Manuscript association method and device, equipment, storage medium and program product

Abstract

The invention provides a manuscript association method and device, equipment, a storage medium and a program product. The manuscript association method comprises the steps that at least one second manuscript similar to a first manuscript is determined from manuscripts to be screened; splicing the first manuscript and the at least one second manuscript to obtain at least one spliced text; wherein one spliced text is obtained by splicing one first manuscript and one second manuscript; inputting the at least one spliced text into a classification model to obtain classification results corresponding to the at least one spliced text; wherein the classification model is used for predicting relevancy between manuscripts contained in the spliced text; and determining a target manuscript related to the first manuscript from the at least one second manuscript according to the classification result corresponding to the at least one spliced text. According to the invention, the efficiency of establishing the association relationship between manuscripts can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a manuscript association method, device, equipment, storage medium and program product. Background Art

[0002] During the due diligence process for a company's listing, a walk-through test is an audit technique used to verify the process of a transaction from its original occurrence to its final recording in the financial statements. By checking the transaction documents between the proposed listed company and its upstream and downstream companies, the authenticity of the transaction can be verified, regulatory requirements can be met, and a solid foundation can be laid for the company's successful listing.

[0003] Currently, the relationship between manuscripts is usually constructed manually, and the key information in the manuscripts is manually identified, and then the key information in the manuscripts with related relationships is verified.

[0004] Manually constructing the relationship between working papers requires business personnel to find and determine the relationship between working papers in a large number of working papers according to different walkthrough test scenarios and the actual situation of each enterprise. This not only tests the professional ability of business personnel, but also requires them to have an in-depth understanding of the actual operation of the enterprise. It is a time-consuming and laborious task, and it is prone to errors, resulting in high labor costs and difficulty in traceability. Therefore, manually constructing the relationship between working papers leads to low efficiency in constructing the relationship between working papers. Summary of the invention

[0005] The present application provides a manuscript association method, device, equipment, storage medium and program product to solve the problem of low efficiency in establishing association relationships between manuscripts.

[0006] According to a first aspect of an embodiment of the present application, a manuscript association method is provided, comprising:

[0007] Determine at least one second manuscript similar to the first manuscript from the manuscripts to be screened;

[0008] Splicing the first draft and at least one of the second drafts to obtain at least one spliced ​​text; wherein one of the spliced ​​texts is obtained by splicing one of the first draft and one of the second draft;

[0009] Inputting at least one of the spliced ​​texts into a classification model to obtain a classification result corresponding to at least one of the spliced ​​texts; wherein the classification model is used to predict the relevance between the manuscripts contained in the spliced ​​text;

[0010] According to the classification result corresponding to at least one of the spliced ​​texts, a target manuscript related to the first manuscript is determined from at least one of the second manuscripts.

[0011] Optionally, the process of the classification model obtaining the classification result corresponding to at least one of the concatenated texts comprises:

[0012] Generate at least one third vector of the concatenated text;

[0013] According to at least one of the third vectors, a probability that the first draft is associated with at least one of the second drafts is determined, and a classification result is generated based on the probability.

[0014] Optionally, determining at least one second draft similar to the first draft from the drafts to be screened includes:

[0015] Inputting the first draft and the draft to be screened into the retrieval model to obtain the similarity between the first draft and the draft to be screened;

[0016] Based on the similarity between the first draft and the draft to be screened, at least one second draft similar to the first draft is determined from the drafts to be screened.

[0017] Optionally, the process of obtaining the similarity between the first manuscript and the manuscript to be screened by the retrieval model includes:

[0018] Generate a first vector of the first manuscript and a second vector of the manuscript to be screened;

[0019] A similarity between the first vector and the second vector is calculated.

[0020] Optionally, the classification model is trained based on sample data, and the sample data includes a first sample spliced ​​text and a second sample spliced ​​text, the first sample spliced ​​text includes the spliced ​​text of a sample manuscript and manuscripts unrelated to the sample manuscript, and the second sample spliced ​​text includes the spliced ​​text of the sample manuscript and manuscripts related to the sample manuscript.

[0021] Optionally, the method further comprises:

[0022] Determine at least one similar sample manuscript to the sample manuscript from the sample manuscripts to be screened;

[0023] From at least one of the similar sample manuscripts, filter out irrelevant manuscripts of the sample manuscript, and filter out relevant manuscripts of the sample manuscript;

[0024] splicing the sample manuscript and irrelevant manuscripts of the sample manuscript to obtain the first sample spliced ​​text;

[0025] The sample manuscript and the related manuscripts of the sample manuscript are spliced ​​together to obtain the second sample spliced ​​text.

[0026] Optionally, screening out irrelevant drafts of the sample draft from at least one of the similar sample drafts includes:

[0027] Calculating the number of overlapping characters between the sample manuscript and at least one similar sample manuscript;

[0028] According to the number of overlapping characters, irrelevant drafts of the sample draft are screened out from at least one similar sample draft.

[0029] Optionally, the method further comprises:

[0030] Determining the category of the first manuscript and determining the category of the target manuscript;

[0031] Extracting key information from the first draft according to the category of the first draft;

[0032] Extracting key information from the target manuscript according to the category of the target manuscript;

[0033] The key information in the first draft is compared with the key information in the target draft to obtain a comparison result, wherein the comparison result includes a relationship between the key information in the first draft and the key information in the target draft.

[0034] Optionally, determining the category of the first manuscript includes:

[0035] Determining whether the first draft satisfies a preset rule;

[0036] If the first draft satisfies the preset rule, generating a category of the first draft according to the preset rule;

[0037] If the first draft does not satisfy the preset rules, a first prompt word is generated based on the first draft, and the first prompt word is input into the large model to obtain the category of the first draft output by the large model, wherein the first prompt word is used to indicate the category of the first draft generated by the large model.

[0038] Optionally, extracting key information from the first draft according to the category of the first draft includes:

[0039] According to the category of the first draft, key parameters to be extracted are obtained, and definitions of the key parameters are acquired;

[0040] Generate a second prompt word based on the first draft, the key parameter and the definition of the key parameter;

[0041] The second prompt word is input into the large model to obtain key information in the first draft output by the large model, wherein the second prompt word is used to instruct the large model to extract key information in the first draft, and the key information includes the key parameters and the values ​​of the key parameters.

[0042] According to a second aspect of an embodiment of the present application, a manuscript associating device is provided, comprising:

[0043] A screening unit, used to determine at least one second draft similar to the first draft from the drafts to be screened;

[0044] A splicing unit, used for splicing the first draft and at least one of the second drafts to obtain at least one spliced ​​text; wherein one of the spliced ​​texts is obtained by splicing one of the first draft and one of the second draft;

[0045] A classification unit, used for inputting at least one of the spliced ​​texts into a classification model to obtain a classification result corresponding to at least one of the spliced ​​texts; wherein the classification model is used for predicting the relevance between the manuscripts contained in the spliced ​​texts;

[0046] A processing unit is used to determine a target manuscript related to the first manuscript from at least one of the second manuscripts according to the classification results corresponding to at least one of the spliced ​​texts.

[0047] According to a third aspect of an embodiment of the present application, there is provided an electronic device, including a memory and a processor;

[0048] The memory is connected to the processor and is used to store programs;

[0049] The processor is used to implement the manuscript association method described in the first aspect by running the program in the memory.

[0050] According to a fourth aspect of an embodiment of the present application, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the manuscript association method described in the first aspect is implemented.

[0051] According to a fifth aspect of an embodiment of the present application, a computer program product is provided, comprising computer program instructions, which, when executed by a processor, enable the processor to execute the manuscript association method described in the first aspect.

