File information identification method, system and device and computer program product

Abstract

The invention discloses a file information identification method, system and device and a computer program product. The method relates to the field of financial science and technology, and comprises the steps that a to-be-identified file sent by a block chain platform is received, and the to-be-identified file is sent to the block chain platform by a management platform; a model comparison rule is obtained, a target model is determined from the model comparison rule according to the file type of the to-be-recognized file, the model comparison rule comprises a corresponding relation between multiple file types and a preset model, and the preset model is used for recognizing document information of the to-be-recognized file; and identifying the to-be-identified file by using the target model to obtain file identification information. According to the method and the device, the problems of low accuracy of selecting the identification model and low accuracy and low efficiency of identifying the character information when the identification model is used for identifying the file in the related technology are solved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and specifically, to a method, system, device and computer program product for identifying file information. Background Art

[0002] In order to meet the needs of financial institutions for data security, digital safes can usually be used for secure data storage. Digital safes, also known as blockchain electronic filing cabinets and blockchain electronic evidence cabinets, are a platform for secure storage and management of digital files based on blockchain technology. Through the characteristics of blockchain technology such as immutability, traceability, and decentralization, it provides users with safe and reliable digital file storage and management services. However, in terms of processing and storage efficiency, blockchain digital safes are insufficient compared to traditional database systems. For example, the time required for transaction confirmation is long and the cost of data reading and writing is high. The above defects are particularly obvious in application scenarios that require large-scale data processing. For example, when financial institutions need to perform OCR (Optical Character Recognition) processing on a large number of documents, the efficiency will be greatly reduced.

[0003] In current technology, the file processing of digital safes usually relies on users to manually select and call different large model interfaces to perform OCR recognition. Although this method provides users with a certain degree of flexibility, it still leads to low processing efficiency, especially when facing the OCR recognition needs of large-scale unstructured files. Manual OCR recognition one by one is not only time-consuming, but also significantly increases the burden on the blockchain platform, such as transaction cost burden and network congestion risk. In addition, the existing pre-classification algorithm can only correspond to the large model level, and the selection of further subdivided models (such as models optimized for files of specific types or sources) cannot be realized. It is difficult to accurately match more fine-grained OCR models, which limits the accuracy of file recognition to a certain extent, and manual operations may also lead to consistency problems and errors, which in turn affect the overall recognition quality and the smoothness of business processes.

[0004] When using recognition models to recognize files in related technologies, there is a problem that the accuracy of the selected recognition model is low, resulting in low precision and low efficiency in recognizing text information. Currently, no effective solution has been proposed. Summary of the invention

[0005] The main purpose of the present application is to provide a method, system, device and computer program product for identifying file information, so as to solve the problem in the related art that when using recognition models to identify files, the accuracy of the selected recognition model is low, resulting in low precision and low efficiency in recognizing text information.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for identifying file information is provided. The method comprises: receiving a file to be identified sent by a blockchain platform, wherein the file to be identified is sent by a management platform to the blockchain platform; obtaining a model comparison rule, and determining a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; and identifying the file to be identified using the target model to obtain file identification information.

[0007] Further, in the case where the file classifier includes N pre-classifiers, N is a positive integer, and the model comparison rules are determined in the following manner: receiving M pre-trained information files sent by a batch trigger of the blockchain platform, wherein M is a positive integer; performing feature extraction on each pre-trained information file to obtain M file features; having the N pre-classifiers process each pre-trained information file in turn to obtain M processing results, wherein each processing result is used to characterize the correspondence between the file feature of each pre-trained information file and a preset model; and constructing the model comparison rules according to the M processing results.

[0008] Further, in the case where the file classifier includes three pre-classifiers, N pre-classifiers process each pre-trained information file in turn to obtain M processing results including: for a pre-trained information file, the file type of the pre-trained information file is obtained, and a first recognition model is determined according to the file features; the first pre-classifier in the file classifier calls the first recognition model to recognize the pre-trained information file to obtain a recognition message; the second pre-classifier in the file classifier calls the semantic vector model and the clustering algorithm in turn to process the recognition message to obtain a first sub-processing result, and matches the second recognition model according to the first sub-processing result; the second pre-classifier calls the second recognition model to recognize the first sub-processing result to obtain a first recognition result; the third pre-classifier in the file classifier calls the semantic vector model and the clustering algorithm in turn to process the first recognition result to obtain a second sub-processing result, and matches the third recognition model according to the second sub-processing result; the third pre-classifier calls the third recognition model to recognize the second sub-processing result to obtain a second recognition result, and determines the processing result according to the third recognition model.

[0009] Furthermore, the second pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the recognition message, and obtains the first sub-processing result including: using the regular expression disassembly rule to disassemble the recognition message to obtain L disassembly results, wherein each disassembly result includes multiple area sets, each area set includes multiple areas, each area is used to represent the information of the vocabulary in the recognition message, each area set represents the information of the sentence in the recognition message, and L is a positive integer; for one disassembly result, the disassembly result is input into the semantic vector model, and P first semantic vectors are output, wherein one first semantic vector is associated with one area, and P is a positive integer; for one disassembly result, the frequency weight of each area in the disassembly result is obtained, and the first semantic vector dimension value matrix is ​​determined according to the P first semantic vectors and the frequency weight of each area; the L first semantic vector dimension value matrices are merged using the clustering algorithm to obtain the first sub-processing result.

[0010] Furthermore, before using a clustering algorithm to merge the L first semantic vector dimension value matrices to obtain a first sub-processing result, the method also includes: for a first semantic vector dimension value matrix, calculating the Euclidean distance between every two regions in the first semantic vector dimension value matrix to obtain K Euclidean distances, and calculating the sum of the K Euclidean distances to obtain a result difference value, where K is a positive integer; sorting the L result difference values ​​to obtain a difference value sorting; using a clustering algorithm to merge the L first semantic vector dimension value matrices to obtain a first sub-processing result includes: using a clustering algorithm to merge the L first semantic vector dimension value matrices according to the difference value sorting to obtain Q clusters, where Q is a positive integer; obtaining the total number of first models of the candidate second recognition models stored in the text recognition platform, and when Q is equal to the total number of first models, constructing the first sub-processing result based on the Q clusters.

[0011] Furthermore, the third pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the first recognition result, and obtains a second sub-processing result including: extracting Q clusters from the first recognition result; inputting the Q clusters into the semantic vector model, and outputting R second semantic vectors, where R is a positive integer; obtaining the frequency weight of each second semantic vector, and determining the second semantic vector dimension value matrix according to each second semantic vector and the frequency weight of each second semantic vector; merging the R second semantic vector dimension value matrices using a clustering algorithm to obtain Y file clusters, where Y is a positive integer; obtaining the total number of second models of the candidate third recognition model stored in the text recognition platform, and when Y is equal to the total number of second models, constructing the second sub-processing result based on the Y file clusters.

