Data processing method and device, electronic device, storage medium and program product

Abstract

The present disclosure provides a data processing method and device, an electronic device, a storage medium and a program product, which relate to the field of artificial intelligence technology. The method comprises: based on the first prompt information, determining the first context information associated with the element to be extracted from the audit text through the information extraction model; based on the second prompt information, determining the extraction result corresponding to the element to be extracted from the first context information through the information extraction model; and determining the position information of the extraction result in the audit text according to the first context information. The embodiment of the present disclosure realizes the precise positioning of the original position of the element to be extracted in the audit text.

Description

Background Art

[0002] When reviewing certain specific texts such as business contracts, it is necessary to extract and locate the elements that require attention. Since similar elements may appear multiple times in different positions in the text, element positioning is prone to confusion and inaccurate positioning.

[0003] Therefore, how to accurately extract elements and determine their locations has become one of the hot issues in the field of data processing.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute relevant technology known to ordinary technicians in the field. Summary of the invention

[0005] The present disclosure provides a data processing method and device, electronic equipment, storage medium and program product, which can realize the precise positioning of elements in text.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, there is provided a data processing method, comprising:

[0008] Based on the first prompt information, determining first context information associated with the element to be extracted from the review text through an information extraction model;

[0009] Based on the second prompt information, determining an extraction result corresponding to the element to be extracted from the first context information through the information extraction model;

[0010] The location information of the extraction result in the review text is determined according to the first context information.

[0011] According to another aspect of the present disclosure, there is also provided a data processing device, including:

[0012] A context information determination module, used to determine first context information associated with the element to be extracted from the audit text through an information extraction model based on the first prompt information;

[0013] An extraction result determination module, used to determine an extraction result corresponding to the element to be extracted from the first context information through the information extraction model based on the second prompt information;

[0014] A position information determination module is used to determine the position information of the extraction result in the review text according to the first context information.

[0015] According to another aspect of the present disclosure, an electronic device is also provided, which includes: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the above-mentioned data processing method by executing the executable instructions.

[0016] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned data processing method is implemented.

[0017] According to another aspect of the present disclosure, a computer program product is also provided, including a computer program, and when the computer program is executed by a processor, the above-mentioned data processing method is implemented.

[0018] The data processing method and device, electronic device, storage medium and program product provided in the embodiments of the present disclosure. The scheme provided in the embodiments of the present disclosure firstly extracts the first context information associated with the element to be extracted from the audit text based on the first prompt information and using the information extraction model. The first context information is the original text related to the element to be extracted extracted from the audit text, which can avoid the problem that the extraction result is easily confused when there are multiple similar elements to be extracted in the audit text at the same time; and then based on the second prompt information, the information extraction model is used to determine the extraction result corresponding to the element to be extracted in the audit text from the first context information. The first context information is used as the basis for the element to be extracted, so as to achieve the accuracy of the extraction of the element to be extracted. Compared with extracting elements from the full text of the audit text, extracting the corresponding extraction result from the first context information narrows the extraction range, so as to improve the extraction efficiency on the one hand and the accuracy of the extraction on the other hand. The first context information is further used to determine the position information of the extraction result in the audit text, so as to achieve the rapid and accurate positioning of the original text position of the extraction result in the audit text.

[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0021] Figure 1 A schematic diagram of the structure of a data processing system in an embodiment of the present disclosure is shown.

[0022] Figure 2 A flow chart of a data processing method in an embodiment of the present disclosure is shown.

[0023] Figure 3 A schematic diagram showing a question-and-answer example of a data processing method in an embodiment of the present disclosure.

[0024] Figure 4 A question-and-answer example framework diagram of a data processing method in an embodiment of the present disclosure is shown.

[0025] Figure 5 An architectural diagram showing a data processing method in an embodiment of the present disclosure.

[0026] Figure 6 A schematic structural diagram of a data processing device in an embodiment of the present disclosure is shown.

[0027] Figure 7 A structural block diagram of an electronic device showing a data processing method in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0029] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0030] The specific implementation of the embodiment of the present disclosure is described in detail below with reference to the accompanying drawings.

[0031] Figure 1 The schematic diagram of the structure of an exemplary data processing system to which the data processing method in the embodiment of the present disclosure can be applied is shown. The data processing method is used in the data processing system to extract elements, so as to achieve accurate positioning of the original text in the audit text by the extraction result. Figure 1 As shown, the system structure may include a terminal device 101 , a network 102 and a server 103 .

[0032] The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103, and can be a wired network or a wireless network.

[0033] Optionally, the wireless network or wired network described above uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. are used to represent data exchanged through the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above data communication technologies.

[0034] The terminal device 101 can be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, wearable devices, augmented reality devices, virtual reality devices, etc.

[0035] Optionally, the client of the application installed in different terminal devices 101 is the same, or the client of the same type of application based on different operating systems. Based on the different terminal platforms, the specific form of the client of the application can also be different, for example, the application client can be a mobile client, a PC client, etc.

[0036] The server 103 may be a server that provides various services, such as a background management server that provides support for the device operated by the user using the terminal device 101. The background management server may analyze and process the received request and other data, and feed back the processing results to the terminal device.

[0037] Optionally, server 103 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application is not limited here.

[0038] Those skilled in the art will know that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration, and any number of terminal devices, networks and servers may be provided according to actual needs, and the embodiments of the present disclosure do not limit this.

[0039] Under the above system architecture, a data processing method is provided in an embodiment of the present disclosure, and the method can be executed by any electronic device with computing and processing capabilities.