[0052] In the present application, at least one second draft similar to the first draft is determined from the drafts to be screened, and a target draft related to the first draft is determined from the at least one second draft. The first draft and the target draft are drafts with an associated relationship, and the drafts are screened twice. In the second screening process, the first draft and the at least one second draft are spliced ​​to obtain at least one spliced ​​text, wherein a spliced ​​text is obtained by splicing a first draft and a second draft. The at least one spliced ​​text is input into the classification model to obtain the classification results corresponding to the at least one spliced ​​text. The classification model is used to predict the correlation between the drafts contained in the spliced ​​text, and then according to the classification results corresponding to the at least one spliced ​​text, the target draft related to the first draft is determined from the at least one second draft. The second screening of the drafts is achieved through the classification model, and the associated relationship between the drafts is automatically constructed. Compared with the manually constructed associated relationship between the drafts, the efficiency of constructing the associated relationship between the drafts is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0054] Figure 1 A schematic diagram of a process of manually checking manuscript information in the prior art;

[0055] Figure 2 A schematic diagram of a process flow of a manuscript association method provided in an embodiment of the present application;

[0056] Figure 3 A schematic diagram of the structure of a dual retrieval model provided in an embodiment of the present application;

[0057] Figure 4 A schematic diagram of the association relationship between manuscripts provided in an embodiment of the present application;

[0058] Figure 5 A schematic diagram of a process flow of a manuscript association method provided in an embodiment of the present application;

[0059] Figure 6 A schematic diagram of a flow chart for determining a category of a first draft provided in an embodiment of the present application;

[0060] Figure 7 A schematic diagram of preset rules for a draft of a category "supplier list" provided in an embodiment of the present application;

[0061] Figure 8 A schematic diagram of a first prompt word provided in an embodiment of the present application;

[0062] Fig. 9 A schematic diagram of a process of step 502 provided in an embodiment of the present application;

[0063] Fig.10 A schematic diagram of a second prompt word provided in an embodiment of the present application;

[0064] Fig.11 A schematic diagram of a third prompt word provided in an embodiment of the present application;

[0065] Fig.12 A schematic diagram of the structure of a manuscript association device provided in an embodiment of the present application;

[0066] Fig.13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0067] During the due diligence process for a company's listing, a walk-through test is an audit technique used to verify the processing of a transaction from its original occurrence to its final recording in the financial statements. It is of great significance to construct the relationship between the upstream and downstream working papers of the proposed listed company. The upstream and downstream related working papers usually include purchase orders, invoices, transportation documents, etc. By checking the transaction working papers between the proposed listed company and the upstream and downstream companies, the authenticity of the transaction can be verified, ensuring that the company has not fabricated transactions or made fraudulent accounting, helping to verify the company's revenue, confirm whether the costs are accurate, discover possible risks, meet regulatory requirements, ensure the effective operation of internal controls and the accuracy of financial reports, and lay a solid foundation for the company's successful listing.

[0068] Although establishing the relationship between working papers is crucial for a company's due diligence before going public, the process of establishing the relationship between working papers currently relies mainly on manual operations. Currently, the relationship between working papers is usually established manually, and the key information in the working papers is manually identified, and then the key information in the working papers with related relationships is verified.

[0069] like Figure 1 The figure shows a flow chart of manually checking manuscript information in the prior art. Figure 1 In the process of manually checking the manuscript information, the process mainly includes:

[0070] (1) Manually sort out the manuscript categories.

[0071] When performing walkthrough testing, staff are faced with a difficult task: from a large number of drafts, according to different walkthrough test scenarios, accurately sort out the draft categories and extract the drafts corresponding to the walkthrough test scenarios.

[0072] For example, in the procurement walkthrough test scenario, it is necessary to filter out the documents such as purchase requisitions, purchase orders, warehouse receipts, invoices, accounting vouchers, and payment documents from the massive amount of documents. In the sales walkthrough test scenario, it is necessary to identify and separate the documents such as sales contracts, sales orders, warehouse delivery orders, logistics orders, revenue confirmation documents, invoices, accounting vouchers, receipt vouchers, and logistics settlement documents from the massive amount of documents. In the confirmation test scenario, it is necessary to accurately select the documents such as the scanned copies of the outgoing letters, the express delivery slips of the outgoing letters, the scanned copies of the reply letters, the express delivery slips of the reply letters, the employee roster, the list of related parties, and the industrial and commercial verification documents of customers and suppliers from the numerous documents.

[0073] The sorted working papers are then uploaded to the information management system for systematic and classified management. The process of sorting working papers not only requires staff to have a high degree of concentration and meticulous discrimination, but also requires them to have a deep understanding of the content and structure of the working papers to ensure that each working paper can be correctly classified, laying a solid foundation for subsequent audit and analysis work.

[0074] (2) Manually record key information of the manuscript.

[0075] When performing a walkthrough test, it is necessary to focus on specific key information (i.e. key elements) for various draft information. Taking the procurement walkthrough test scenario as an example, when processing the draft of the warehouse entry form, special attention should be paid to key information such as "warehouse entry date", "warehouse entry variety", "warehouse entry quantity" and "warehouse entry amount". The formats of warehouse entry forms vary from company to company, and the marking methods for "warehouse entry date" are also different. Some companies mark it as "arrival date", while others may use the expression "incoming information". Although the expressions are different, the core meaning is the same, referring to the time point when the goods arrive at the designated location. Business personnel need to extract and pay attention to these key information from various drafts according to the requirements of different business scenarios. Subsequently, this key information is entered into the company's information management system to achieve unified management and analysis of data to support business process monitoring and decision making.

[0076] (3) Manually organize the manuscripts into a large relational table.

[0077] Through the previous two steps, the information management system contains the classification information and key information of the working papers. It is necessary to conduct a detailed comparative analysis of the key information in the working papers based on the specific conditions of each enterprise and in combination with different walkthrough test scenario processes. At present, most investment banks manually sort out the associations between the working papers, and then store the associations between the working papers in the enterprise's information management system according to a multi-directory structure. According to different walkthrough test scenarios, data can be selected from a specific directory to form a large table.

[0078] (4) Review and verification of cross-references

[0079] Manually compare the cross-references of the draft in the large table. These cross-references include but are not limited to: time comparison, quantity comparison, variety comparison, amount comparison, name comparison, express order number comparison, etc. For example, in the sales walkthrough test scenario, it may be necessary to compare whether the "sales amount" in the sales contract is greater than or equal to the "sales amount" in the sales order, whether the "sales amount" in the sales order is greater than or equal to the "outbound amount" in the outbound order, and whether the "outbound amount" in the accounting voucher and the "receipt amount" in the receipt are equal. For abnormal situations, it is necessary to summarize and generate a comparison result table. The comparison result table should contain all the aforementioned fields, so that all comparison results can be clearly seen, so as to better understand and manage the company's information.

[0080] Manually constructing the relationship between working papers requires business personnel to find and determine the relationship between working papers in the retained enterprise information management system in a large number of working papers according to different walkthrough test scenarios and the actual situation of each enterprise. This not only tests the professional ability of business personnel, but also requires them to have an in-depth understanding of the actual operation of the enterprise. Business personnel need to find the starting point, middle point and end point of the walkthrough test in order to conduct effective testing. However, the relationship between working papers is very complex, which may be one-to-one, one-to-many, many-to-many or many-to-one. Therefore, manually constructing the relationship between working papers is a time-consuming and laborious task. According to experience, an initial public offering (IPO) project requires more than a dozen professionals to spend about a year to sort out the relationship between these working papers. In this way, the transaction working papers with upstream and downstream enterprises can be effectively checked, the authenticity of the transaction can be verified, and regulatory requirements can be met. Manually constructing the relationship between working papers is a time-consuming and laborious task, and it is prone to errors, resulting in high labor costs and difficulty in tracing, which greatly reduces the efficiency of due diligence. Therefore, manually constructing the association relationship between manuscripts leads to low efficiency in constructing the association relationship between manuscripts.

[0081] In order to solve the problem of low efficiency in establishing association relationships between manuscripts, the present application provides a manuscript association method, device, equipment, storage medium and program product.

[0082] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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 are within the scope of protection of this application.