[0012] Furthermore, matching the second recognition model according to the first sub-processing result includes: obtaining vector features of Q clusters associated with the first sub-processing result, and obtaining model parameters of multiple candidate second recognition models stored in the text recognition platform; matching the model parameters of each candidate second recognition model according to the vector features of the Q clusters to obtain the second recognition model.

[0013] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a file information identification system is provided. The system includes: a management platform, which is used to obtain the file to be identified and upload the file to be identified to the blockchain platform; the blockchain platform is used to receive the file to be identified and forward the file to be identified to the file classifier; the file classifier is used to obtain the model comparison rule when receiving the file to be identified, determine the target model from the model comparison rule according to the file type of the file to be identified, identify the file to be identified using the target model, obtain file identification information, and store the file identification information to the management platform through the data distribution protocol.

[0014] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a device for identifying file information is provided. The device includes: a receiving unit, which is used to receive a file to be identified sent by a blockchain platform, wherein the file to be identified is sent by a management platform to the blockchain platform; an acquisition unit, which is used to acquire a model comparison rule, and determine a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; and an identification unit, which is used to identify the file to be identified using the target model to obtain file identification information.

[0015] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute any of the above-mentioned file information identification methods.

[0016] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory stores an executable program, and the processor is used to run the program, wherein when the one or more programs are executed by one or more processors, the one or more processors implement any one of the above-mentioned file information recognition methods.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided. The computer program product includes a computer program, wherein when the computer program is executed by a processor, any one of the above-mentioned file information identification methods is implemented.

[0018] In an embodiment of the present application, a file information identification method is adopted, by receiving a file to be identified sent by a blockchain platform, wherein the file to be identified is sent by a management platform to the blockchain platform; a model comparison rule is obtained, and a target model is determined from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; the file to be identified is identified using the target model to obtain file identification information, thereby achieving the purpose of improving the accuracy of model selection, thereby achieving the technical effect of accurate identification and classification of files and improving file identification efficiency, and further solving the problem that when using a recognition model to identify files, the accuracy of selecting a recognition model is low, resulting in low accuracy and low efficiency in recognizing text information. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 It is a hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for identifying file information;

[0021] Figure 2 is a flowchart of a method for identifying file information provided in an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a system for identifying file information provided according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of an optional file information identification system provided according to an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of a device for identifying file information provided in an embodiment of the present application;

[0025] Figure 6 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set up between this system and the relevant user or organization. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0029] It should be noted that the collected information used in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse to use it.

[0030] Example 1

[0031] According to an embodiment of the present application, a method embodiment for identifying file information is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 It is a hardware structure diagram of a computer terminal (or mobile device) for implementing a method for identifying file information, such as Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more ( Figure 1102a, 102b, ..., 102n are used to illustrate) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU (Microcontroller Unit) or a programmable logic device FPGA (Field-Programmable Gate Array)), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a keyboard, a cursor control device, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0033] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0034] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the file information identification method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned file information identification method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0035] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC) and a network interface, which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0036] The display may be, for example, a touch screen liquid crystal display (LCD), which enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0037] Under the above operating environment, this application provides Figure 2 The identification method of the file information shown. Figure 2 is a flowchart of a method for identifying file information provided in an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0038] Step S201, receiving a file to be identified sent by the blockchain platform, wherein the file to be identified is sent by the management platform to the blockchain platform.

[0039] Specifically, when performing OCR (Optical Character Recognition) on a file to be recognized, in order to improve the recognition accuracy of the file, when the management platform receives the file to be recognized, the management platform can send the file to be recognized to the blockchain platform, and then the smart contract in the blockchain platform will send it to the file classifier, where the smart contract is an execution protocol based on blockchain technology, which includes rules and conditions for file processing and classification.

[0040] It should be noted that the smart contract includes multiple functional modules, such as batch triggers, online triggers and recognition result modules. The batch trigger is used to start when the files in the blockchain platform accumulate to a certain number or meet specific conditions, and then trigger the model training and batch recognition process; the online trigger is used to process newly uploaded files in real time and send them to the real-time call module for OCR recognition.

[0041] Step S202, obtaining a model comparison rule, and determining a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified.

[0042] Specifically, the model comparison rules are stored in the comparison rule module, and the comparison rule module receives query requests from the real-time call module or other components of the file classifier through a query interface. That is, after the file classifier receives the file to be identified, it sends a call request to the comparison rule module, and then extracts the model comparison rules from it. After using the pre-classification algorithm to identify the type of the file to be identified and obtaining the file type, the OCR recognition model that is most suitable as the target model is quickly and accurately selected from the model comparison rules based on the file type, so as to achieve efficient recognition and classification of the file content, wherein the file type may include the file's metadata (such as file name, file format, upload time, etc.), image features (such as layout, resolution, special tags, etc.) and contextual information (such as the project and department to which the file belongs).

[0043] Step S203: using the target model to identify the file to be identified, and obtaining file identification information.

[0044] Specifically, after determining the target model, the model is called through the cloud service interface to perform OCR recognition on the file to be identified to obtain initial file identification information, and then the information is preprocessed, such as data cleaning, structured conversion, etc., to ensure the accuracy and uniformity of the output results, and then obtain file identification information.

[0045] Furthermore, the file classifier can feed back the processed file identification information to the management platform through the data distribution service (DDS) for further data processing and business operations.

[0046] The file information identification method provided by the embodiment of the present application receives a file to be identified sent by a blockchain platform, wherein the file to be identified is sent by a management platform to the blockchain platform; obtains a model comparison rule, and determines a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; uses the target model to identify the file to be identified to obtain file identification information, thereby solving the problem in the related art that when using the recognition model to identify the file, the accuracy of the selected recognition model is low, resulting in low accuracy and low efficiency in the recognition of text information. By determining the target model for identifying the received file to be identified according to the model comparison rule, the target model is used to identify the file to be identified, thereby obtaining file identification information, thereby achieving the technical effect of accurate identification and classification of files and improving file identification efficiency.

[0047] In order to improve the accuracy of model selection, optionally, in the file information identification method provided in the embodiment of the present application, when the file classifier includes N pre-classifiers, N is a positive integer, and the model comparison rule is determined in the following manner: receiving M pre-trained information files sent by a batch trigger of the blockchain platform, wherein M is a positive integer; performing feature extraction on each pre-trained information file to obtain M file features; having the N pre-classifiers process each pre-trained information file in turn to obtain M processing results, wherein each processing result is used to characterize the correspondence between the file features of each pre-trained information file and a preset model; and constructing a model comparison rule based on the M processing results.

[0048] Before determining the target model of the file to be identified, it is necessary to obtain the model comparison rules from the comparison rule module, and the model comparison rules can be obtained through multiple pre-trained information files. Specifically, after the management platform uploads the pre-trained information file to the blockchain platform, the blockchain platform identifies the number of files of the pre-trained information file. When the number of pre-trained information files accumulates to the set file threshold, all files are sent to the file classifier through the batch trigger on the blockchain platform.