[0040] In some embodiments, the data processing method provided in the embodiments of the present disclosure may be executed by a terminal device of the above-mentioned system architecture; in other embodiments, the data processing method provided in the embodiments of the present disclosure may be executed by a server in the above-mentioned system architecture; in other embodiments, the data processing method provided in the embodiments of the present disclosure may be implemented by a terminal device and a server in the above-mentioned system architecture through interaction.

[0041] In business exchanges and activities, in order to protect the interests of the parties involved and provide legal protection, the parties involved need to sign a large number of contracts. In order to ensure the legality and compliance of the contract and avoid legal and financial risks, the contract needs to be reviewed before it is finally signed. Contract review requires professional legal personnel to conduct a comprehensive, clause-by-clause, and multiple rounds of review of the contract text, which takes a lot of time for legal personnel and has a very low review efficiency. At the same time, due to negligence of personnel or different understandings of the terms, review errors and omissions may occur.

[0042] Smart contract review based on artificial intelligence technology can overcome some problems caused by manual review and has the following advantages: It can significantly improve the efficiency of review and reduce the time and cost of manual review. It can timely identify risks and problems in the contract and provide suggestions for manual review. It can avoid errors and omissions caused by human negligence or lack of professional knowledge.

[0043] Smart contract review needs to extract the elements in the contract to be reviewed first. The extracted elements can be, for example, shipping time, delivery time, payment amount, payment method, delivery address, means of transport, and mode of transport, etc. Then, the extracted elements are compared according to the defined expert rules to find out the problems in the contract to be reviewed. Legal personnel are required to finally confirm whether the review results are correct. If there is no function to locate the original text segment where the elements are located, legal personnel need to find the keywords corresponding to the element extraction results from a dozen or even hundreds of pages of contracts before they can find the original text segment where the elements are located. When the element extraction results appear multiple times in the contract, it takes more time to locate the correct original text segment. Generally, smart contract review will contain more than 50 elements, which will cause legal personnel to spend a lot of time reviewing the smart review results. The long legal review time is also one of the reasons why smart contract review has not been widely popularized.

[0044] In response to the above situation, some commonly used solutions for locating original text fragments in contract review have also appeared in related technologies. One way is to use an artificial intelligence model to extract elements from the contract to obtain element extraction results, and match the element extraction results with the contract text line by line; if the match is successful, the line number corresponding to the contract text is returned. However, this method has the following problems: when the element extraction result appears multiple times in the contract text, multiple line numbers will be returned, resulting in inaccurate element positioning of the original text fragment. In addition, the element extraction value is generated by an artificial intelligence model, rather than directly derived from the original text, which will cause the element extraction result to be rewritten in terms of expression, making it impossible to locate the original text. For example, in the related art, there is a solution for extracting elements using an artificial intelligence model. The text input to the artificial intelligence model is "Based on the public consultation between Party A and Party B, the total contract price for 100 pieces of commodity A is [13,000] yuan", and the element to be extracted is "the total contract price". The output of the artificial intelligence model is output word by word. One output result may be "the total contract price is 13,000 yuan", and the other output result may be "the total contract price is 13,000 yuan". The above two output results are the output of the artificial intelligence model after understanding the text, filtering out the special symbols "[", "]" and "," in the text, and it is possible to rewrite Arabic numerals as text characters, so that the extraction result is inconsistent with the original content in description, resulting in the inability to locate the original text.

[0045] In order to solve the above technical problems, the present application proposes a data processing method that can achieve accurate feature extraction and positioning. Figure 2 A schematic diagram of a data processing method in an embodiment of the present disclosure is shown. Figure 2 As shown, the data processing method provided in the embodiment of the present disclosure includes the following steps:

[0046] Step S202: Based on the first prompt information, determine the first context information associated with the element to be extracted from the review text through the information extraction model.

[0047] Step S204: Based on the second prompt information, an extraction result corresponding to the element to be extracted is determined from the first context information through an information extraction model.

[0048] Step S206: Determine the position information of the extraction result in the review text according to the first context information.

[0049] In the embodiment, firstly, based on the first prompt information, the first context information associated with the element to be extracted is extracted from the audit text using the information extraction model. The first context information is the original text related to the element to be extracted extracted from the audit text, which ensures that it will not be modified in description by the information extraction model, and can avoid the problem that the extraction result is easily confused when there are multiple similar elements to be extracted in the audit text at the same time; and then based on the second prompt information, the information extraction model is used to determine the extraction result corresponding to the element to be extracted in the audit text from the first context information. The first context information is used as the basis for the element to be extracted, and the accuracy of the extraction of the element to be extracted is achieved. Compared with extracting elements from the full text of the audit text, extracting the corresponding extraction result from the first context information narrows the extraction range, so on the one hand, the efficiency of extraction is improved, and on the other hand, the accuracy of extraction is also improved. The first context information is further used to determine the position information of the extraction result in the audit text, so that the extraction result can be quickly and accurately located in the original text position in the audit text.

[0050] In an embodiment, the above-mentioned text to be reviewed can be a text containing multiple contents such as a contract to be reviewed, an assignment to be reviewed, a project proposal to be reviewed, etc. The above-mentioned information extraction model is a model that performs the task of extracting the original text content from the review text according to the prompt information. The prompt information and the review text are input into the information extraction model, the information extraction model performs semantic understanding on the review text, establishes and executes the corresponding task based on the prompt information, and the task is used to extract the original content of the target text from the review text according to the prompt information. In one embodiment, a large language model is used as the information extraction model. Large Language Model (LLM) is a natural language processing model with a large number of parameters. This model usually requires a large amount of computing resources and training data to train, so it can handle various complex natural language tasks, including language understanding, generation, translation, etc. The large language model is usually used to input questions and answer the input questions. The questions input to the large language model are generally described in natural language. For example, the question input to the large language model is "What is the boiling point of water?", then the answer output by the large language model is "Under standard atmospheric pressure (101.325kPa, also known as one atmosphere) , The boiling point of water is 100°C. The large language model is able to handle a variety of complex natural language tasks, so it can also be applied to text review to achieve rapid positioning of elements to be extracted. The answer output by the large language model is usually a single response to the question, and this answer will output all the content according to the set output logic. In addition to the large language model, the information extraction model in this embodiment can also be other types of generative language dialogue models, such as a generative artificial intelligence model.