[0083] Exemplary Implementation Environment

[0084] The manuscript association method according to the embodiment of the present disclosure can be executed by an electronic device such as a terminal device or a server. The terminal device can be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The server can be an independent physical server, a server cluster composed of multiple physical servers, or a cloud server capable of cloud computing. The method can be implemented by a processor calling a computer-readable program instruction stored in a memory.

[0085] Exemplary Methods

[0086] See also Figure 2 In an exemplary embodiment, a manuscript association method is provided. Figure 2 As shown in the figure, the process of the manuscript association method mainly includes:

[0087] Step 201: determine at least one second draft similar to the first draft from the drafts to be screened.

[0088] In the exemplary embodiment, the working document refers to the original document that is kept for reference as the original basis. For example, in the procurement walkthrough test scenario, the categories of working documents include purchase requisitions, purchase orders, warehouse receipts, invoices, accounting vouchers, and payment documents, etc.; in the sales walkthrough test scenario, the categories of working documents include sales contracts, sales orders, warehouse delivery orders, logistics orders, revenue confirmation documents, invoices, accounting vouchers, receipt vouchers, and logistics settlement documents, etc.; in the confirmation test scenario, the categories of working documents include scanned copies of outgoing letters, express delivery orders for outgoing letters, scanned copies of reply letters, express delivery orders for reply letters, employee rosters, lists of related parties, and industrial and commercial verification working documents of customers and suppliers, etc.

[0089] In an exemplary embodiment, the manuscripts to be screened refer to a series of manuscripts from which a target manuscript is to be screened, and the target manuscript is related to the first manuscript.

[0090] In some embodiments, step 201 may be to calculate the number of overlapping characters between the draft to be screened and the first draft, and determine at least one second draft similar to the first draft from the draft to be screened based on the number of overlapping characters, wherein the number of overlapping characters between the second draft and the first draft is greater than a threshold value for the number of overlapping characters.

[0091] In some other embodiments, step 201 includes:

[0092] The first draft and the draft to be screened are input into the retrieval model to obtain the similarity between the first draft and the draft to be screened; based on the similarity between the first draft and the draft to be screened, at least one second draft similar to the first draft is determined from the drafts to be screened.

[0093] Among them, the retrieval model is used to predict the similarity between the first manuscript and the manuscript to be screened.

[0094] In an exemplary embodiment, based on the similarity between the first draft and the draft to be screened, at least one second draft similar to the first draft is determined from the draft to be screened, which can be implemented inside the retrieval model or outside the retrieval model.

[0095] In an exemplary embodiment, based on the similarity between the first draft and the draft to be screened, there are multiple implementation methods for determining at least one second draft similar to the first draft from the draft to be screened, including but not limited to the following implementation methods:

[0096] Method 1

[0097] The similarities between the first draft and the draft to be screened are sorted in descending order to obtain a first sorting result; the drafts to be screened corresponding to the first K similarities in the first sorting result are used as at least one second draft similar to the first draft. The value of K can be adjusted.

[0098] Method 2

[0099] Compare the similarity with the similarity threshold; and use the manuscript to be screened corresponding to the similarity greater than the similarity threshold as at least one second manuscript similar to the first manuscript. The similarity threshold can be adjusted.

[0100] The similarity between the first draft and the draft to be screened is automatically acquired through the retrieval model. Further, based on the similarity between the first draft and the draft to be screened, at least one second draft similar to the first draft is automatically determined from the drafts to be screened. The first screening of the drafts is achieved through the retrieval model, and the association relationship between the drafts is automatically constructed. Compared with manually constructing the association relationship between the drafts, the efficiency of constructing the association relationship between the drafts is improved.

[0101] In some embodiments, the process of the retrieval model obtaining the similarity between the first draft and the draft to be screened includes:

[0102] Calculate the Jaccard Index between the first draft and the draft to be screened;

[0103] The Jaccard similarity coefficient is used as the similarity between the first draft and the draft to be screened.

[0104] In an exemplary embodiment, the Jaccard similarity coefficient is used to compare the difference and similarity between two samples.

[0105] In some other embodiments, the process of obtaining the similarity between the first manuscript and the manuscript to be screened by the retrieval model includes:

[0106] Generate a first vector of the first manuscript and a second vector of the manuscript to be screened; and calculate the similarity between the first vector and the second vector.

[0107] In an exemplary embodiment, generating a first vector of the first draft and a second vector of the draft to be screened may be by inputting the first draft and at least one draft to be screened into BERT (Bidirectional Encoder Representation from Transformers) respectively, and obtaining the first vector of the first draft and the second vector of the at least one draft to be screened output by BERT. Inputting the first draft and at least one draft to be screened into BERT respectively may be by inputting the first draft and at least one draft to be screened into the same BERT model in sequence, or by inputting the first draft into a BERT model and inputting at least one draft to be screened into another BERT model in sequence. Inputting the first draft into a BERT model and inputting at least one draft to be screened into another BERT model in sequence, and using a dual-tower structure model to generate the first vector of the first draft and the second vector of the draft to be screened, which can be processed in parallel, thereby improving the efficiency of generating the first vector of the first draft and the second vector of the draft to be screened.

[0108] In an exemplary embodiment, the similarity between the first vector and the second vector is calculated by using ES, Faiss or Milvus. ES (Elasticsearch) is a search server. Faiss is a library for high-dimensional vector indexing and clustering, supports multiple distance metrics, and performs well on large-scale data sets. Milvus is an open source feature vector similarity search engine.

[0109] Step 202, splicing the first draft and at least one second draft to obtain at least one spliced ​​text.

[0110] Among them, a spliced ​​text is obtained by splicing a first draft and a second draft.

[0111] Step 203: input at least one concatenated text into a classification model to obtain a classification result corresponding to each of the at least one concatenated texts.

[0112] Among them, the classification model is used to predict the relevance between the manuscripts contained in the spliced ​​text.

[0113] In an exemplary embodiment, the classification result may include the probability that the first draft and the second draft contained in the spliced ​​text are related; the classification result may also include that the first draft and the second draft contained in the spliced ​​text are related, or that the first draft and the second draft contained in the spliced ​​text are not related. The classification result includes that the first draft and the second draft contained in the spliced ​​text are related, or that the first draft and the second draft contained in the spliced ​​text are not related, and the first label value may be used to indicate that the first draft and the second draft contained in the spliced ​​text are related, and the second label value may be used to indicate that the first draft and the second draft contained in the spliced ​​text are not related. For example, the first label value is 1 and the second label value is 0.

[0114] In some embodiments, the process of obtaining the classification result corresponding to at least one concatenated text by the classification model includes:

[0115] Extracting keywords from at least one of the concatenated texts;

[0116] At least one classification result corresponding to each of the at least one concatenated texts is generated according to the at least one keyword.

[0117] In some other embodiments, the process of obtaining the classification result corresponding to at least one concatenated text by the classification model includes:

[0118] generating at least one third vector of the concatenated text;

[0119] Based on the at least one third vector, a probability that the first draft is associated with the at least one second draft is determined, and a classification result is generated based on the probability.

[0120] In an exemplary embodiment, generating the third vector of at least one concatenated text may be inputting the at least one concatenated text into BERT in sequence to obtain the third vector of at least one concatenated text output by BERT.

[0121] In an exemplary embodiment, there are many ways to generate classification results based on probability, including but not limited to the following ways:

[0122] Method 1

[0123] Directly use the probability as the classification result.

[0124] Method 2

[0125] Compare at least one probability with a probability threshold; determine the classification result corresponding to the spliced ​​text corresponding to a probability greater than the probability threshold as the first draft and the second draft included in the spliced ​​text are related; determine the classification result corresponding to the spliced ​​text corresponding to a probability less than or equal to the probability threshold as the first draft and the second draft included in the spliced ​​text are not related. The probability threshold can be adjusted.