[0049] After the file classifier receives multiple pre-training information files, the original data and metadata of the files are used to train the file classifier. Each layer in the file classifier corresponds to a model at a different level, for example, from a vertical large model to a segmented model, and then to a further segmented model, gradually refining the granularity of file classification. Specifically, natural language processing and other technologies can be used to extract features from these pre-training information files to obtain multiple file features, where file features can include image features, text features, metadata features, etc. Image features can include file layout, fonts, colors, etc.; text features can include keywords, semantic structures, etc.; metadata features can include file type, format, upload time, and other information.

[0050] Furthermore, the multiple pre-classifiers included in the file classifier process the pre-training information file in turn according to each of the above-mentioned file features to obtain corresponding processing results. It should be noted that the above-mentioned multiple pre-classifiers can constitute a model tree, and each layer of pre-classifiers corresponds to a different model level. For example, when there are three pre-classifiers, each layer of pre-classifiers corresponds to a vertical large model, a subdivided model, and a further subdivided model. When each pre-classifier is processing, each layer can be reprocessed and aggregated based on the recognition results of the previous layer, and then call the next layer of more subdivided models for training, that is, each layer of pre-classifiers calls the corresponding preset model for OCR recognition, and after generating the recognition results, sends them to the next layer of pre-classifiers.

[0051] After the processing results are output based on multiple pre-classifiers, since the processing results not only include the recognized document information, but also include the matching relationship between the file features and the model, the corresponding relationship of the model can be established based on the processing results, thereby obtaining the model comparison rules. This embodiment uses progressive pre-classifier processing to ultimately generate multiple processing results, and then constructs model comparison rules based on the processing results, and then the most suitable OCR recognition model can be selected through the model comparison rules. The determination of the model comparison rules can not only improve the accuracy of OCR recognition, reduce the recognition errors caused by improper model selection, but also optimize the efficiency of file processing and resource utilization, providing strong technical support for the digital transformation of financial institutions.

[0052] Optionally, in the file information identification method provided in the embodiment of the present application, when the file classifier includes three pre-classifiers, N pre-classifiers process each pre-trained information file in turn to obtain M processing results including: for a pre-trained information file, obtaining the file type of the pre-trained information file, and determining a first recognition model according to the file features; the first pre-classifier in the file classifier calls the first recognition model to identify the pre-trained information file to obtain an identification message; the second pre-classifier in the file classifier calls the semantic vector model and the clustering algorithm in turn to process the identification message to obtain a first sub-processing result, and matches the second recognition model according to the first sub-processing result; the second pre-classifier calls the second recognition model to identify the first sub-processing result to obtain a first recognition result; the third pre-classifier in the file classifier calls the semantic vector model and the clustering algorithm in turn to process the first recognition result to obtain a second sub-processing result, and matches the third recognition model according to the second sub-processing result; the third pre-classifier calls the third recognition model to identify the second sub-processing result to obtain a second recognition result, and determines the processing result according to the third recognition model.

[0053] Specifically, when the file classifier includes three pre-classifiers (i.e., a first pre-classifier, a second pre-classifier, and a third pre-classifier), when processing each pre-trained information file, the semantic vector model and the clustering algorithm can be utilized by the above-mentioned multi-level pre-classifier, and different recognition models can be called at the same time to obtain the corresponding processing results, wherein the recognition accuracy of each pre-classifier is different, the recognition accuracy of the first pre-classifier is the lowest, the recognition accuracy of the second pre-classifier is medium, and the recognition accuracy of the third pre-classifier is the highest.

[0054] When processing each pre-training information file, a preliminary analysis of the file can be performed first, wherein the analysis method may include pre-processing, feature extraction, file name analysis, metadata inspection, etc., to extract the key features of the file, that is, to obtain the file type. Subsequently, a pre-classification algorithm (such as a machine learning model) can be called according to the file type to preliminarily classify the file into the first recognition model to which it belongs, to ensure that the file can be processed by the most appropriate model, wherein the first recognition model can be a "card recognition" recognition model, a "general form and text recognition" recognition model, a "bill recognition" recognition model, etc.

[0055] Furthermore, the first pre-classifier can call the first recognition model to perform OCR recognition on the pre-trained information file and generate a preliminary recognition result, and then pre-process and aggregate the result to obtain a recognition message, and use the recognition message as the input of the second pre-classifier to enter a more fine-grained classification and recognition process, wherein the recognition message may include text information, table data, labels in a specific format, etc. obtained by OCR recognition, which provides a basis for subsequent in-depth processing and classification, and can also significantly improve the processing speed.

[0056] After the second pre-classifier receives the identification message, it can call the semantic vector model and clustering algorithm to perform in-depth analysis on the message content, wherein the semantic vector model can convert the key information in the identification message into a vector representation, and the clustering algorithm classifies the information according to the similarity between the vectors, thereby generating a first sub-processing result, wherein the processing result includes the semantic structure of the file content and the clustering information of the key information. Then, based on the first sub-processing result, it can match a more accurate second identification model, such as a segmentation model for "product parameter description".

[0057] Furthermore, the second pre-classifier calls the second recognition model to recognize the first sub-processing result, and obtains the first recognition result based on the "segmentation model" to accurately recognize the file content. Then, the third pre-classifier uses the semantic vector model and clustering algorithm to recognize and cluster the information of the first recognition result again, and generates the second sub-processing result, wherein the second sub-processing result not only contains more accurate file content information, but also provides a basis for matching the third recognition model with the finest granularity. Then, the third recognition model can be matched based on the second sub-processing result, such as a proprietary model for financial products.

[0058] Finally, the third pre-classifier calls the third recognition model to perform final recognition on the second sub-processing result, obtains the second recognition result containing the finest-grained file information, and uses the third recognition model to determine the processing result of the file. This embodiment processes the pre-trained information file by using three pre-classifiers and a strategy of dynamically calling the recognition model, and can select the most suitable recognition model according to the complexity of the file content and specific needs, thereby significantly improving the accuracy and efficiency of OCR recognition, and avoiding the problem of low recognition efficiency that may occur when a single model processes a large number of heterogeneous files.

[0059] In order to further refine the classification of files so as to call a more specific recognition model for more accurate OCR recognition, optionally, in the file information recognition method provided in the embodiment of the present application, the second pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the recognition message, and obtains the first sub-processing result including: using regular expression disassembly rules to disassemble the recognition message to obtain L disassembly results, wherein each disassembly result includes multiple area sets, each area set includes multiple areas, each area is used to represent the information of vocabulary in the recognition message, each area set represents the information of sentences in the recognition message, and L is a positive integer; for one disassembly result, the disassembly result is input into the semantic vector model, and P first semantic vectors are output, wherein one first semantic vector is associated with one area, and P is a positive integer; for one disassembly result, the frequency weight of each area in the disassembly result is obtained, and the first semantic vector dimension value matrix is ​​determined according to the P first semantic vectors and the frequency weight of each area; the L first semantic vector dimension value matrices are merged using the clustering algorithm to obtain the first sub-processing result.