[0051] The first prompt information and the second prompt information are used to guide the information extraction model to analyze the review text, thereby outputting the first context information and the extraction result.

[0052] The first prompt information in the disclosed embodiment refers to relevant prompt information for instructing the information extraction model to determine the content associated with the element to be extracted from the audit text. The content associated with the element to be extracted includes the text description located before the element to be extracted and / or the text description located after the element to be extracted in the audit text. In the disclosed embodiment, the content associated with the element to be extracted determined from the audit text can be used as the first context information. The first prompt information and the audit text are input into the information extraction model together, the information extraction model parses and understands the audit text, and the first prompt information instructs the information extraction model to extract the content associated with the element to be extracted from the understood audit text.

[0053] In some embodiments, the first prompt information refers to the relevant prompt information for instructing the information extraction model to extract the original content associated with the element to be extracted from the audit text, and directly use the original content associated with the element to be extracted in the audit text as the first context information without rewriting or deleting it. Exemplarily, the first context information refers to the original text associated with the element to be extracted extracted from the audit text according to the first prompt information.

[0054] In an embodiment, a first prompt message is "Please quote the original text to answer the question "[Elements to be extracted]", please make sure your answer directly quotes the original text without any modification or explanation". The above first prompt message is input into the information extraction model, and the information extraction model can extract the content associated with the "elements to be extracted" as the first context information based on the original text (i.e., the review text) understood by the input. For example, the element to be extracted is "payment completion time", and the original text of the review text is "Delivery time: Party A must complete the payment within [7] working days after Party B delivers the product". The first prompt message is "Please quote the original text of the review text to answer the question "payment completion time", please make sure your answer directly quotes the original text without any modification or explanation". The original text of the above review text and the first prompt message are input into the information extraction model, and the information extraction model outputs the original content associated with the "payment completion time" "Party A must complete the payment within [7] working days after Party B delivers the product" as the first context information.

[0055] In the disclosed embodiment, the second prompt information refers to the relevant prompt information used to instruct the information extraction model to determine the content associated with the element to be extracted from the first context information. The content associated with the element to be extracted determined from the first context information is used as the extraction result. Based on the above basis of using the information extraction model to determine the first context information, the information extraction model already has the review text, the first prompt information and the first context information. At this time, the second prompt information is input into the information extraction model to instruct the information extraction model to further understand the first context information and extract the content corresponding to the second prompt information from the understood context information as the extraction result. Based on the above-obtained "Party A must complete the payment within [7] working days after Party B delivers the product" as the first context information, the second prompt information is "According to the first context information, answer the following question: payment time", and the second prompt information instructs the information extraction model to extract more specific related content "within 7 working days after the delivery of the product" from the first context information, and takes "within 7 working days after the delivery of the product" as the extraction result of the element to be extracted "payment completion time".

[0056] In an exemplary embodiment, the process of determining the first prompt information and the second prompt information includes:

[0057] Construct questions based on the elements to be extracted;

[0058] Constructing a question-and-answer example; wherein the question-and-answer example is used to guide the information extraction model to decompose the question into multiple tasks, so as to guide the information extraction model to answer the multiple tasks respectively; the information extraction model answers the question described in natural language;

[0059] According to the question-and-answer example and the question asked, the first prompt information and the second prompt information are determined.

[0060] In the embodiment, questions are generally constructed in a template manner. For example, one template is "Please quote the original text entered to answer the question "ABC". Please make sure that your answer directly quotes the original text without any modification or explanation." Replace "ABC" in the above template with "elements to be extracted" to obtain the corresponding question.

[0061] This embodiment divides the extraction of the elements to be extracted from the review text into two steps. The first step is to extract the first context information associated with the elements to be extracted from the review text, and the second step is to determine the extraction result corresponding to the elements to be extracted from the first context information. When a large language model is used as an information extraction model, the large language model outputs a single answer to a single question, so the large language model needs to be trained. This implementation trains the large language model by constructing question-and-answer examples, instructing the large language model to decompose the questions into multiple tasks and answer these multiple tasks respectively. For example, the problem of obtaining the elements to be extracted from the review text is divided into two tasks. The first task is to determine the first context information from the review text based on the first prompt information, and the second task is to determine the extraction result corresponding to the elements to be extracted from the first context information based on the second prompt information.

[0062] In one embodiment, the process of constructing the question and answer example includes:

[0063] Get sample questions, sample texts, sample contexts, and sample extraction results;

[0064] Constructing a sample template based on the sample question, sample text, sample context and sample extraction results;

[0065] Determine the answer output format based on the sample template;

[0066] Determine the question and answer examples based on the sample template and answer output format.

[0067] Figure 3 A question-and-answer example framework diagram of a data processing method in an embodiment of the present disclosure is shown, such as Figure 3As shown, the question-answering example of this embodiment at least includes: a prompt word partial framework 311, an example template partial framework 312, and an answer output format partial framework 313.