[0126] In some embodiments, the classification model is trained based on sample data, and the sample data includes a first sample spliced ​​text and a second sample spliced ​​text. The first sample spliced ​​text includes the spliced ​​text of the sample manuscript and unrelated manuscripts of the sample manuscript, and the second sample spliced ​​text includes the spliced ​​text of the sample manuscript and related manuscripts of the sample manuscript.

[0127] The classification model is trained by using sample data so that the classification model can accurately predict whether the manuscripts contained in the spliced ​​text are related or unrelated.

[0128] In the exemplary embodiment, the classification model is generally trained, and the built-in pre-trained weight file is not used directly. Because the built-in pre-trained weight file is trained based on ordinary natural language, and the data characteristics of the manuscript are greatly different from ordinary natural language, the classification model is generally trained.

[0129] In some embodiments, the sample data further includes a fourth prompt word, and the fourth prompt word is used to indicate a manuscript determined by the classification model to be related to the sample manuscript.

[0130] In some embodiments, the fourth prompt word includes the sample manuscript itself and the intention of finding manuscripts related to the sample manuscript.

[0131] In some embodiments, the fourth prompt word also includes the upstream and downstream relationship between manuscripts. For example, the fourth prompt word is used to indicate that the classification model determines the downstream manuscript related to the sample manuscript, and the fourth prompt word includes that the manuscript related to the sample manuscript in the second sample concatenated text is the upstream manuscript of the sample manuscript, or the manuscript related to the sample manuscript in the second sample concatenated text is the downstream manuscript of the sample manuscript.

[0132] The upstream and downstream drafts are defined according to the order in which the drafts are generated. The upstream draft is generated first, and the downstream draft is generated later. The upstream and downstream relationship is the relative relationship between the drafts. For example, the purchase requisition is generated first, and then the purchase order is generated. The purchase requisition is the upstream draft of the purchase order, and the purchase order is the downstream draft of the purchase requisition; the purchase order is generated first, and then the payment order is generated. The purchase order is the upstream draft of the payment order, and the payment order is the downstream draft of the purchase order.

[0133] According to the upstream and downstream relationship between the manuscripts in the fourth prompt word, the classification model can record the upstream and downstream relationship between the manuscripts. When the classification model obtains the classification result corresponding to at least one spliced ​​text, the upstream and downstream relationship between the manuscripts contained in the spliced ​​text can also be obtained, and the downstream manuscript related to the first manuscript can be determined from the target manuscript.

[0134] Determining the upstream and downstream relationships between working papers can facilitate the verification of key information in related working papers and can quickly determine the starting point, middle point and end point of the verification.

[0135] In some embodiments, the process of obtaining the first sample concatenated text and the second sample concatenated text includes:

[0136] Calculate the number of overlapping characters between the sample manuscript and the sample manuscript to be screened;

[0137] According to the number of overlapping characters, irrelevant manuscripts of the sample manuscripts are screened out from the sample manuscripts to be screened, and relevant manuscripts of the sample manuscripts are screened out;

[0138] splicing the sample manuscript and the irrelevant manuscripts of the sample manuscript to obtain a first sample spliced ​​text;

[0139] The sample draft and the related drafts of the sample draft are spliced ​​together to obtain a second sample spliced ​​text.

[0140] In some other embodiments, the process of obtaining the first sample concatenated text and the second sample concatenated text includes:

[0141] Determine at least one similar sample manuscript to the sample manuscript from the sample manuscripts to be screened;

[0142] From at least one similar sample draft, screen out irrelevant drafts of the sample draft, and screen out relevant drafts of the sample draft;

[0143] splicing the sample manuscript and the irrelevant manuscripts of the sample manuscript to obtain a first sample spliced ​​text;

[0144] The sample draft and the related drafts of the sample draft are spliced ​​together to obtain a second sample spliced ​​text.

[0145] The sample data used for classification model training is obtained by screening out irrelevant and relevant drafts of a sample draft from at least one similar sample draft of the sample draft, and then concatenating the text with the sample draft, so that the classification model can better distinguish relevant and irrelevant drafts among similar drafts.

[0146] In some embodiments, filtering out irrelevant manuscripts of a sample manuscript from at least one similar sample manuscript includes:

[0147] Calculating the edit distance between the sample manuscript and at least one similar sample manuscript;

[0148] Based on the edit distance, irrelevant manuscripts of the sample manuscript are filtered out from at least one similar sample manuscript.

[0149] In an exemplary embodiment, the edit distance refers to the minimum number of edit operations required to convert one character string into another. The larger the edit distance, the less related the similar sample draft is to the sample draft. Similar sample drafts with an edit distance greater than a preset distance value may be determined as irrelevant drafts of the sample draft.

[0150] In some other embodiments, filtering out irrelevant drafts of a sample draft from at least one similar sample draft includes:

[0151] Calculating the number of overlapping characters between the sample manuscript and at least one similar sample manuscript;

[0152] Based on the number of overlapping characters, irrelevant manuscripts of the sample manuscript are filtered out from at least one similar sample manuscript.

[0153] In the exemplary embodiment, the number of overlapping characters between the irrelevant draft of the sample draft and the sample draft is greater than the lower limit of the number of overlapping characters and less than the upper limit of the number of overlapping characters. The number of overlapping characters is less than the upper limit of the number of overlapping characters, which can avoid the screening out of the relevant draft of the sample draft due to excessive overlap; the number of overlapping characters is greater than the lower limit of the number of overlapping characters, which can make the irrelevant draft of the sample draft and the sample draft that are screened out both similar and different, and the irrelevant draft of the sample draft and the sample draft that are screened out have antagonism.

[0154] The lower limit and upper limit of the number of overlapping characters can be adjusted in different scenarios.

[0155] In some embodiments, the retrieval model is trained based on retrieval sample data, and the retrieval sample data includes retrieval sample manuscripts, related manuscripts of the retrieval sample manuscripts, and irrelevant manuscripts of the retrieval sample manuscripts.

[0156] By training the retrieval model with retrieval sample data, the Euclidean distance between two unrelated manuscripts is enlarged in the high-dimensional vector space, while the Euclidean distance between related manuscripts is reduced, so that the retrieval model can accurately obtain the similarity between the first manuscript and the manuscript to be screened, and determine at least one second manuscript similar to the first manuscript from the manuscripts to be screened, the at least one second manuscript contains more manuscripts related to the first manuscript, and the at least one second manuscript contains fewer manuscripts unrelated to the first manuscript.

[0157] In the exemplary embodiment, the retrieval model is generally trained, and the built-in pre-trained weight file is not used directly. Because the built-in pre-trained weight file is trained based on ordinary natural language, and the data characteristics of the manuscript are greatly different from ordinary natural language, the retrieval model is generally trained.

[0158] In some embodiments, the retrieval sample data includes m pairs consisting of a fifth prompt word and related manuscripts of the retrieval sample manuscript, and n pairs consisting of the fifth prompt word and irrelevant manuscripts of the retrieval sample manuscript; the fifth prompt word is used to instruct the retrieval model to determine a manuscript similar to the sample manuscript, and the fifth prompt word includes the retrieval sample manuscript itself and the intention of finding similar manuscripts to the retrieval sample manuscript.

[0159] Among them, m and n can be equal or unequal. When m and n are unequal, the absolute value of the difference between m and n must be less than the threshold to ensure data balance.

[0160] In some embodiments, the fifth prompt word also includes the upstream and downstream relationship between manuscripts. For example, the fifth prompt word is used to indicate that the retrieval model determines a downstream manuscript similar to the sample manuscript, and the fifth prompt word includes that the related manuscript of the retrieved sample manuscript is the upstream manuscript of the retrieved sample manuscript, or the related manuscript of the retrieved sample manuscript is the downstream manuscript of the retrieved sample manuscript.

[0161] According to the upstream and downstream relationship between the manuscripts in the fifth prompt word, the retrieval model can record the upstream and downstream relationship between the manuscripts, and send the upstream and downstream relationship between the manuscripts to the classification model, so that the classification model can determine the downstream manuscript related to the first manuscript from the target manuscript.