[0060] After the second preclassifier receives the identification message sent by the first preclassifier, deeper features can be extracted from the identification message to determine the more detailed model to which the file belongs. Specifically, in order to process different types of identification messages, the set regular expression disassembly rules can be used for information identification. Different types of identification messages can be stored in different control groups. At this time, the corresponding disassembly rules can be different, wherein the rules can accurately identify and separate different information units in the message, including the region of vocabulary information and the region set of sentence information. By inputting the identification message into the regular expression disassembly rule, and then disassembling it into a disassembly result containing multiple region sets, for example, the identification message can be disassembled into "region" and "region set" dimensions by a line break separator or a specific separator, \\n is an escaped line break separator, \\t is an escaped table separator, and the relationship between the region and the region set can be expressed as: identification message raw_sentences = ["region 1 region 2 region 3", "region 4 region 5"......], the double quotes are the region set n, and the regions are separated by spaces. It should be noted that the decomposed regions can represent and identify the independent information of the words in the message, while the region set can represent the complete meaning of the sentences in the message. The above hierarchical decomposition method helps to more accurately capture and understand the semantic structure in the message, and provides a more precise basis for subsequent information classification.

[0061] Furthermore, each decomposition result is input into the semantic vector model. The model captures the contextual semantics in the text according to the content of the "region set", and then converts each region into a three-dimensional or higher-dimensional first semantic vector, which can represent the semantic features of vocabulary and sentences in numerical form. After determining the frequency weight of each region by means of inverse document frequency statistics, a first semantic vector dimension value matrix is ​​constructed according to the first semantic vector generated by the decomposition result and the calculated frequency weight, wherein inverse document frequency means that the importance of a region is inversely proportional to the frequency of occurrence of the region in the document set. By allocating frequency weights, the relative importance of each region can be more accurately evaluated. Each row of the first semantic vector dimension value matrix represents the first semantic vector of a region, and each column represents the value of a specific semantic dimension. The frequency weight is used as the weighted value of the semantic vector to reflect the actual importance of the region in the document.

[0062] Further, clustering algorithms can be used for merging processing, and the similarity between regions or region sets can be determined by calculating the distance between vectors, and then the most similar regions or region sets can be gradually merged to form higher-level categories, that is, to obtain the first sub-processing result, wherein the clustering algorithm can be an agglomerative hierarchical clustering algorithm, and the first sub-processing result can be represented as a tree-like category structure, which not only includes the deep classification of regions and region sets, but also reflects the semantic association between each information unit. This application uses regular expression disassembly rules to perform deep processing on the recognition message, thereby generating a refined first sub-processing result, which can more deeply understand the semantic structure and information weight of the recognition message, improve the accuracy of parsing the document content, and provide a basis for the subsequent pre-classifier to call a more fine-grained OCR recognition model.

[0063] Optionally, in the file information recognition method provided in the embodiment of the present application, before using a clustering algorithm to merge L first semantic vector dimension value matrices to obtain a first sub-processing result, the method also includes: for a first semantic vector dimension value matrix, calculating the Euclidean distance between every two regions in the first semantic vector dimension value matrix to obtain K Euclidean distances, and calculating the sum of the K Euclidean distances to obtain a result difference value, where K is a positive integer; sorting the L result difference values ​​to obtain a difference value sorting; using a clustering algorithm to merge the L first semantic vector dimension value matrices to obtain the first sub-processing result includes: using a clustering algorithm to merge the L first semantic vector dimension value matrices according to the difference value sorting to obtain Q clusters, where Q is a positive integer; obtaining the total number of first models of the candidate second recognition models stored in the text recognition platform, and when Q is equal to the total number of first models, constructing the first sub-processing result based on the Q clusters.

[0064] Specifically, before using the clustering algorithm to merge the matrices, the Euclidean distance between any two regions (i.e., rows) in the first semantic vector dimension value matrix can be calculated by calculating the square root of the sum of the square differences of the corresponding dimensional values ​​of the two vectors, and multiple Euclidean distances can be obtained. This distance can accurately reflect the differences between the semantic vectors, and then these Euclidean distance values ​​are added to generate a result difference value that reflects the overall degree of difference between the regions in the matrix, which provides a key indicator for subsequent clustering analysis.

[0065] Furthermore, the result difference values ​​corresponding to each first semantic vector dimension value matrix are sorted to obtain a difference value sorting, and then the priority of cluster merging is determined by minimizing the difference within the region and maximizing the difference between regions. That is, the agglomerative hierarchical clustering algorithm can be used to merge the above-mentioned first semantic vector dimension value matrix according to the difference value sorting, and the most similar regions are gradually merged from each region until the preset conditions are met, wherein the merging process follows the principle of "merging the nearest cluster first". Multiple clusters can be obtained through the processing of the clustering algorithm. At this time, it is also necessary to compare the number of clusters with the total number of first models of the candidate second recognition model stored in the text recognition platform. If the two numbers are not equal, it is necessary to merge them again using the above merging method until the number of clusters is consistent with the number of models, and then the merged clusters are subjected to data cleaning and format conversion operations to form the first sub-processing result, which can ensure that each formed cluster matches a segmentation model for more in-depth OCR recognition. This embodiment uses Euclidean distance calculation, difference value sorting and clustering algorithm to merge the matrix. The generated first sub-processing result can not only improve the accuracy of information classification and model calling, but also accurately evaluate and classify the semantic features of information units, significantly improve the accuracy of information classification, and provide a clear classification basis for subsequent deep recognition.

[0066] Optionally, in the file information identification method provided in the embodiment of the present application, the third pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the first recognition result, and obtains the second sub-processing result including: extracting Q clusters from the first recognition result; inputting the Q clusters into the semantic vector model, and outputting R second semantic vectors, where R is a positive integer; obtaining the frequency weight of each second semantic vector, and determining the second semantic vector dimension value matrix according to each second semantic vector and the frequency weight of each second semantic vector; merging the R second semantic vector dimension value matrices using a clustering algorithm to obtain Y file clusters, where Y is a positive integer; obtaining the total number of second models of the candidate third recognition model stored in the text recognition platform, and when Y is equal to the total number of second models, constructing the second sub-processing result based on the Y file clusters.

[0067] Specifically, when the third pre-classifier processes the first recognition result output by the second pre-classifier, a plurality of clusters representing the information unit set after the preliminary classification of the file content can be extracted from the first recognition result. Then, the plurality of clusters are used as input and sent to the semantic vector model for processing, thereby outputting a second semantic vector that can more comprehensively reflect the semantic features of the information in the cluster.

