata;

[0256] A second parsing module 830 is used to parse the multimodal information using a business rule adapted to a target business scenario to which the unstructured file belongs, so as to obtain business metadata;

[0257] The generating module 840 is used to generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

[0258] In a possible implementation of the embodiment of the present disclosure, the generation module 840 is used to: obtain basic attributes and complex attributes of an unstructured file; generate technical metadata of the unstructured file based on the basic attributes and complex attributes; and generate file metadata of the unstructured file based on semantic metadata, business metadata, and technical metadata.

[0259] In a possible implementation of the embodiment of the present disclosure, the generation module 840 is used to: obtain a first nested relationship between objects in an unstructured file; detect the file content of the unstructured file to obtain the encoding method and content quality of the unstructured file; and obtain complex attributes of the unstructured file based on at least one of the first nested relationship, the encoding method, and the content quality.

[0260] In a possible implementation of the embodiment of the present disclosure, a generation module 840 is used to: obtain file traceability information of an unstructured file; obtain a nested relationship graph associated with the unstructured file; wherein the nested relationship graph is used to indicate a second nested relationship of the unstructured file in the storage system; and obtain complex attributes of the unstructured file based on at least one of the first nested relationship, encoding method, content quality, file traceability information, and the nested relationship graph.

[0261] In a possible implementation of the embodiment of the present disclosure, the semantic reasoning module 820 is used to: perform semantic reasoning on multimodal information to obtain multiple key fields in a target business scenario; generate contextual relationships between multiple key fields based on context information of the multiple key fields in the multimodal information; and generate semantic metadata based on the multiple key fields and the contextual relationships between the multiple key fields.

[0262] In a possible implementation of the embodiment of the present disclosure, the semantic reasoning module 820 is used to: obtain a prompt template adapted to the target business scenario; wherein the prompt template is used to instruct the large language model to perform a semantic reasoning task, as well as semantic information to be extracted in the target business scenario; generate prompt information based on multimodal information and the prompt template; call the large language model to perform semantic reasoning on the prompt information to obtain semantic metadata.

[0263] In a possible implementation manner of the embodiment of the present disclosure, the prompt template also includes a reference example, wherein the reference example includes a reference text in a target business scenario and semantic information in the reference text.

[0264] In a possible implementation of the embodiments of the present disclosure, the multimodal information includes at least one of the following: text content in the file content; image embedded content obtained by character recognition of the image in response to the file content including an image; audio embedded content obtained by speech recognition of the audio in response to the file content including audio; and video text description obtained by visual analysis of the video in response to the file content including a video.

[0265] In a possible implementation of the embodiment of the present disclosure, the video text description is obtained by using the following modules:

[0266] An extraction module, configured to extract key frames from the video in response to the file content including the video, to obtain key video frames;

[0267] An analysis module, used for visually analyzing key video frames to obtain frame text descriptions of the key video frames;

[0268] The combining module is used to combine the frame text descriptions of each key video frame according to the arrangement order of each key video frame in the video to obtain a video text description.

[0269] In a possible implementation of the embodiment of the present disclosure, the large model-based file metadata generation device 800 further includes:

[0270] The first processing module is used to vectorize the file metadata to obtain an embedded vector of the file metadata;

[0271] The storage module is used to store the embedded vector into a vector database; wherein the vector database is used for semantic retrieval.

[0272] In a possible implementation of the embodiment of the present disclosure, the large model-based file metadata generation device 800 further includes:

[0273] A second processing module is used for vectorizing the natural language query text in response to receiving the natural language query text to obtain a query vector;

[0274] A retrieval module, used to retrieve a target embedding vector that is the nearest neighbor to the query vector from each embedding vector in the vector database;

[0275] An acquisition module, used for acquiring target metadata corresponding to the target embedding vector and a target file associated with the target metadata;

[0276] The sending module is used to send the target file and target metadata.

[0277] The device for generating file metadata based on a large model of the disclosed embodiment can capture more comprehensive, rich and accurate multimodal information by performing multimodal analysis on the file content of unstructured files, and then perform semantic reasoning on the complete and accurate multimodal information, which can deeply mine implicit and deep-level semantic metadata, thereby significantly improving the accuracy and practicality of semantic metadata. In addition, the use of business rules that are closely adapted to the business scenarios to which unstructured files belong to parse multimodal information can ensure that the generated business metadata is highly consistent with the actual business needs under the target business scenario, thereby optimizing and improving the efficiency and accuracy of business processing. Furthermore, the integration of semantic metadata and business metadata to generate file metadata for unstructured files can improve the rationality, richness and reliability of business metadata generation, thereby providing an accurate and reliable data source for unstructured data management, and providing strong support for enterprise data governance, business decision-making and intelligent applications.

[0278] In order to implement the above embodiments, the present disclosure also provides an electronic device, which may include at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for generating file metadata based on a large model proposed in any of the above embodiments of the present disclosure.

[0279] In order to implement the above embodiments, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the method for generating file metadata based on a large model proposed in any of the above embodiments of the present disclosure.

[0280] In order to implement the above embodiments, the present disclosure further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the method for generating file metadata based on a large model proposed in any of the above embodiments of the present disclosure.

[0281] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0282] Fig. 9 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown.

[0283] Among them, the electronic device may include the server and the client in the above-mentioned embodiment. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0284] like Fig. 9As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 902 or a computer program loaded from a storage unit 907 to a RAM (Random Access Memory) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An I / O (Input / Output) interface 905 is also connected to the bus 904.