[0162] In some embodiments, only the classification model may be trained without training the retrieval model; or both the classification model and the retrieval model may be trained.

[0163] Step 204, determining a target manuscript related to the first manuscript from at least one second manuscript according to the classification result corresponding to at least one spliced ​​text.

[0164] Through the classification model, it is possible to automatically obtain the classification results corresponding to at least one of the spliced ​​texts, and further, based on the classification results corresponding to at least one of the spliced ​​texts, automatically determine the target manuscript related to the first manuscript from at least one second manuscript, and implement the second screening of the manuscripts through the classification model, and automatically build the association relationship between the manuscripts. Compared with manually building the association relationship between the manuscripts, the efficiency of building the association relationship between the manuscripts is improved.

[0165] At least one second draft similar to the first draft is determined from the drafts to be screened, and a target draft related to the first draft is determined from the at least one second draft. The first draft and the target draft are drafts with an associated relationship. The drafts are screened twice, and a target draft related to the first draft is determined from the drafts to be screened. This can improve the correlation between the target draft and the first draft, and improve the accuracy of constructing the associated relationship between the drafts.

[0166] In an exemplary embodiment, according to the classification results corresponding to at least one concatenated text, a target manuscript related to the first manuscript is determined from at least one second manuscript, which can be implemented inside the classification model or outside the classification model.

[0167] In an exemplary embodiment, there are multiple implementation methods for determining a target draft related to the first draft from at least one second draft according to the classification result corresponding to each of the at least one concatenated texts, including but not limited to the following methods:

[0168] Method 1

[0169] The classification result includes the probability that the first draft and the second draft contained in the concatenated text are related.

[0170] The probabilities are sorted in descending order to obtain a second sorting result; the second drafts corresponding to the first N probabilities in the second sorting result are used as target drafts related to the first draft, wherein the value of N can be adjusted.

[0171] Method 2

[0172] The classification result includes the probability that the first draft and the second draft contained in the concatenated text are related.

[0173] Compare at least one probability with a probability threshold; and use the second draft corresponding to the probability greater than the probability threshold as at least one target draft related to the first draft. The probability threshold can be adjusted.

[0174] Method 3

[0175] The classification result includes whether the first draft and the second draft included in the spliced ​​text are related, or whether the first draft and the second draft included in the spliced ​​text are not related.

[0176] The second draft corresponding to the classification result related to the first draft and the second draft contained in the spliced ​​text is used as the target draft related to the first draft.

[0177] In some embodiments, Figure 3 As shown, the dual retrieval model includes two parts. The first part is the retrieval model, and the second part is the classification model. The retrieval model adopts a double-tower structure model. Figure 3 In the method, the first manuscript Q is input into a BERT model in the retrieval model, and the manuscript to be screened Chunk is input into another BERT model in the retrieval model in turn to obtain a first vector (i.e., vector representation) of the first manuscript and a second vector (i.e., vector representation) of the manuscript to be screened, and the similarity between the first vector and the second vector is calculated by using the ES, Faiss or Milvus tool in the retrieval model, and the similarities between the first manuscript and the manuscript to be screened are sorted in descending order to obtain a first sorting result, and the second manuscripts corresponding to the top K similarities in the first sorting result (Top K) are used as at least one second manuscript similar to the first manuscript, wherein the value of K can be adjusted. The first draft Q and at least one second draft (Top K) are spliced ​​to obtain at least one spliced ​​text, wherein the spliced ​​text is obtained by splicing the first draft and the second draft, the at least one spliced ​​text is input into the BERT in the classification model, and a third vector of the at least one spliced ​​text is obtained, the third vector of the at least one spliced ​​text is input into the softmax layer in the classification model, and the probability of the first draft and the second draft contained in the spliced ​​text being related is obtained, the probabilities are sorted in descending order, and a second sorting result is obtained, and the second drafts corresponding to the top N probabilities in the second sorting result (Top N) are used as target drafts related to the first draft, wherein the value of N can be adjusted.

[0178] In some embodiments, Figure 4 The figure shows a schematic diagram of the relationship between manuscripts. Figure 4 is a directed acyclic graph. Taking the procurement walkthrough test scenario as an example, after confirming the procurement application form, the business personnel can perform semantic retrieval in a large number of drafts, find the most relevant topN drafts, and sort them according to the probability value. Among the N target drafts, the business personnel can accurately select the correct downstream draft and quickly build the association relationship between drafts to form a Figure 4 The knowledge graph shown. Figure 4In the verification, the purchase requisition is the starting point, purchase order 1 and purchase order 2 are downstream drafts related to the purchase requisition, delivery order 1 and delivery order 2 are downstream drafts related to purchase order 1, payment order 1 and payment order 2 are downstream drafts related to purchase order 2, logistics order 1 is the downstream draft related to delivery order 1, logistics order 2 and logistics order 3 are the downstream drafts related to delivery order 2, invoice 1 is the downstream draft related to logistics order 1 and logistics order 2, invoice 2 is the downstream draft related to logistics order 3, Invoice 3 is the downstream draft related to payment order 1, invoice 4 is the downstream draft related to payment order 2, receipt order 1 is the downstream draft related to invoice 1 and invoice 2, receipt order 2 is the downstream draft related to invoice 3 and invoice 4, receipt order 3 is the downstream draft related to invoice 4, receipt order 1 is the downstream draft related to receipt order 1, receipt order 2 is the downstream draft related to receipt order 2, receipt order 3 is the downstream draft related to receipt order 3, and receipt orders 1, 2 and 3 are the verification end points.

[0179] The relationship between manuscripts not only supports the walkthrough testing business, but also supports a wider range of data mining tasks.

[0180] In some embodiments, Figure 5 As shown, the manuscript association method also includes:

[0181] Step 501, determine the category of the first manuscript, and determine the category of the target manuscript.

[0182] In an exemplary embodiment, in a procurement walkthrough test scenario, the categories of working papers include purchase requisitions, purchase orders, warehouse receipts, invoices, accounting vouchers, and payment documents; in a sales walkthrough test scenario, the categories of working papers include sales contracts, sales orders, warehouse delivery notes, logistics notes, revenue confirmation documents, invoices, accounting vouchers, receipt vouchers, and logistics settlement documents; in a confirmation test scenario, the categories of working papers include scanned copies of outgoing letters, express delivery notes for outgoing letters, scanned copies of reply letters, express delivery notes for reply letters, employee rosters, lists of related parties, and customer and supplier industrial and commercial verification working papers.

[0183] In an exemplary embodiment, before step 501, the first draft and the target draft are parsed into a text format. For example, an OCR (Optical Character Recognition) tool is needed to recognize the text and tables in the first draft and the target draft. It should be noted that in order to be more compatible with the input format of the large model, the table recognized by OCR needs to be converted into Markdown (a lightweight markup language) format.

[0184] In some embodiments, determining the category of the first manuscript includes:

[0185] A first prompt word is generated based on the first draft, the first prompt word is input into the large model, and the category of the first draft output by the large model is obtained.

[0186] In other embodiments, Figure 6 As shown, determine the category of the first draft, including:

[0187] Step 601, determine whether the first draft meets the preset rules, if so, execute step 602, otherwise, execute step 603.

[0188] In the exemplary embodiment, according to actual observation, most manuscripts (more than 80%) contain obvious identification information. Based on this identification information, a series of preset rules can be designed for quick judgment. Figure 7 The figure shows a schematic diagram of the preset rules designed for the draft of the category "Supplier List". Figure 7 In the "keyWords_Or" field, it is judged as true as long as any string in the value list appears in the draft text; the "keyWords_And" field indicates that it is judged as true only when all strings in the value list appear in the draft text; the "keyWords_Not" field indicates that it is judged as true only when none of the strings in the value list appears in the draft text. When all three conditions are met, the id of the draft category is returned. Figure 7 In the example, only when any string of {"supplier name", "supplier full name", "company name", "supplier list", "information verification"} appears in the draft text, and when "supplier" appears in the draft text, the id = 32 of the draft category "supplier list" is returned. The draft rule design of other categories is similar to the sample category draft.