[0068] Among them, the prompt word part framework 311 is used to set the role of the information extraction model, guide the information extraction model to perform the task corresponding to the role and output the corresponding content. Specifically, the information extraction model is used as a text review expert, then the prompt word part framework 311 includes: "You are now a text review expert, you need to perform the following tasks step by step according to the review text to obtain the elements to be extracted: Task 1: Please quote the original text to answer the question "{elements to be extracted}", please make sure that your answer directly quotes the original text without any modification or explanation. Task 2: Based on the answer result of Task 1, answer the following question: {Extraction result}". Through the above prompt word part framework 311, the information extraction model is positioned as a text review expert, so as to review the review text. In the prompt word part framework 311, Task 1 is a specific implementation method of the first prompt information, and Task 2 is a specific implementation method of the second prompt information.

[0069] The example template partial framework 312 is used to provide learning examples to the information extraction model so as to train the information extraction model to understand the audit text in the form of examples. N examples are obtained, each of which includes an example question, an example text, an example context, and an example extraction result. The example question, the example text, the example context, and the example extraction result are used as training data to train the information extraction model to construct an example template. The goal of training the information extraction model is to learn the example context as the first context information and the example extraction result as the extraction result of the element to be extracted.

[0070] The answer output format partial framework 313 is used to specify the output format of the information extraction model. After learning the example template, the information extraction model outputs the answers corresponding to the above-mentioned tasks 1 and 2 in accordance with the output format specified by the answer output format partial framework 313. Specifically, it may include "Please learn the above example template and output the {elements to be extracted} results in the following format: Task 1 answer: first context information; Task 2 answer: extraction result; if relevant content cannot be extracted, please use the "unknown" answer. Input text: {review text}".

[0071] The framework of the question-and-answer example can match content according to the usage scenario, thereby realizing the multi-scenario adaptation of the question-and-answer example framework. By using the large language model as the information extraction model's efficient and accurate understanding of natural language, the first prompt information and the second prompt information are generated according to the question-and-answer example and the question, providing a basis for the precise positioning of the elements to be extracted.

[0072] For example, the sample question is to answer "delivery time", the sample text is "Party B shall deliver the equipment to Party A's designated delivery location within [7] working days from the date of the effective date of the contract, and complete the (installation, commissioning) delivery of the equipment", the sample context is "Party B shall deliver the equipment to Party A's designated delivery location within 7 working days from the date of the effective date of the contract, and complete the installation, commissioning and delivery of the equipment", and the sample extraction result is "7 working days". The above examples are a set of examples. In actual application, multiple sets of examples can be used to strengthen the training guidance of the large language model. According to the sample question, sample text, sample context and sample extraction results, an example template is constructed, and the sample question is decomposed into multiple sub-questions in the example template. The method of decomposing a complex question into multiple sub-questions for solution and answer is called Few-Shot. This method can guide the large language model to decompose a complex question into multiple sub-questions for solution and answer when answering questions, thereby enhancing the reasoning ability of the large language model and achieving more accurate question answering. Based on the constructed example template and the format of the question constructed by the elements to be extracted, the answer output format of the large language model can be obtained. According to the example template and the answer output format, the question and answer example can be determined.

[0073] Figure 4 A specific schematic diagram of a question-and-answer example of a data processing method in an embodiment of the present disclosure is shown as follows: Figure 4 As shown, the question-answering example 41 includes at least: a prompt word part 411, an example template part 412, and an answer output format part 413. The prompt word part is used to use the large language model as the role of the questioned person, so that the large language model outputs the first context information and the extraction result according to the review text.

[0074] Taking contract review as an example, the prompt word part 411 guides the large language model to answer the question as the review expert being asked.

[0075] Taking contract review as an example, sample questions, sample text, sample context and sample extraction results are obtained from historically reviewed contract texts to construct a sample template; as shown in the sample template part 412, two examples are listed.

[0076] Example 1 requires extracting a specific delivery time, and the example question includes "specific delivery time"; the example text includes "Delivery time: Party B shall deliver the equipment to Party A's designated delivery location within [7] working days from the date the contract takes effect, and complete the (installation, commissioning) delivery of the equipment at the same time"; the example context is the original text content associated with the instance question extracted from the example text, including "Party B shall deliver the equipment to Party A's designated delivery location within [7] working days from the date the contract takes effect, and complete the (installation, commissioning) delivery of the equipment at the same time"; the example extraction result is the corresponding extraction result extracted from the example context, including "- Delivery time-: 7 working days".

[0077] Example 2 requires extracting a specific delivery time. The example question includes "specific delivery time"; the example text includes "Project results delivery time: Party B shall complete the software and implementation customized service development (including technical testing) agreed in this Agreement on [January]

[25]

[2023] and deliver the function to Party A"; the example context is the original text content associated with the example question extracted from the example text, including "Party B shall complete the software and implementation customized service development (including technical testing) agreed in this Agreement on [January]

[25]

[2023] and deliver the function to Party A"; the example extraction result is the corresponding extraction result extracted from the example context, including "- Delivery time-: January 25, 2023".

[0078] The answer output format is used to specify the format of the information extraction model output answer. It is necessary to guide the information extraction model to output the answer result according to the specified style and answer output format. Take the above task 1 as the first prompt information, output the corresponding first context information, take the above task 2 as the second prompt information, and output the extraction result of the corresponding element to be extracted.

[0079] In an embodiment, based on the first prompt information, the first context information associated with the element to be extracted is determined from the review text through an information extraction model, including: using a large language model to perform semantic understanding on the review text, and determining the first context information associated with the element to be extracted from the review text according to the first prompt information.