[0068] The semantic vector difference of the above vectors can be further calculated, that is, the frequency weight of each second semantic vector is calculated based on the inverse document frequency method, and then the second semantic vector dimension value matrix is ​​constructed based on the second semantic vector and the calculated frequency weight, wherein each row of the matrix corresponds to a second semantic vector, each column represents a specific semantic dimension value, and the frequency weight is used as the weighting coefficient of each vector element in the matrix, reflecting the importance of the information unit in the document. Then the Euclidean distance between different matrices is calculated to determine the semantic difference, and then the second semantic vector dimension value matrix is ​​processed according to the difference value using a clustering algorithm, that is, based on the similarity between the second semantic vectors, multiple file clusters are obtained, wherein each file cluster can represent the classification result after further subdivision of the file content.

[0069] Furthermore, when the number of file clusters does not match the number of candidate third recognition models stored in the text recognition platform, it is necessary to merge them using the above-mentioned merging method until the number of the two is equal, and finally form the second sub-processing result based on the merged clusters, thereby providing an accurate classification basis for calling the third recognition model, ensuring that the subsequent recognition process can be more accurately targeted at specific types of files. This embodiment can perform multi-level in-depth understanding and classification of file content by invoking the semantic vector model and agglomerative hierarchical clustering algorithm for processing by the third pre-classifier, significantly improving the accuracy of OCR recognition, optimizing the efficiency of subsequent model selection and calling, reducing resource consumption, and enhancing the automation and intelligence of information processing, providing financial institutions with a more efficient and accurate means of file information management.

[0070] Optionally, in the file information recognition method provided in an embodiment of the present application, matching the second recognition model according to the first sub-processing result includes: obtaining vector features of Q clusters associated with the first sub-processing result, and obtaining model parameters of multiple candidate second recognition models stored in the text recognition platform; matching the model parameters of each candidate second recognition model according to the vector features of the Q clusters to obtain the second recognition model.

[0071] Specifically, in order to determine the second recognition model, firstly, the vector features of the cluster associated with the first sub-processing result can be obtained, which contain the semantic vector and frequency weight information of each information unit in the cluster, and are used to describe the comprehensive semantic attributes and importance of the cluster. Then, the model parameters of multiple candidate second recognition models are obtained from the text recognition platform, and these parameters include but are not limited to the training data set of the model, model type, model structure, hyperparameter settings, etc.

[0072] The vector features of the cluster can be further used to match the model parameters of each candidate second recognition model, for example, to calculate the similarity or compatibility between the cluster features and the model parameters; the potential performance of the model for cluster content recognition can also be evaluated by comparing the semantic vector distribution of the cluster with the semantic feature distribution of the model training data set. Finally, the optimal second recognition model is determined based on the matching results between the vector features of the cluster and the parameters of each candidate second recognition model. This embodiment selects the second recognition model that best suits the cluster features through intelligent matching. This model selection method based on feature matching significantly improves the accuracy of OCR recognition.

[0073] This application also provides Figure 3 The identification system of the file information shown. Figure 3 is a schematic diagram of a file information recognition system provided according to an embodiment of the present application, such as Figure 3 As shown, the system includes:

[0074] The management platform 301 is used to obtain the file to be identified and upload the file to be identified to the blockchain platform 302;

[0075] Specifically, the management platform 301 may be a financial institution operation and management platform 301 , which is capable of receiving products and documents waiting for identification, and then uploading the files to the blockchain platform 302 .

[0076] The blockchain platform 302 is used to receive the file to be identified and forward the file to be identified to the file classifier 303;

[0077] Specifically, the smart contract in the blockchain platform 302 can send the files to be identified into the file classifier 303, wherein the smart contract also includes multiple functional modules, such as batch triggers, online triggers and recognition result modules. The batch trigger can be activated when the files in the blockchain platform 302 accumulate to a certain number or meet specific conditions, and then trigger the model training and batch recognition process; the online trigger can process the newly uploaded files in real time and send them to the file classifier 303 for OCR recognition.

[0078] The file classifier 303 is used to obtain the model comparison rules when receiving the file to be identified, determine the target model from the model comparison rules according to the file type of the file to be identified, identify the file to be identified using the target model to obtain file identification information, and store the file identification information to the management platform 301 through the data distribution protocol.

[0079] Specifically, after the file classifier 303 receives the file to be identified, it extracts the model comparison rules and uses the pre-classification algorithm to identify the type of the file to be identified to obtain the file type. Then, based on the file type, the most suitable OCR recognition model as the target model is quickly and accurately selected from the model comparison rules to achieve efficient recognition and classification of the file content, wherein the file type may include the file's metadata (such as file name, file format, upload time, etc.), image features (such as layout, resolution, special marks, etc.) and context information (such as the project and department to which the file belongs).

[0080] After determining the target model, the file classifier 303 can call the target model through the cloud service interface to perform OCR recognition on the file to be recognized, obtain initial file recognition information, and then obtain file recognition information after preprocessing the information, such as data cleaning, structural conversion, etc., to ensure the accuracy and uniformity of the output results. Finally, the file recognition information is fed back to the management platform 301 through the data distribution service (Data Distribution Service: DDS) for further data processing and business operations. Among them, the DDS mechanism can ensure efficient and reliable data transmission, and maintain the real-time and integrity of the data even in the scenario of parallel processing of a large number of files.

[0081] The file information recognition system provided by the embodiment of the present application is used to obtain the file to be recognized through the management platform 301, and upload the file to be recognized to the blockchain platform 302; the blockchain platform 302 is used to receive the file to be recognized and forward the file to be recognized to the file classifier 303; the file classifier 303 is used to obtain the model comparison rule when receiving the file to be recognized, determine the target model from the model comparison rule according to the file type of the file to be recognized, use the target model to recognize the file to be recognized, and obtain file identification information, and is also used to store the file identification information to the management platform 301 through the data distribution protocol, which solves the problem that when using the recognition model to recognize the file in the related art, the accuracy of selecting the recognition model is low, resulting in low accuracy and low efficiency in recognizing text information. Through the interactive operation of the association platform 301, the blockchain platform 302 and the file classifier 303, the target model for identifying the received file to be recognized is determined according to the model comparison rule, and the file to be recognized is recognized using the target model, so as to obtain file identification information, thereby achieving the technical effect of accurately identifying and classifying files and improving file recognition efficiency.

[0082] The present application also provides a file information identification system. Figure 4 is a schematic diagram of an optional file information identification system provided according to an embodiment of the present application, such as Figure 4As shown, the system includes: a financial institution operation and management platform, a blockchain platform, a file pre-classification model trainer, and an optical character recognition platform (also known as a text recognition platform).

[0083] It should be noted that the recognition model for identifying the file information of financial files can be screened from the model comparison rules, and the information in the model comparison rules can be constructed based on multiple pre-trained financial files, that is, when the financial institution operation and management platform uploads the pre-trained financial files to the blockchain platform, the smart contract is activated through the digital safe in the blockchain platform, and the batch trigger in the smart contract monitors the number of unprocessed pre-trained financial files accumulated in the digital safe. When the number of files reaches the preset threshold, the batch trigger will package these files and upload them to the file pre-classification model trainer.