[0189] Step 602: Generate a category of the first draft according to a preset rule.

[0190] In an exemplary embodiment, the category of the first draft is generated according to preset rules. The category corresponding to the id returned by the preset rules can be directly used as the category of the first draft; or the major category to which the category corresponding to the id returned by the preset rules belongs can be used as the category of the first draft. For example, the rule matching can identify checks and bills of exchange, and the checks and bills of exchange can be processed as remittance vouchers during post-processing, and the category of the first draft is remittance vouchers; or a custom category mapped to the category corresponding to the id returned by the preset rules can be used as the category of the first draft.

[0191] Step 603: Generate a first prompt word based on the first draft, input the first prompt word into the large model, and obtain the category of the first draft output by the large model.

[0192] The first prompt word is used to indicate the category of the first draft generated by the large model.

[0193] In the exemplary embodiment, a large model (also called a foundation model) refers to a machine learning model with large-scale parameters and complex computing structure, which can process massive data and complete various complex tasks, such as natural language processing, computer vision, speech recognition, etc. Large models usually have billions or even hundreds of billions of parameters.

[0194] In an exemplary embodiment, the large model may specifically be a large language model (Large Language Models, LLMs for short), which refers to a deep learning model trained using a large amount of text data, and is intended to understand and generate natural language text.

[0195] In the exemplary embodiment, the purpose of the first prompt word is to let the big model classify the manuscript text and directly output the result to avoid including other noise information, just like asking the big model to do a judgment question, wrong is wrong, right is right. Figure 8 The figure shows a schematic diagram of a first prompt word. Figure 8 In the example, the "draftStr" variable is the draft text passed into the big model, and "self.llm" represents the big language model. The first prompt word allows the big model to make judgments based on the draft text. Although the big model may have hallucinations and occasionally output noise data, these exceptions can be easily captured and the results can be regenerated through subsequent program exception handling.

[0196] In an exemplary embodiment, after obtaining the output result of the large model, the output result of the large model can be directly used as the category of the first draft, or the output result of the large model can include redundant explanatory information and also include the category of the first draft. A script can be designed to directly extract the category of the first draft, avoiding the need to let the large model perform reasoning again, thereby saving time.

[0197] First, rule matching is used to determine whether the first draft meets the preset rules. Since rule matching is faster for draft classification, and large models are slower for draft classification, if the first draft meets the preset rules, the category of the first draft is generated according to the preset rules, which can improve the speed of draft classification; if the first draft does not meet the preset rules, in the case of rule matching failure, the large model is used for draft classification to ensure that the draft category can be generated. The category of each draft can be obtained quickly and accurately.

[0198] The process of determining the category of the target draft is similar to the process of determining the category of the first draft. For details, please refer to the process of determining the category of the first draft, which will not be explained in detail here.

[0199] Step 502: extract key information from the first draft according to the category of the first draft.

[0200] In some embodiments, step 502 includes:

[0201] According to the category of the first draft, key parameters to be extracted are obtained;

[0202] According to the key parameter, find the value of the key parameter from the first draft;

[0203] The key parameter and the value of the key parameter are determined as key information in the first draft.

[0204] In other embodiments, Fig. 9 As shown, step 502 includes:

[0205] Step 901, according to the category of the first draft, obtain the key parameters to be extracted and obtain the definition of the key parameters.

[0206] In the exemplary embodiment, the mapping relationship between the draft category and the key parameters, as well as the definition of the key parameters, are pre-stored. For example, the key parameter "warehouse entry date" is defined as the time point when the goods arrive at the predetermined location. The marking method of "warehouse entry date" may be different in each enterprise. Some enterprises mark it as "arrival date" and some enterprises mark it as "arrival information". Although the expression forms are different, the core meaning is the same, all referring to the time point when the goods arrive at the predetermined location. Therefore, according to the definition of the key parameters, the parameters with the same meaning as the key parameters but different expressions can also be identified to avoid missing key information.

[0207] Step 902: Generate a second prompt word based on the first draft, the key parameters and the definition of the key parameters.

[0208] In some embodiments, step 902 includes:

[0209] Based on the first draft, key parameters, definitions of key parameters and preset formats, a second prompt word is generated, wherein the second prompt word is also used to instruct the large model to output the key information in the first draft according to the preset format.

[0210] In the exemplary embodiment, the preset format refers to a unified interactive format for data interaction with back-end development programmers. For example, the preset format can be a json format or an XML format, etc., preferably a json format. The following explanation is given using the json format as an example, but it is not limited to this.

[0211] JSON stands for JavaScript Object Notation, which is a lightweight data exchange format. XML is the abbreviation of eXtensible Markup Language, which is an extensible markup language and a simple data storage language.

[0212] In an exemplary embodiment, Fig.10 Shown is a schematic diagram of a second prompt word. Fig.10 In the example, the variable "self.jsonExample" represents the json format template of the draft output of this category, and the variable "draftTxt" is the plain text representation of the draft after OCR processing. The advantage of this design is that the large model is free of noise and redundant information as much as possible.

[0213] Step 903, input the second prompt word into the large model to obtain key information in the first draft output by the large model.

[0214] The second prompt word is used to instruct the large model to extract key information in the first draft, and the key information includes key parameters and the values ​​of the key parameters.

[0215] In some embodiments, after step 903, the manuscript association method further includes:

[0216] If the format of the key information in the first draft is not a preset format, generating a third prompt word based on the key information in the first draft;

[0217] The third prompt word is input into the large model to obtain key information in the first draft output by the large model in a preset format, wherein the third prompt word is used to instruct the large model to correct the format of the key information in the first draft and output the key information in the first draft according to the preset format.

[0218] Since the large model may have hallucinations, it is necessary to judge and process the output content of the large model. Using the open source python (a programming language) tool to judge whether the format of the large model output is correct, correcting the json output format is a complex task. According to the third prompt word, let the large model correct the format of the key information in the first draft, output the key information in the first draft according to the preset format, and let the large model correct the output format by itself, which can reduce the complexity of correcting the format of the key information in the first draft.

[0219] In an exemplary embodiment, Fig.11 Shown is a schematic diagram of a third prompt word. Fig.11In the example, the "errorStr" variable represents the json format output by the large model that needs to be corrected. This process will be repeated until the large model can output the correct json format.

[0220] In an exemplary embodiment, after step 903, a post-processing module may be set to process key information in the first draft output by the large model. For example, the post-processing module may perform standardized output of certain fields, such as time representation in different formats, which the post-processing module may unify into the same format.

[0221] At present, when dealing with the working papers of companies to be listed, the working papers are not only huge in number, but also varied in types. It is necessary to conduct detailed classification and accurate matching of walkthrough test scenarios in order to effectively extract key information. First, due to the large number of working papers involved, it is difficult for business personnel to avoid negligence and errors when sorting out the working papers, even if they are not lacking in professional ability. These errors may be unintentional, but their existence brings hidden dangers to the accuracy of the key information in the working papers, and once they occur, it is often difficult to trace the specific source, thus affecting the quality of the whole sorting of the key information in the working papers. Secondly, when the key information is entered into the enterprise management system, due to the cumbersome operation and the large number, the occurrence of errors is almost inevitable. These errors in key information, whether due to negligence or other reasons, may cause abnormalities in the subsequent review and verification work, and thus affect the company's initial public offering process. Therefore, manual identification of key information in the working papers leads to a high error rate in the identification of key information in the working papers.

[0222] In this application, the key information in the manuscript is extracted by a large model, which not only improves the efficiency of identifying the key information in the manuscript, but also improves the accuracy of identifying the key information in the manuscript compared to manual identification of the key information in the manuscript.

[0223] Step 503, extracting key information from the target draft according to the category of the target draft.

[0224] Step 504, compare the key information in the first draft with the key information in the target draft to obtain a comparison result.