[0080] Based on the second prompt information, an extraction result corresponding to the element to be extracted is determined from the first context information through an information extraction model, including: using a large language model to perform semantic understanding on the first context information, and determining the extraction result corresponding to the element to be extracted from the first context information according to the second prompt information.

[0081] The method provided by the embodiment of the present disclosure decomposes the problem of obtaining the extraction result of the element to be extracted into two steps.

[0082] The first step is to determine the first context information corresponding to the element to be extracted, which is to use the large language model to perform semantic understanding on the audit text, guide the large language model, answer the content related to the element to be extracted, and use the answer content as the answer to obtain the first context information; wherein, the first context information is an approximate context, which refers to the original content corresponding to the element to be extracted output by the large language model after the large language model performs semantic understanding on the audit text. The first context information obtained by using the large language model is output character by character, which is essentially the original text. Determining the first context information corresponding to the element to be extracted allows the large language model to quickly locate the context where the element to be extracted is located, and can avoid the problem of easy confusion of extraction results when there are multiple similar elements to be extracted in the audit text at the same time. For example, when the contract contains both Party A's contact address and Party A's invoicing information address, the large language model is likely to confuse the results of the two when extracting the invoicing information address. The first context information of this embodiment can clearly locate the invoicing information address.

[0083] The second step is to determine the extraction result corresponding to the element to be extracted. The first context information corresponding to the element to be extracted is semantically understood by using a large language model. The extraction result corresponding to the element to be extracted is determined from the first context information corresponding to the element to be extracted according to the second prompt information. Since the first context information includes the context of the element to be extracted, the extraction result corresponding to the element to be extracted can be extracted by the first context information. The extraction result of the element to be extracted is determined based on the first context information. The first context information is used as the basis for locating the original text of the element to be extracted, and the original text segment of the element to be extracted in the audit text can be accurately located. Before the extraction of the element to be extracted, the context of the element to be extracted is obtained first, which can solve the problem that the extraction result is easily confused when there are multiple similar elements to be extracted in the audit text, and the accuracy of the extraction of the element to be extracted is achieved. Compared with extracting elements from the full text of the audit text, the extraction result corresponding to the first context information in this embodiment is extracted from The extraction range is narrowed, which improves the efficiency of extraction on the one hand, and improves the accuracy of extraction on the other hand.

[0084] In an embodiment, determining the position information of the extraction result in the audit text according to the first context information includes:

[0085] Step S502: Divide the audit text according to the division rule to obtain text segments, and vectorize the text segments to obtain text segment vectors.

[0086] Step S504: vectorize the first context information to determine a first context vector.

[0087] Step S506: Determine the position information of the extraction result in the audit text according to the similarity between the text segment vector and the first context vector.

[0088] Based on the above, the first context information obtained by the information extraction model is obtained by the information extraction model, which may not be completely consistent with the original text, such as having more or fewer characters, but the semantics are exactly the same as the original text. Therefore, the specific location of the element to be extracted cannot be obtained by the context alone, so the context needs to be matched with the audit text. The audit text is a text with a large number of characters and segmented. The audit text itself has line numbers and page numbers. When it is necessary to locate the element to be extracted, the line number and page number can be used to locate the specific location. If the context is matched directly with the audit text, the matching efficiency will be relatively low. Each time a to-be-extracted element is matched, the full text of the audit text needs to be matched once. When there are multiple similar elements, multiple matches are required, which is time-consuming and inefficient. Therefore, it is necessary to divide the audit text into multiple text segments according to the division rules, and input the text segments into the pre-trained vector model for vectorization to obtain the text segment vector. Similarly, the first context information is input into the pre-trained vector model for vectorization to obtain the first context vector. The pre-trained vector model can adopt the GTE model (Gegeral Text Embeddings). The text segment vector and the first context vector are compared for similarity to determine the position information of the extraction result in the audit text. This embodiment solves the problem of low efficiency caused by the large amount of calculation for text-to-text matching. By comparing the text segment vector and the first context vector for similarity, the amount of calculation between the vectors is significantly reduced, and the matching accuracy between the vectors is higher.

[0089] In an embodiment, the division rule includes any one or more of a character rule and a line break rule; when the division rule includes a character rule, a text segment includes a set number of characters; when the division rule includes a line break rule, a text segment includes characters located between two adjacent line breaks. There are multiple ways to divide the audit text. By dividing the audit text in multiple ways, the audit text can be formed into segments of multiple different dimensions, and multiple text segments of different types can be determined, which can reduce the probability of errors during vector matching.

[0090] This embodiment adopts character rules and line break rules, and the number of text fragments obtained is multiple. When the character rule is adopted, the audit text is divided according to the set number of characters to obtain multiple text fragments, and each text fragment includes a set number of characters. For example, the number of characters K is set to 32, and the audit text is divided into parts according to the number of 32 characters to obtain multiple text fragments. When the line break rule is adopted, the audit text is divided according to the line break to obtain multiple text fragments, and the characters between two adjacent line breaks constitute a text fragment, that is, the paragraph is divided to obtain multiple text fragments. For example, the audit text includes 4000 words and 50 paragraphs. When the character rule is adopted, 125 text fragments are obtained, and the 125 text fragment vectors are vectorized to obtain text fragment vectors; when the line break rule is adopted, 50 text fragments are obtained, and the 50 text fragment vectors are vectorized to obtain 50 text fragment vectors.

[0091] In the embodiment, the text segment vector includes multiple text segment vectors corresponding to the multiple text segments; each text segment has a position identifier; according to the similarity between the text segment vector and the first context vector, determining the position information of the extraction result in the audit text includes:

[0092] Step S602: determining a first vector from a plurality of text segment vectors according to the similarity between the first context vector and the text segment vector.