[0084] Furthermore, after receiving the above-mentioned file, since the file pre-classification model trainer includes three pre-classifiers (i.e., the first pre-classifier, the second pre-classifier and the third pre-classifier), multiple pre-classifiers can process the file in turn according to each of the above-mentioned file features to obtain corresponding processing results.

[0085] Specifically, each file can first be preliminarily classified using a pre-classification algorithm, and then the first recognition model associated with each file can be obtained, such as "card recognition", "general form and text recognition", and "bill recognition", thereby realizing the coarse-grained classification of the "vertical large model". Then, each file is further processed and aggregated according to the results output by the pre-classification algorithm: first, by setting differentiated regular expression rules, the recognition message output by the first recognition model is disassembled into "region" and "region set" dimensions, and then the disassembled results are input into the semantic vector model for calculation, and then the frequency weight of the calculated semantic vector is determined, and the semantic vector dimension value matrix is ​​constructed from the frequency weight and the semantic vector. Then, the above matrix is ​​merged using a clustering algorithm to obtain a preliminary classification result, that is, the first sub-processing result, and the second recognition model is determined using this result.

[0086] Furthermore, the file pre-classification model trainer calls the second recognition model to recognize the first sub-processing result, and obtains the first recognition result based on the "segmentation model" to accurately recognize the file content. The first recognition result is then processed again to generate the second sub-processing result. Finally, the third recognition model can be matched based on the second sub-processing result. Finally, the model comparison rules can be constructed based on the third recognition model and stored in the optical character recognition platform.

[0087] When the digital safe of the financial institution's management platform receives the real-time uploaded financial files, the online trigger in the smart contract sends the files to the real-time call module of the file pre-classification model trainer. The real-time call module can select the recognition model with the highest recognition accuracy from the model comparison rules for file recognition based on the image features, metadata and other information of the financial files. For example, for a fund product manual, the proprietary model of "Company A B Index Fund" can be called for in-depth recognition to obtain specific fund code, issue date, issue price and other information.

[0088] After the recognition model has finished recognizing the financial documents, the generated recognition results can be transmitted back to the financial institution's business management platform through the data distribution service for subsequent automatic filling, verification, statistics and other operations, thereby ensuring the real-time update and secure transmission of data, and providing the platform with structured and accurate file information. This embodiment not only improves the recognition efficiency and accuracy by recognizing and intelligently classifying financial documents, but also reduces the possible problems caused by manual intervention and reduces the error rate. At the same time, the real-time recognition results and data security are ensured through the data distribution service, providing strong support for the digital transformation of financial institutions.

[0089] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0090] Example 2

[0091] The present application also provides a file information recognition device. It should be noted that the file information recognition device of the present application can be used to execute the file information recognition method provided by the present application. The file information recognition device provided by the present application is introduced below.

[0092] According to an embodiment of the present application, a device for implementing the above-mentioned file information identification method is also provided. Figure 5 is a schematic diagram of a device for identifying file information provided in an embodiment of the present application, such as Figure 5 As shown, the device includes: a receiving unit 50, an acquiring unit 51, and an identifying unit 52.

[0093] A receiving unit 50 is used to receive a file to be identified sent by the blockchain platform, wherein the file to be identified is sent by the management platform to the blockchain platform;

[0094] An acquisition unit 51 is used to acquire a model comparison rule, and determine a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence relationship between multiple file types and preset models, and the preset model is used to identify document information of the file to be identified;

[0095] The identification unit 52 is used to identify the file to be identified by using the target model to obtain file identification information.

[0096] The file information identification device provided in the embodiment of the present application receives the file to be identified sent by the blockchain platform through the receiving unit 50, wherein the file to be identified is sent by the management platform to the blockchain platform; the acquisition unit 51 acquires the model comparison rule, and determines the target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes the correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; the identification unit 52 uses the target model to identify the file to be identified and obtains file identification information, which solves the problem that when using the recognition model to identify the file in the related art, the accuracy of selecting the recognition model is low, resulting in low accuracy and low efficiency in identifying the text information. By determining the target model for identifying the received file to be identified according to the model comparison rule, the target model is used to identify the file to be identified, thereby obtaining the file identification information, thereby achieving the technical effect of accurately identifying and classifying the file and improving the efficiency of file identification.

[0097] Optionally, in the file information identification device provided in the embodiment of the present application, the acquisition unit 51 includes: a receiving module, used to receive M pre-trained information files sent by a batch trigger of the blockchain platform, where M is a positive integer; a first extraction module, used to extract features from each pre-trained information file to obtain M file features; a processing module, used to process each pre-trained information file in turn by N pre-classifiers to obtain M processing results, wherein each processing result is used to characterize the correspondence between the file features of each pre-trained information file and a preset model; a construction module, used to construct a model comparison rule according to the M processing results.

[0098] Optionally, in the file information identification device provided in the embodiment of the present application, the acquisition unit 51 includes: a first acquisition module, which is used to obtain the file type of a pre-trained information file for a pre-trained information file, and determine the first recognition model according to the file characteristics; a first calling module, which is used by the first pre-classifier in the file classifier to call the first recognition model to identify the pre-trained information file and obtain an identification message; the first calling module is used by the second pre-classifier in the file classifier to call the semantic vector model and the clustering algorithm in sequence to process the identification message to obtain a first sub-processing result, and match the second recognition model according to the first sub-processing result; the first recognition module is used by the second pre-classifier to call the second recognition model to identify the first sub-processing result to obtain a first recognition result; the second calling module is used by the third pre-classifier in the file classifier to call the semantic vector model and the clustering algorithm in sequence to process the first recognition result to obtain a second sub-processing result, and match the third recognition model according to the second sub-processing result; the second recognition module is used by the third pre-classifier to call the third recognition model to identify the second sub-processing result to obtain a second recognition result, and determine the processing result according to the third recognition model.

[0099] Optionally, in the file information recognition device provided in the embodiment of the present application, the acquisition unit 51 includes: a disassembly module, which is used to disassemble the recognition message using regular expression disassembly rules to obtain L disassembly results, wherein each disassembly result includes multiple area sets, each area set includes multiple areas, each area is used to represent the information of the vocabulary in the recognition message, each area set represents the information of the sentence in the recognition message, and L is a positive integer; a first input module, which is used to input the disassembly result into a semantic vector model for one disassembly result, and output P first semantic vectors, wherein one first semantic vector is associated with one area, and P is a positive integer; a second acquisition module, which is used to obtain the frequency weight of each area in the disassembly result for one disassembly result, and determine the first semantic vector dimension value matrix according to the P first semantic vectors and the frequency weight of each area; a first merging module, which is used to merge the L first semantic vector dimension value matrices using a clustering algorithm to obtain a first sub-processing result.