[0225] The comparison result includes the relationship between the key information in the first draft and the key information in the target draft.

[0226] In the exemplary embodiment, the relationship includes a size relationship, an equal relationship, a containment relationship, an equivalent relationship, etc. For example, when the key information is an amount, quantity, etc., the comparison result includes a size relationship or an equal relationship between the key information in the first draft and the key information in the target draft; when the key information is a name, variety, etc., the comparison result includes a containment relationship or an equivalent relationship between the key information in the first draft and the key information in the target draft.

[0227] The classification model is used to implement a second screening of the drafts, and the relationship between the drafts is automatically constructed. Compared with manually constructing the relationship between the drafts, the efficiency of constructing the relationship between the drafts is improved. Since the efficiency of constructing the relationship between the drafts is improved, the efficiency of comparing the key information in the drafts with related relationships is further improved.

[0228] In summary, in the present application, at least one second draft similar to the first draft is determined from the drafts to be screened, and a target draft related to the first draft is determined from the at least one second draft. The first draft and the target draft are drafts with an associated relationship, and the drafts are screened twice. In the second screening process, the first draft and the at least one second draft are spliced ​​to obtain at least one spliced ​​text, wherein a spliced ​​text is obtained by splicing a first draft and a second draft. The at least one spliced ​​text is input into the classification model to obtain the classification results corresponding to the at least one spliced ​​text. The classification model is used to predict the correlation between the drafts contained in the spliced ​​text, and then according to the classification results corresponding to the at least one spliced ​​text, the target draft related to the first draft is determined from the at least one second draft. The second screening of the draft is achieved through the classification model, and the association relationship between the drafts is automatically constructed. Compared with the manual construction of the association relationship between the drafts, the construction efficiency of the association relationship between the drafts is improved.

[0229] Exemplary Devices

[0230] Correspondingly, the embodiment of the present application also provides a manuscript association device, such as Fig.12 As shown, the manuscript association device includes:

[0231] A screening unit 1201 is used to determine at least one second draft similar to the first draft from the drafts to be screened;

[0232] The splicing unit 1202 is used to splice the first draft and at least one of the second drafts to obtain at least one spliced ​​text; wherein the spliced ​​text is obtained by splicing one of the first draft and one of the second draft;

[0233] The classification unit 1203 is used to input at least one of the spliced ​​texts into a classification model to obtain a classification result corresponding to at least one of the spliced ​​texts; wherein the classification model is used to predict the relevance between the manuscripts contained in the spliced ​​texts;

[0234] The processing unit 1204 is used to determine a target manuscript related to the first manuscript from at least one of the second manuscripts according to the classification results corresponding to at least one of the spliced ​​texts.

[0235] Optionally, the classification unit 1203 is specifically used to generate at least one third vector of the spliced ​​text; determine the probability that the first manuscript and at least one second manuscript are related according to at least one third vector, and generate a classification result based on the probability.

[0236] Optionally, the screening unit 1201 includes:

[0237] A retrieval subunit, used for inputting the first draft and the draft to be screened into the retrieval model to obtain the similarity between the first draft and the draft to be screened;

[0238] The first processing subunit is used to determine at least one second draft similar to the first draft from the drafts to be screened based on the similarity between the first draft and the draft to be screened.

[0239] Optionally, the retrieval subunit is specifically used to generate a first vector of the first manuscript and a second vector of the manuscript to be screened; and calculate the similarity between the first vector and the second vector.

[0240] Optionally, the classification model is trained based on sample data, and the sample data includes a first sample spliced ​​text and a second sample spliced ​​text, the first sample spliced ​​text includes the spliced ​​text of a sample manuscript and manuscripts unrelated to the sample manuscript, and the second sample spliced ​​text includes the spliced ​​text of the sample manuscript and manuscripts related to the sample manuscript.

[0241] Optionally, the manuscript associating device further includes:

[0242] A first determining unit, used to determine at least one similar sample manuscript to the sample manuscript from the sample manuscripts to be screened;

[0243] A sample screening unit, used to screen out irrelevant drafts of the sample draft from at least one of the similar sample drafts, and screen out relevant drafts of the sample draft;

[0244] A first sample splicing unit, used for splicing the sample manuscript and irrelevant manuscripts of the sample manuscript to obtain the first sample splicing text;

[0245] The second sample splicing unit is used to splice the sample manuscript and the related manuscripts of the sample manuscript to obtain the second sample splicing text.

[0246] Optionally, the sample screening unit is specifically configured to calculate the number of overlapping characters between the sample draft and at least one similar sample draft; and screen out irrelevant drafts of the sample draft from at least one similar sample draft based on the number of overlapping characters.

[0247] Optionally, the manuscript associating device further includes:

[0248] A second determining unit, used to determine the category of the first draft and the category of the target draft;

[0249] A first extraction unit, configured to extract key information from the first draft according to the category of the first draft;

[0250] A second extraction unit, used for extracting key information from the target manuscript according to the category of the target manuscript;

[0251] The comparison unit is used to compare the key information in the first draft with the key information in the target draft to obtain a comparison result, wherein the comparison result includes the relationship between the key information in the first draft and the key information in the target draft.

[0252] Optionally, the second determination unit is specifically used to determine whether the first draft satisfies preset rules; if the first draft satisfies the preset rules, a category of the first draft is generated according to the preset rules; if the first draft does not satisfy the preset rules, a first prompt word is generated based on the first draft, the first prompt word is input into the large model, and the category of the first draft output by the large model is obtained, wherein the first prompt word is used to indicate the category of the first draft generated by the large model.

[0253] Optionally, the first extraction unit is specifically used to obtain key parameters to be extracted according to the category of the first draft, and obtain the definition of the key parameters; generate a second prompt word based on the first draft, the key parameters and the definition of the key parameters; input the second prompt word into the large model to obtain key information in the first draft output by the large model, wherein the second prompt word is used to instruct the large model to extract the key information in the first draft, and the key information includes the key parameters and the values ​​of the key parameters.

[0254] The draft association device provided in this embodiment belongs to the same application concept as the draft association method provided in the above embodiments of this application, can execute the draft association method provided in any of the above embodiments of this application, and has the corresponding functional modules and beneficial effects of executing the draft association method. For technical details not fully described in this embodiment, please refer to the specific processing content of the draft association method provided in the above embodiments of this application, which will not be repeated here.

[0255] The functions implemented by the above screening unit 1201, splicing unit 1202, classification unit 1203 and processing unit 1204 can be implemented by the same or different processors respectively, and the embodiments of the present application are not limited thereto.

[0256] It should be understood that the units in the above devices can be implemented in the form of a processor calling software. For example, the device includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit of the device, wherein the processor can be a general-purpose processor, such as a CPU or a microprocessor, etc., and the memory can be a memory in the device or a memory outside the device. Alternatively, the units in the device can be implemented in the form of hardware circuits, and the functions of some or all units can be realized by designing the hardware circuits. The hardware circuit can be understood as one or more processors; for example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all of the above units are realized by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit can be implemented by PLD, taking FPGA as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by the configuration file, so as to realize the functions of some or all of the above units. All units of the above devices can be implemented in the form of a processor calling software, or in the form of hardware circuits, or in part by a processor calling software, and the remaining part is implemented in the form of hardware circuits.

[0257] In an embodiment of the present application, a processor is a circuit with the ability to process signals. In one implementation, the processor may be a circuit with the ability to read and run instructions, such as a CPU, a microprocessor, a GPU, or a DSP; in another implementation, the processor may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as a hardware circuit implemented by an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document to implement the hardware circuit configuration can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, DPU, etc.

[0258] It can be seen that each unit in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0259] In addition, all or part of the units in the above device can be integrated together, or can be implemented independently. In one implementation, these units are integrated together and implemented in the form of a SOC. The SOC may include at least one processor for implementing any of the above methods or implementing the functions of each unit of the device. The type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.