[0093] Step S604: Determine the position information of the extraction result in the audit text according to the position identifier corresponding to the first vector, the position identifier corresponding to the first vector refers to the position identifier corresponding to the target text segment, and the target text segment refers to the text segment corresponding to the first vector.

[0094] Each text segment has its own location identifier, including line number, page number and paragraph number. When the first context vector is compared with the text segment vector for similarity,

[0095] The meaning expressed by the first context information is similar to the meaning expressed by the original text to be reviewed. The first context vector is obtained through vectorization, and multiple text fragment vectors are obtained by dividing the review text and then vectorizing it. The original text is not modified in content, so it can represent the original text of the review text. The similarity between the first context vector and the multiple text fragment vectors is calculated, and the first context vector is used to perform similarity matching with the multiple text fragment vectors to filter out the first vector from the multiple text fragment vectors.

[0096] Specifically, when calculating the similarity between the first context vector and the multiple text segment vectors, the cosine similarity is used for calculation, that is, the cosine similarity cosθ1 between the first context vector and the multiple text segment vectors is calculated. The closer the cosine similarity cosθ1 is to 1, the larger the value of the cosine similarity cosθ1 is, which means that the two vectors are more similar, indicating that the text meanings of the two vectors are closer. The closer the cosine similarity cosθ1 is to 0, the smaller the value of the cosine similarity cosθ1 is, which means that the two vectors are less similar, indicating that the text meanings of the two vectors are more different.

[0097] After calculating the cosine similarity between the first context vector and each of the multiple text segment vectors, the cosine similarities are sorted from large to small, and the text segment vector with the largest cosine similarity is the most similar to the first context vector. In one case, the text segment vector with the largest cosine similarity is determined as the first vector, which has the highest requirements for the positioning of the extraction result; in another case, the text segment vectors corresponding to the top N (N is a positive integer) cosine similarities are determined as the first vector, which is relatively generalized for the positioning of the extraction result and can achieve a wider positioning of the extraction result at the cost of a certain degree of accuracy.

[0098] After obtaining the first vector, obtain the text segment corresponding to the first vector, obtain the corresponding position identifier according to the text segment, and obtain the position information of the extraction result in the audit text through the position identifier. By calculating the cosine similarity between the first context vector and the text segment vector, find the first vector closest to the first context vector in the text segment vector, and then determine the position identifier of the first vector to determine the clerical information of the extraction result in the audit text, so as to achieve accurate positioning of the position of the element to be extracted in the audit text.

[0099] In the embodiment, the text segment vector includes: a plurality of first-category text segment vectors and a plurality of second-category text segment vectors; the similarity between the text segment vector and the first context vector includes the similarity between each first-category text segment vector and the first context vector, and the similarity between each second-category text segment vector and the first context vector;

[0100] According to the similarity between the text segment vector and the first context vector, the position information of the extraction result in the audit text is determined, including:

[0101] Step S702: determining a second vector from a plurality of first-category text segment vectors based on the similarity between each first-category text segment vector and the first context vector.

[0102] Step S704: determining a third vector from a plurality of second-category text segment vectors based on the similarity between each second-category text segment vector and the first context vector.

[0103] Step S706: Determine the position information of the extraction result in the audit text according to the position identifier corresponding to the second vector and the position identifier corresponding to the third vector.

[0104] The text segment vectors are divided into different types of text segments according to different division rules and then vectorized to obtain first-category text segment vectors and second-category text segment vectors. For example, the multiple text segments obtained by dividing the audit text according to the character rule are multiple first-category text segments, and the multiple first-category text segments are vectorized to obtain multiple first-category text segment vectors. The multiple text segments obtained by dividing the audit text according to the line break rule are multiple second-category text segments, and the multiple second-category text segments are vectorized to obtain multiple second-category text segment vectors. In the case where the text segment vectors are divided into two types, the similarity calculation also needs to be calculated separately.

[0105] Calculate the cosine similarity between each first-category text segment vector and the first context vector, sort the cosine similarities from large to small, and determine the first-category text segment vector with the largest cosine similarity or the first-category text segment vectors corresponding to the first N cosine similarities as the second vector; determine the corresponding first-category text segment according to the second vector, and determine the corresponding position identifier according to the corresponding first-category text segment, so as to obtain the position identifier corresponding to the second vector.

[0106] Calculate the cosine similarity between each second-category text segment vector and the first context vector, sort the cosine similarities from large to small, and determine the second-category text segment vector with the largest cosine similarity or the second-category text segment vectors corresponding to the first N cosine similarities as the third vector; determine the corresponding second-category text segment according to the third vector, and determine the corresponding position identifier according to the corresponding second-category text segment, so as to obtain the position identifier corresponding to the third vector.

[0107] According to the position identifier corresponding to the second vector and the position identifier corresponding to the third vector, the position information of the extraction result in the audit text is determined.

[0108] After determining that the first-category text segment vector and the second-category text segment vector with the largest cosine similarity are the most similar to the context in terms of the specific meaning of the text, the position of the element to be extracted in the audit text can be accurately located by locating the position markers that divide the text segments.

[0109] In the embodiment, determining the position information of the extraction result in the audit text according to the position identifier corresponding to the second vector and the position identifier corresponding to the third vector includes:

[0110] The position identifier corresponding to the second vector and the position identifier corresponding to the third vector are processed together to determine the line number, page number and paragraph number of the extraction result; according to the line number, page number and paragraph number of the extraction result, the position of the extraction result in the audit text is determined. The position identifier contains the line number, page number and paragraph number, so the position of the extraction result in the audit text can be directly and accurately located.