[0100] Optionally, in the file information recognition device provided in the embodiment of the present application, the device includes: a calculation unit, which is used to calculate the Euclidean distance between each two regions in a first semantic vector dimensional value matrix for a first semantic vector dimensional value matrix before merging L first semantic vector dimensional value matrices using a clustering algorithm to obtain a first sub-processing result, and calculate the sum of the K Euclidean distances to obtain a result difference value, wherein K is a positive integer; a sorting unit, which is used to sort the L result difference values ​​to obtain a difference value sorting; the acquisition unit 51 includes: a second merging module, which is used to merge the L first semantic vector dimensional value matrices according to the difference value sorting by a clustering algorithm to obtain Q clusters, wherein Q is a positive integer; a third acquisition module, which is used to obtain the total number of first models of the candidate second recognition models stored in the text recognition platform, and when Q is equal to the total number of first models, the first sub-processing result is constructed based on the Q clusters.

[0101] Optionally, in the file information identification device provided in the embodiment of the present application, the acquisition unit 51 includes: a second extraction module, used to extract Q clusters from the first recognition result; a second input module, used to input the Q clusters into the semantic vector model, and output R second semantic vectors, where R is a positive integer; a fourth acquisition module, used to obtain the frequency weight of each second semantic vector, and determine the second semantic vector dimension value matrix according to each second semantic vector and the frequency weight of each second semantic vector; a second merging module, used to merge the R second semantic vector dimension value matrices using a clustering algorithm to obtain Y file clusters, where Y is a positive integer; a fifth acquisition module, used to obtain the total number of second models of the candidate third recognition model stored in the text recognition platform, and when Y is equal to the total number of second models, a second sub-processing result is constructed based on the Y file clusters.

[0102] Optionally, in the file information recognition device provided in the embodiment of the present application, the acquisition unit 51 includes: a sixth acquisition module, used to obtain the vector features of Q clusters associated with the first sub-processing result, and obtain the model parameters of multiple candidate second recognition models stored in the text recognition platform; a matching module, used to match the model parameters of each candidate second recognition model according to the vector features of the Q clusters to obtain a second recognition model.

[0103] It should be noted that the above-mentioned receiving unit 50, acquiring unit 51, and identifying unit 52 correspond to steps S201 to S203 in Example 1, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n), and the above-mentioned units can also be part of the device and can be run in the computer terminal 10 provided in Example 1.

[0104] Example 3

[0105] The embodiment of the present application may provide a computer terminal, which may be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal may also be replaced by a terminal device such as a mobile terminal or an electronic device.

[0106] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of the computer network.

[0107] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the file information identification method: receiving the file to be identified sent by the blockchain platform, wherein the file to be identified is sent by the management platform to the blockchain platform; obtaining the model comparison rule, and determining the target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes the correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; using the target model to identify the file to be identified to obtain file identification information.

[0108] Optionally, the computer terminal may execute the following program codes of the following steps in the file information identification method: receiving M pre-trained information files sent by a batch trigger of the blockchain platform, where M is a positive integer; performing feature extraction on each pre-trained information file to obtain M file features; having N pre-classifiers process each pre-trained information file in turn to obtain M processing results, where each processing result is used to characterize the correspondence between the file features of each pre-trained information file and a preset model; and constructing a model comparison rule based on the M processing results.

[0109] Optionally, the above-mentioned computer terminal can execute the program code of the following steps in the file information identification method: for a pre-trained information file, obtain the file type of the pre-trained information file, and determine the first identification model according to the file characteristics; the first pre-classifier in the file classifier calls the first identification model to identify the pre-trained information file to obtain an identification message; the second pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the identification message to obtain a first sub-processing result, and matches the second identification model according to the first sub-processing result; the second pre-classifier calls the second identification model to identify the first sub-processing result to obtain a first identification result; the third pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the first identification result to obtain a second sub-processing result, and matches the third identification model according to the second sub-processing result; the third pre-classifier calls the third identification model to identify the second sub-processing result to obtain a second identification result, and determines the processing result according to the third identification model.

[0110] Optionally, the computer terminal can execute the program code of the following steps in the file information identification method: using regular expression disassembly rules to disassemble the identification message to obtain L disassembly results, wherein each disassembly result includes multiple area sets, each area set includes multiple areas, each area is used to represent the information of the vocabulary in the identification message, each area set represents the information of the sentence in the identification message, and L is a positive integer; for one disassembly result, input the disassembly result into a semantic vector model, and output P first semantic vectors, wherein one first semantic vector is associated with one area, and P is a positive integer; for one disassembly result, obtain the frequency weight of each area in the disassembly result, and determine the first semantic vector dimension value matrix according to the P first semantic vectors and the frequency weight of each area; use a clustering algorithm to merge the L first semantic vector dimension value matrices to obtain a first sub-processing result.

[0111] Optionally, the computer terminal can execute the following program codes of the following steps in the file information recognition method: for a first semantic vector dimension value matrix, calculate the Euclidean distance between every two regions in the first semantic vector dimension value matrix to obtain K Euclidean distances, and calculate the sum of the K Euclidean distances to obtain a result difference value, where K is a positive integer; sort the L result difference values ​​to obtain a difference value sorting; use a clustering algorithm to merge the L first semantic vector dimension value matrices to obtain a first sub-processing result including: using a clustering algorithm to merge the L first semantic vector dimension value matrices according to the difference value sorting to obtain Q clusters, where Q is a positive integer; obtain the total number of first models of the candidate second recognition models stored in the text recognition platform, and when Q is equal to the total number of first models, construct the first sub-processing result based on the Q clusters.

[0112] Optionally, the computer terminal can execute the following program codes of the following steps in the file information identification method: extract Q clusters from the first recognition result; input the Q clusters into the semantic vector model, and output R second semantic vectors, where R is a positive integer; obtain the frequency weight of each second semantic vector, and determine the second semantic vector dimension value matrix according to each second semantic vector and the frequency weight of each second semantic vector; use a clustering algorithm to merge the R second semantic vector dimension value matrices to obtain Y file clusters, where Y is a positive integer; obtain the total number of second models of the candidate third recognition model stored in the text recognition platform, and when Y is equal to the total number of second models, construct a second sub-processing result based on the Y file clusters.

[0113] Optionally, the above-mentioned computer terminal can execute the program code of the following steps in the file information recognition method: obtaining the vector features of Q clusters associated with the first sub-processing results, and obtaining the model parameters of multiple candidate second recognition models stored in the text recognition platform; matching the model parameters of each candidate second recognition model according to the vector features of the Q clusters to obtain the second recognition model.

[0114] Optionally, Figure 6 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (only one is shown) processor 602, memory 604, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0115] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the file information identification method and device in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned file information identification method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0116] The processor can call the information and application programs stored in the memory through the transmission device to execute the above steps in the above file information identification method.