[0260] Exemplary Electronic Devices

[0261] Another embodiment of the present application also provides an electronic device, see Fig.13 As shown, the device includes:

[0262] Memory 200 and processor 210;

[0263] The memory 200 is connected to the processor 210 and is used to store programs;

[0264] The processor 210 is used to implement the manuscript association method disclosed in any of the above embodiments by running the program stored in the memory 200.

[0265] Specifically, the electronic device may further include: a bus, a communication interface 220 , an input device 230 and an output device 240 .

[0266] The processor 210, the memory 200, the communication interface 220, the input device 230 and the output device 240 are connected to each other via a bus.

[0267] A bus may include a pathway that transfers information between components of a computer system.

[0268] The processor 210 may be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the scheme of the present invention. It may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0269] The processor 210 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0270] The memory 200 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include a program code, and the program code includes a computer operation instruction. More specifically, the memory 200 may include a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM), other types of dynamic storage devices that can store information and instructions, a disk storage, a flash, and the like.

[0271] The input device 230 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0272] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0273] The communication interface 220 may include any transceiver or the like to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0274] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement each step of any manuscript association method provided in the above embodiments of the present application.

[0275] Exemplary computer program products and storage media

[0276] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the manuscript association method according to various embodiments of the present application described in any of the above embodiments of this specification.

[0277] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0278] In addition, the embodiment of the present application may also be a storage medium on which a computer program is stored. The computer program is executed by a processor to execute the steps of the manuscript association method according to various embodiments of the present application described in any of the above embodiments of this specification, and specifically the following steps may be implemented:

[0279] Step 201: determine at least one second draft similar to the first draft from the drafts to be screened.

[0280] Step 202, splicing the first draft and at least one second draft to obtain at least one spliced ​​text.

[0281] Among them, a spliced ​​text is obtained by splicing a first draft and a second draft.

[0282] Step 203: input at least one concatenated text into a classification model to obtain a classification result corresponding to each of the at least one concatenated texts.

[0283] Among them, the classification model is used to predict the relevance between the manuscripts contained in the spliced ​​text.

[0284] Step 204, determining a target manuscript related to the first manuscript from at least one second manuscript according to the classification result corresponding to at least one spliced ​​text.

[0285] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0286] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0287] The steps in the methods of each embodiment of the present application can be adjusted in order, combined and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0288] The modules and sub-modules in the devices and terminals in the various embodiments of the present application can be combined, divided and deleted according to actual needs.

[0289] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, 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 an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0290] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0291] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0292] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0293] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly by hardware, software units executed by a processor, or a combination of the two. The software units may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0294] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0295] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A manuscript association method, characterized in that: include: Determine at least one second manuscript similar to the first manuscript from the manuscripts to be screened; Splicing the first draft and at least one of the second drafts to obtain at least one spliced ​​text; wherein one of the spliced ​​texts is obtained by splicing one of the first draft and one of the second draft; Inputting at least one of the spliced ​​texts into a classification model to obtain a classification result corresponding to at least one of the spliced ​​texts; wherein the classification model is used to predict the relevance between the manuscripts contained in the spliced ​​text; According to the classification result corresponding to at least one of the spliced ​​texts, a target manuscript related to the first manuscript is determined from at least one of the second manuscripts.

2. The manuscript association method according to claim 1, characterized in that: The process of the classification model obtaining the classification result corresponding to at least one of the spliced ​​texts comprises: Generate at least one third vector of the concatenated text; According to at least one of the third vectors, a probability that the first draft is associated with at least one of the second drafts is determined, and a classification result is generated based on the probability.

3. The manuscript association method according to claim 1, characterized in that: The step of determining at least one second draft similar to the first draft from the drafts to be screened comprises: Inputting the first draft and the draft to be screened into the retrieval model to obtain the similarity between the first draft and the draft to be screened; Based on the similarity between the first draft and the draft to be screened, at least one second draft similar to the first draft is determined from the drafts to be screened.

4. The manuscript association method according to claim 3, characterized in that: The process of obtaining the similarity between the first manuscript and the manuscript to be screened by the retrieval model includes: Generate a first vector of the first manuscript and a second vector of the manuscript to be screened; A similarity between the first vector and the second vector is calculated.

5. The manuscript association method according to claim 1, characterized in that: The classification model is trained based on sample data, and the sample data includes a first sample spliced ​​text and a second sample spliced ​​text. The first sample spliced ​​text includes the spliced ​​text of a sample manuscript and manuscripts unrelated to the sample manuscript, and the second sample spliced ​​text includes the spliced ​​text of the sample manuscript and manuscripts related to the sample manuscript.

6. The manuscript association method according to claim 5, characterized in that: The method further comprises: Determine at least one similar sample manuscript to the sample manuscript from the sample manuscripts to be screened; From at least one of the similar sample manuscripts, filter out irrelevant manuscripts of the sample manuscript, and filter out relevant manuscripts of the sample manuscript; splicing the sample manuscript and irrelevant manuscripts of the sample manuscript to obtain the first sample spliced ​​text; The sample manuscript and the related manuscripts of the sample manuscript are spliced ​​together to obtain the second sample spliced ​​text.

7. The manuscript association method according to claim 6, characterized in that: The step of screening out irrelevant drafts of the sample draft from at least one of the similar sample drafts includes: Calculating the number of overlapping characters between the sample manuscript and at least one similar sample manuscript; According to the number of overlapping characters, irrelevant drafts of the sample draft are screened out from at least one similar sample draft.

8. The manuscript association method according to any one of claims 1 to 7, characterized in that: The method further comprises: Determining the category of the first manuscript and determining the category of the target manuscript; Extracting key information from the first draft according to the category of the first draft; Extracting key information from the target manuscript according to the category of the target manuscript; The key information in the first draft is compared with the key information in the target draft to obtain a comparison result, wherein the comparison result includes a relationship between the key information in the first draft and the key information in the target draft.

9. The manuscript association method according to claim 8, characterized in that: The determining the category of the first draft includes: Determining whether the first draft satisfies a preset rule; If the first draft satisfies the preset rule, generating a category of the first draft according to the preset rule; If the first draft does not satisfy the preset rules, a first prompt word is generated based on the first draft, and the first prompt word is input into the large model to obtain the category of the first draft output by the large model, wherein the first prompt word is used to indicate the category of the first draft generated by the large model.

10. The manuscript association method according to claim 8, characterized in that: The step of extracting key information from the first draft according to the category of the first draft includes: According to the category of the first draft, key parameters to be extracted are obtained, and definitions of the key parameters are acquired; Generate a second prompt word based on the first draft, the key parameter and the definition of the key parameter; The second prompt word is input into the large model to obtain key information in the first draft output by the large model, wherein the second prompt word is used to instruct the large model to extract key information in the first draft, and the key information includes the key parameters and the values ​​of the key parameters.

11. A manuscript association device, characterized in that: include: A screening unit, used to determine at least one second draft similar to the first draft from the drafts to be screened; A splicing unit, used for splicing the first draft and at least one of the second drafts to obtain at least one spliced ​​text; wherein one of the spliced ​​texts is obtained by splicing one of the first draft and one of the second draft; A classification unit, used for inputting at least one of the spliced ​​texts into a classification model to obtain a classification result corresponding to at least one of the spliced ​​texts; wherein the classification model is used for predicting the relevance between the manuscripts contained in the spliced ​​texts; A processing unit is used to determine a target manuscript related to the first manuscript from at least one of the second manuscripts according to the classification results corresponding to at least one of the spliced ​​texts.

12. An electronic device, characterized in that: including memory and processor; The memory is connected to the processor and is used to store programs; The processor is used to implement the manuscript association method as described in any one of claims 1 to 10 by running the program in the memory.

13. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the manuscript association method according to any one of claims 1 to 10 is implemented.

14. A computer program product, characterized in that The method comprises computer program instructions, which, when executed by a processor, enable the processor to execute the manuscript association method as described in any one of claims 1 to 10.

Images