[0111] The disclosed embodiment uses a large language model to obtain the first context information and extraction results, and uses the first context information as the basis for locating the original text of the element to be extracted, thereby achieving the accuracy of extracting the element to be extracted. Extracting the corresponding extraction results from the first context information narrows the extraction range, which improves the efficiency of extraction on the one hand, and the accuracy of extraction on the other hand; vectorizing the first context information to obtain a first context vector; dividing and vectorizing the review text to obtain multiple text segment vectors, using the first context vector to retrieve the element original text segment with the highest similarity from the multiple text segment vectors, and returning the union of the two segment line numbers as the line number of the review text segment corresponding to the element to be extracted, which can achieve accurate positioning of the first context segment where the element to be extracted is located. Reduce the time that legal personnel need to associate the element to be extracted with the original text position multiple times when reviewing the review results.

[0112] In an embodiment, determining the position information of the extraction result in the review text according to the first context information includes:

[0113] Step S802: deleting the first context information and the punctuation marks in the review text to obtain first context characters and review text characters;

[0114] Step S804: using the first N characters and the last N characters of the first context character to match the characters of the audit text; wherein N is an integer greater than or equal to 1;

[0115] Step S806: determine the position information of the successfully matched characters in the audit text as the position information of the extraction result in the audit text.

[0116] In the embodiment, when only punctuation marks are different between the first context information and the audit text, another solution for quickly locating the elements to be extracted is provided. First, regular expressions are used to delete punctuation marks for the context information and the audit text corresponding to the elements to be extracted, and the first context characters and the audit text characters are determined. The first N characters of the first context characters and the last N characters of the first context characters are taken as keywords, and matching is performed from the audit text characters. If the match is successful, the position of the successful match in the audit text characters is determined as the position of the elements to be extracted in the audit text. Since only punctuation marks are different between the first context information and the audit text, the characters obtained after deleting the punctuation marks are similar or even the same. Then, by dividing a context character into the first N characters and the last N characters, the corresponding position can be directly matched in the audit text characters. The value of N determines the speed and accuracy of the search. When N is large, the search speed is slow, but the accuracy is high; when N is small, the search speed is fast, but the accuracy is reduced. In the specific implementation, the value of N needs to be determined according to the content and length of the audit text. In this way, since regular expressions are used to delete punctuation marks, different punctuation marks are deleted, which can improve the speed of locating the elements to be extracted.

[0117] Figure 5 A framework diagram of a data processing method in an embodiment of the present disclosure is shown, Figure 5 As shown, when locating elements, a question-and-answer example is constructed according to the elements to be extracted, and the question-and-answer example and the audit text are input into the large language model. The large language model is used to determine the first context information and the extraction result corresponding to the elements to be extracted from the audit text based on the question-and-answer example, and the first context information is vectorized by the pre-trained vector model to obtain a first context vector; the audit text is divided according to the set number of characters, and a plurality of first-category text segment vectors are obtained by vectorization through the pre-trained vector model, and the vectors are input into vector database 1; the audit text is divided according to the line breaks, and a plurality of second-category text segment vectors are obtained by vectorization through the pre-trained vector model, and the vectors are input into vector database 2; the first The context vector is matched with the vector database 1 for similarity to obtain the most similar Top1 in the vector database 1, the first-category text fragment vector with the largest cosine similarity is determined as the second vector, and the first position information of the corresponding text fragment in the audit text is determined; the first context vector is matched with the vector database 2 for similarity to obtain the most similar Top1 in the vector database 2, the second-category text fragment vector with the largest cosine similarity is determined as the third vector, and the second position information of the corresponding text fragment in the audit text is determined; the position of the extraction result in the audit text is determined according to the union of the position identifier corresponding to the second vector and the position identifier corresponding to the third vector.

[0118] One of the key steps of the disclosed embodiment is to solve the factor extraction problem in steps. The first step is to generate the approximate context where the factor is located, and the second step is to extract the extraction result of the factor according to the approximate context. The second key step of the disclosed embodiment is to segment the contract text according to the fixed number of characters and line breaks, and then obtain the vectors of the corresponding original text segments through the pre-trained vector model, and store them in two vector libraries respectively. The approximate context corresponding to the factor is also passed through the pre-trained vector model, and the original text segments of the factor with the highest similarity are retrieved from the two vector libraries respectively, and the union of the line numbers of the two segments is returned as the line number of the contract text segment corresponding to the factor. Before the factor is extracted, the disclosed embodiment first obtains the approximate context where the factor is located, which can solve the problem that the extracted value is easily confused when there are multiple approximate factors in the contract. At the same time, determining the extraction result based on the first context narrows the extraction range and improves the efficiency and accuracy of the extraction. The approximate context of the element is generated using a large language model, and then the original text segment of the element with the highest similarity is retrieved from the two original text segment vector libraries. The union of the two segment line numbers is returned as the line number of the contract text segment corresponding to the element, which can achieve the precise positioning of the context segment where the element is located. The problem of contract element extraction is solved in steps. The first step is to generate the approximate context where the element is located, and the second step is to extract the extraction result of the element based on the approximate context. In the first step, when solving the confusion problem of the element extraction result, it serves as the basis for locating the original context of the element, narrowing the extraction scope. Vector retrieval can accurately locate the original text position of the element, solving the problem of easy confusion in positioning when there are multiple similar elements to be extracted in the audit text at the same time, and realizing the precise positioning of the original text position of the extraction result in the audit text, reducing the time for legal personnel to repeatedly associate elements with the original text position when reviewing the review results.