[0117] By adopting the embodiment of the present application, a solution for identifying file information is provided. By receiving the file to be identified sent by the blockchain platform, wherein the file to be identified is sent by the management platform to the blockchain platform; obtaining the model comparison rule, and determining the target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes the correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; using the target model to identify the file to be identified, and obtaining the file identification information, the purpose of improving the accuracy of model selection is achieved, and then the problem of low accuracy of selecting the recognition model when using the recognition model to identify the file is solved, resulting in low accuracy and low efficiency of recognizing text information.

[0118] It can be understood by those skilled in the art that Figure 6 The structure shown is for illustration only, and the electronic device may also be a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), a PAD or other terminal device. Figure 6 The structure of the electronic device is not limited. Figure 6 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 6 Different configurations shown.

[0119] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0120] Example 4

[0121] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the file information identification method provided in the first embodiment.

[0122] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0123] Optionally, in this embodiment, the storage medium is configured to store program codes for executing the following steps: receiving a file to be identified sent by the blockchain platform, wherein the file to be identified is sent by the management platform to the blockchain platform; obtaining a model comparison rule, and determining a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify document information of the file to be identified; and identifying the file to be identified using the target model to obtain file identification information.

[0124] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the method for identifying file information.

[0125] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0126] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0128] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0130] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0131] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying file information, characterized in that: Applied to file classifiers, including: Receiving a file to be identified sent by a blockchain platform, wherein the file to be identified is sent by a management platform to the blockchain platform; Obtaining a model comparison rule, and determining a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence between multiple file types and preset models, and the preset model is used to identify the document information of the file to be identified; The target model is used to identify the file to be identified to obtain file identification information.

2. The method according to claim 1, characterized in that In the case where the file classifier includes N pre-classifiers, N is a positive integer, and the model comparison rule is determined in the following manner: Receiving M pre-training information files sent by a batch trigger of the blockchain platform, where M is a positive integer; Perform feature extraction on each pre-training information file to obtain M file features; The N pre-classifiers process each pre-training information file in turn to obtain M processing results, wherein each processing result is used to characterize the corresponding relationship between the file feature of each pre-training information file and a preset model; The model comparison rules are formed according to the M processing results.

3. The method according to claim 2, characterized in that In the case where the file classifier includes three pre-classifiers, the N pre-classifiers process each pre-training information file in turn, and obtain M processing results including: For a pre-training information file, obtaining a file type of the pre-training information file, and determining a first recognition model according to the file features; The first pre-classifier in the file classifier calls the first recognition model to recognize the pre-trained information file to obtain a recognition message; The second pre-classifier in the file classifier sequentially calls a semantic vector model and a clustering algorithm to process the recognition message to obtain a first sub-processing result, and matches a second recognition model according to the first sub-processing result; The second pre-classifier calls the second recognition model to recognize the first sub-processing result to obtain a first recognition result; The third pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the first recognition result to obtain a second sub-processing result, and matches a third recognition model according to the second sub-processing result; The third pre-classifier calls the third recognition model to recognize the second sub-processing result to obtain a second recognition result, and determines the processing result according to the third recognition model.

4. The method according to claim 3, characterized in that The second pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the identified message, and obtains the first sub-processing result including: Decomposing the recognition message using a regular expression decomposition rule to obtain L decomposition results, wherein each decomposition result includes a plurality of region sets, each region set includes a plurality of regions, each region is used to represent information of a vocabulary in the recognition message, each region set represents information of a sentence in the recognition message, and L is a positive integer; For a disassembly result, input the disassembly result into the semantic vector model, and output P first semantic vectors, wherein one first semantic vector is associated with one region, and P is a positive integer; For a disassembly result, obtain the frequency weight of each region in the disassembly result, and determine the first semantic vector dimension value matrix according to the P first semantic vectors and the frequency weight of each region; The clustering algorithm is used to merge the L first semantic vector dimension value matrices to obtain the first sub-processing result.

5. The method according to claim 4, characterized in that Before merging the L first semantic vector dimension value matrices by using the clustering algorithm to obtain the first sub-processing result, the method further includes: For a first semantic vector dimension value matrix, calculate the Euclidean distance between every two regions in the first semantic vector dimension value matrix to obtain K Euclidean distances, and calculate the sum of the K Euclidean distances to obtain a result difference value, where K is a positive integer; Sort the L result difference values ​​to obtain the difference value sorting; The clustering algorithm is used to merge the L first semantic vector dimension value matrices to obtain the first sub-processing result, which includes: The clustering algorithm merges the L first semantic vector dimension value matrices according to the difference value sorting to obtain Q clusters, where Q is a positive integer; The total number of first models of candidate second recognition models stored in the text recognition platform is obtained, and when Q is equal to the total number of the first models, the first sub-processing result is constructed based on the Q clusters.

6. The method according to claim 3, characterized in that The third pre-classifier in the file classifier sequentially calls the semantic vector model and the clustering algorithm to process the first recognition result, and obtains a second sub-processing result including: Extracting Q clusters from the first recognition result; Input the Q clusters into the semantic vector model, and output R second semantic vectors, where R is a positive integer; Obtaining a frequency weight of each second semantic vector, and determining a second semantic vector dimension value matrix according to each second semantic vector and the frequency weight of each second semantic vector; Using the clustering algorithm, the R second semantic vector dimension value matrices are merged to obtain Y file clusters, where Y is a positive integer; The total number of second models of the candidate third recognition models stored in the text recognition platform is obtained, and when Y is equal to the total number of the second models, the second sub-processing result is constructed based on the Y file clusters.

7. The method according to claim 3, characterized in that Matching the second recognition model according to the first sub-processing result includes: Obtaining vector features of Q clusters associated with the first sub-processing result, and obtaining model parameters of multiple candidate second recognition models stored in the text recognition platform; The model parameters of each candidate second recognition model are matched according to the vector features of the Q clusters to obtain the second recognition model.

8. A file information recognition system, characterized in that: include: The management platform is used to obtain the files to be identified and upload them to the blockchain platform; The blockchain platform is used to receive the file to be identified and forward the file to be identified to a file classifier; The file classifier is used to obtain model comparison rules when receiving the file to be identified, determine the target model from the model comparison rules according to the file type of the file to be identified, use the target model to identify the file to be identified to obtain file identification information, and store the file identification information to the management platform through a data distribution protocol.

9. A device for identifying file information, characterized in that: include: A receiving unit, configured to receive a file to be identified sent by a blockchain platform, wherein the file to be identified is sent by a management platform to the blockchain platform; An acquisition unit, used to acquire a model comparison rule, and determine a target model from the model comparison rule according to the file type of the file to be identified, wherein the model comparison rule includes a correspondence relationship between a plurality of file types and a preset model, and the preset model is used to identify the document information of the file to be identified; The recognition unit is used to use the target model to recognize the file to be recognized to obtain file recognition information.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the computer-readable storage medium is located is controlled to execute the file information recognition method according to any one of claims 1 to 7.

11. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the file information recognition method described in any one of claims 1 to 7.

12. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the file information identification method described in any one of claims 1 to 7 are implemented.

Images