[0285] A number of components in the device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0286] The computing unit 901 may be various general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various special AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as the above-mentioned method for generating file metadata based on a large model. For example, in some embodiments, the above-mentioned method for generating file metadata based on a large model may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the method for generating file metadata based on a large model described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to execute the above-mentioned method for generating file metadata based on a large model in any other appropriate manner (for example, by means of firmware).

[0287] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0288] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0289] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0290] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0291] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0292] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services (Virtual Private Server). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0293] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0294] According to the technical solution of the embodiment of the present disclosure, by performing multimodal analysis on the file content of unstructured files, more comprehensive, rich and accurate multimodal information can be captured, and then semantic reasoning can be performed on the complete and accurate multimodal information, and implicit and deep semantic metadata can be deeply mined, thereby significantly improving the accuracy and practicality of semantic metadata. In addition, the use of business rules that are closely adapted to the business scenarios to which the unstructured files belong to parse multimodal information can ensure that the generated business metadata is highly consistent with the actual business needs in the target business scenarios, thereby optimizing and improving the efficiency and accuracy of business processing. Furthermore, the integration of semantic metadata and business metadata to generate file metadata for unstructured files can improve the rationality, richness and reliability of business metadata generation, thereby providing an accurate and reliable data source for unstructured data management, and providing strong support for enterprise data governance, business decision-making and intelligent applications.

[0295] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0296] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for generating file metadata based on a large model, comprising: Perform multimodal analysis on the file content of unstructured files to obtain multimodal information; Performing semantic reasoning on the multimodal information to obtain semantic metadata; The multimodal information is parsed using a business rule adapted to a target business scenario to which the unstructured file belongs, so as to obtain business metadata; Generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

2. The method according to claim 1, wherein: Generating the file metadata of the unstructured file according to the semantic metadata and the business metadata includes: Obtaining basic attributes and complex attributes of the unstructured file; Generating technical metadata of the unstructured file according to the basic attributes and the complex attributes; The file metadata of the unstructured file is generated according to the semantic metadata, the business metadata and the technical metadata.

3. The method according to claim 2, wherein: The complex attributes of the unstructured file are obtained by using the following steps: Acquire a first nested relationship between objects in the unstructured file; Detecting the file content of the unstructured file to obtain the encoding method and content quality of the unstructured file; The complex attribute of the unstructured file is acquired according to at least one of the first nested relationship, the encoding mode and the content quality.

4. The method according to claim 3, wherein: The acquiring the complex attribute of the unstructured file according to at least one of the first nested relationship, the encoding mode and the content quality includes: Obtaining file traceability information of the unstructured file; Acquire a nested relationship graph associated with the unstructured file; wherein the nested relationship graph is used to indicate a second nested relationship of the unstructured file in the storage system; The complex attribute of the unstructured file is obtained according to at least one of the first nested relationship, the encoding method, the content quality, the file source tracing information and the nested relationship graph.

5. The method according to claim 1, wherein: The performing semantic reasoning on the multimodal information to obtain semantic metadata of the unstructured file includes: Performing semantic reasoning on the multimodal information to obtain multiple key fields in the target business scenario; generating a contextual relationship between the multiple key fields based on context information of the multiple key fields in the multimodal information; The semantic metadata is generated according to the multiple key fields and the contextual relationships between the multiple key fields.

6. The method according to claim 1, wherein: The performing semantic reasoning on the multimodal information to obtain semantic metadata of the unstructured file includes: Acquire a prompt template adapted to the target business scenario; wherein the prompt template is used to instruct the large language model to perform a semantic reasoning task and the semantic information to be extracted in the target business scenario; Generate prompt information according to the multimodal information and the prompt template; The large language model is called to perform semantic reasoning on the prompt information to obtain the semantic metadata.

7. The method according to claim 6, wherein: The prompt template also includes reference examples, wherein the reference examples include reference text in the target business scenario and semantic information in the reference text.

8. The method according to any one of claims 1 to 7, wherein: The multimodal information includes at least one of the following: The text content in the file content; In response to the file content including an image, performing character recognition on the image to obtain image embedded content; In response to the file content including audio, performing speech recognition on the audio to obtain audio embedded content; In response to the file content including a video, a text description of the video is obtained by visually analyzing the video.

9. The method according to claim 8, wherein: The video text description is obtained by following the steps below: In response to the file content including a video, extracting key frames from the video to obtain key video frames; Performing visual analysis on the key video frame to obtain a frame text description of the key video frame; The frame text descriptions of the key video frames are combined according to the arrangement order of the key video frames in the video to obtain the video text description.

10. The method according to any one of claims 1 to 7, wherein: The method further comprises: Vectorizing the file metadata to obtain an embedded vector of the file metadata; The embedded vector is stored in a vector database; wherein the vector database is used for semantic retrieval.

11. The method according to claim 10, wherein: The method further comprises: In response to receiving a natural language query text, vectorizing the natural language query text to obtain a query vector; Retrieving a target embedding vector that is closest to the query vector from each embedding vector in the vector database; Acquire target metadata corresponding to the target embedding vector, and a target file associated with the target metadata; The target file and target metadata are sent.

12. A device for generating file metadata based on a large model, comprising: A first parsing module is used to perform multimodal parsing on the file content of the unstructured file to obtain multimodal information; A semantic reasoning module, used for performing semantic reasoning on the multimodal information to obtain semantic metadata; A second parsing module, configured to parse the multimodal information using a business rule adapted to a target business scenario to which the unstructured file belongs, so as to obtain business metadata; A generating module is used to generate file metadata of the unstructured file according to the semantic metadata and the business metadata.

13. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for generating file metadata based on a large model according to any one of claims 1 to 11.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the method for generating file metadata based on a large model according to any one of claims 1-11.

15. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the steps of the method for generating file metadata based on a large model according to any one of claims 1 to 11.

Images