[0119] It should be noted that the acquisition, storage, use, and processing of data in the technical solution disclosed in this disclosure are in compliance with the relevant provisions of national laws and regulations, and various types of data such as personal identity data, operation data, behavioral data, etc. related to individuals, customers, and groups obtained in the embodiments of this disclosure have all been authorized.

[0120] Based on the same inventive concept, the present disclosure also provides a data processing device in the following embodiments. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0121] Figure 6 A schematic diagram of a data processing device in an embodiment of the present disclosure is shown. Figure 6 As shown, the device comprises:

[0122] The context information determination module 601 is used to determine the first context information associated with the element to be extracted from the review text through the information extraction model based on the first prompt information.

[0123] The extraction result determination module 602 is used to determine the extraction result corresponding to the element to be extracted from the first context information through the information extraction model based on the second prompt information.

[0124] The position information determination module 603 is used to determine the position information of the extraction result in the review text according to the first context information.

[0125] It should be noted that the above-mentioned context information determination module 601, extraction result determination module 602 and location information determination module 603 correspond to S202 to S206 in the method embodiment, and the examples and application scenarios implemented by the above-mentioned modules and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned method embodiment. It should be noted that the above-mentioned modules as part of the device can be executed in a computer system such as a set of computer executable instructions.

[0126] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0127] Reference below Figure 7 , the electronic device is exemplarily described in the form of a general-purpose computing device. It should be understood that Figure 7 The electronic device 70 shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0128] like Figure 7 As shown, the electronic device 70 includes: at least one processor 701; at least one memory 702, and one or more I / O interfaces 703, connected between the processor 701 and the memory 702; wherein the memory 702 stores one or more computer programs that can be executed by at least one processor 701, and the one or more computer programs are executed by at least one processor 701 so that the at least one processor 701 can execute the above-mentioned method of searching for web pages.

[0129] Each module in the above electronic device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0130] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0131] The present disclosure also provides a computer-readable storage medium on which a computer program is stored, wherein the computer program implements the above-mentioned data processing method when executed by a processor / processing core. The computer-readable storage medium may be a volatile or non-volatile computer-readable storage medium.

[0132] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above-mentioned data processing method.

[0133] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, which may be a readable signal medium or a readable storage medium. A program product capable of implementing the above-mentioned data processing method of the present disclosure is stored thereon. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above-mentioned "Exemplary Method" section of this specification.

[0134] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0135] In the present disclosure, a computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0136] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0137] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0138] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0139] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0140] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0141] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A data processing method, characterized in that: include: Based on the first prompt information, determining first context information associated with the element to be extracted from the review text through an information extraction model; Based on the second prompt information, determining an extraction result corresponding to the element to be extracted from the first context information through the information extraction model; The location information of the extraction result in the review text is determined according to the first context information.

2. The data processing method according to claim 1, characterized in that: The step of determining the position information of the extraction result in the audit text according to the first context information includes: Dividing the audit text according to the division rule to obtain text segments, and vectorizing the text segments to obtain text segment vectors; Vectorizing the first context information to determine a first context vector; According to the similarity between the text segment vector and the first context vector, the position information of the extraction result in the audit text is determined.

3. The data processing method according to claim 2, characterized in that: The division rule includes any one or more of a character rule and a line break rule; In the case where the segmentation rule includes the character rule, a text segment includes a set number of characters; In the case where the division rule includes the line break rule, a text segment includes characters located between two adjacent line breaks.

4. The data processing method according to claim 2, characterized in that: The number of the text segments is multiple; the text segment vector includes multiple text segment vectors corresponding to the multiple text segments; each text segment has a position identifier; The step of determining the position information of the extraction result in the audit text according to the similarity between the text segment vector and the first context vector includes: Determining a first vector from a plurality of text segment vectors according to a similarity between the first context vector and the text segment vector; According to the position identifier corresponding to the first vector, the position information of the extraction result in the audit text is determined. The position identifier corresponding to the first vector refers to the position identifier corresponding to the target text segment, and the target text segment refers to the text segment corresponding to the first vector.

5. The data processing method according to claim 2, characterized in that: The text segment vector includes: a plurality of first-category text segment vectors and a plurality of second-category text segment vectors; the similarity between the text segment vector and the first context vector includes the similarity between each first-category text segment vector and the first context vector, and the similarity between each second-category text segment vector and the first context vector; The step of determining the position information of the extraction result in the audit text according to the similarity between the text segment vector and the first context vector includes: Determine a second vector from a plurality of first-category text segment vectors based on the similarity between each first-category text segment vector and the first context vector; Determine a third vector from a plurality of second-category text segment vectors based on the similarity between each second-category text segment vector and the first context vector; According to the position identifier corresponding to the second vector and the position identifier corresponding to the third vector, the position information of the extraction result in the review text is determined.

6. The data processing method according to claim 1, characterized in that: The step of determining the position information of the extraction result in the audit text according to the first context information includes: Deleting the first context information and the punctuation marks in the review text to obtain first context characters and review text characters; Using the first N characters and the last N characters of the first context character to match the audit text characters; wherein N is an integer greater than or equal to 1; The position information of the successfully matched characters in the audit text is determined as the position information of the extraction result in the audit text.

7. A data processing device, characterized in that: include: A context information determination module, used to determine first context information associated with the element to be extracted from the audit text through an information extraction model based on the first prompt information; An extraction result determination module, used to determine an extraction result corresponding to the element to be extracted from the first context information through the information extraction model based on the second prompt information; A position information determination module is used to determine the position information of the extraction result in the review text according to the first context information.

8. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the data processing method according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the data processing method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the data processing method according to any one of claims 1 to 6 is implemented.

Images