Content auditing method and device, electronic equipment, storage medium and program product

Abstract

The invention provides a content auditing method and device, electronic equipment, a storage medium and a program product. The method comprises the steps that to-be-processed content is acquired; obtaining at least one to-be-extracted element corresponding to the to-be-processed content; according to the at least one to-be-extracted element, screening out a target element extraction model from the at least one element extraction model; wherein the target element extraction model is used for extracting an element entity corresponding to each first element in the at least one first element, and a set formed by the at least one first element is a subset of a set formed by the at least one element to be extracted; for each target element extraction model, inputting the to-be-processed content into the target element extraction model to obtain a first element entity corresponding to each first element; according to the first element entity corresponding to each target element extraction model, whether the to-be-processed content passes auditing or not is determined, the content can be analyzed from different angles, and therefore the auditing accuracy is improved.

Description

Technical Field

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

[0002] With the popularization and development of the Internet, a large amount of information spreads rapidly on the Internet. However, this also brings some problems, such as false information, malicious attacks, and privacy violations. In order to maintain the security and order of cyberspace, intelligent audit technology came into being.

[0003] Existing intelligent audit methods often rely on a single model, such as keyword-based methods, deep learning-based methods, etc., which are unable to accurately identify malicious content in some cases, resulting in low accuracy of audit results. Summary of the invention

[0004] The present application provides a content review method, device, electronic device, storage medium and program product to improve the accuracy of review results.

[0005] In a first aspect, an embodiment of the present application provides a content review method, comprising:

[0006] Get the content to be processed;

[0007] Acquire at least one element to be extracted corresponding to the content to be processed;

[0008] According to the at least one element to be extracted, a target element extraction model is screened out from at least one element extraction model; wherein the target element extraction model is used to extract element entities corresponding to each first element in at least one first element, wherein the set formed by the at least one first element is a subset of the set formed by the at least one element to be extracted;

[0009] For each target element extraction model, input the content to be processed into the target element extraction model to obtain the first element entity corresponding to each first element;

[0010] According to the first element entities respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the review.

[0011] Optionally, the content to be processed is structured text, and the method further includes:

[0012] Determining the type of the content to be processed;

[0013] Acquire at least one second element, wherein the at least one second element is selected from at least one element to be extracted;

[0014] Determining a target regular expression corresponding to the content to be processed according to the type and at least one second element;

[0015] Using the target regular expression, extracting second element entities corresponding to each second element from the content to be processed;

[0016] Correspondingly, according to the first element entities respectively corresponding to each target element extraction model, determining whether the content to be processed has passed the review includes:

[0017] According to the first element entity and at least one second element entity respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the review.

[0018] Optionally, obtain the content to be processed, including:

[0019] Acquire original content, wherein the original content includes at least one column and content corresponding to each column;

[0020] Determine a first field from at least one field, divide the first field into a plurality of sub-fields for each first field, and determine content corresponding to each of the plurality of sub-fields;

[0021] The content to be processed is determined according to the content corresponding to each sub-field in the plurality of sub-fields.

[0022] Optionally, determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes:

[0023] Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element;

[0024] For any audit indicator of any target element extraction model, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to the target element extraction model, determine the audit result of the audit indicator according to the constraint conditions corresponding to the audit indicator and the first target element entity, and determine the audit result corresponding to the target element extraction model according to the audit results corresponding to each audit indicator;

[0025] According to the audit results corresponding to each target element extraction model, it is determined whether the content to be processed has passed the audit.

[0026] Optionally, determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes:

[0027] Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element;

[0028] For any audit indicator, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to each target element extraction model; and determine the audit result corresponding to the audit indicator according to the constraint condition corresponding to the audit indicator and the first target element entity;

[0029] According to the audit results corresponding to each audit indicator, it is determined whether the content to be processed has passed the audit.

[0030] Optionally, determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes:

[0031] Acquire at least one audit indicator corresponding to the content to be processed, and select a target audit indicator from the at least one audit indicator;

[0032] Determine the second target elements and constraints corresponding to each target audit indicator, wherein the constraints are used to indicate the rules satisfied by the second target element entity corresponding to the second target element;

[0033] For any target audit indicator, perform the following audit operations until the audit result of the target audit indicator is not passed;

[0034] When the audit result of any target audit indicator is not passed, it is determined that the pending content has not passed the audit;

[0035] The audit operation includes:

[0036] Select the second target element entity corresponding to the current target audit indicator from the first element entities corresponding to each target element extraction model;

[0037] When the second target element entity satisfies the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be passed, the next-order target audit indicator is used as the current target audit indicator, and the second target element entity is continuously extracted;

[0038] When the second target element entity does not satisfy the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be failed.

[0039] In a second aspect, an embodiment of the present application provides a content review device, including:

[0040] A first acquisition module, used to acquire content to be processed;

[0041] A second acquisition module, used to acquire at least one element to be extracted corresponding to the content to be processed;

[0042] A screening module, used to screen out a target element extraction model from at least one element extraction model according to the at least one element to be extracted; wherein the target element extraction model is used to extract element entities corresponding to each first element in at least one first element, wherein the set formed by the at least one first element is a subset of the set formed by the at least one element to be extracted;

[0043] An input module, for inputting the to-be-processed content into each target element extraction model to obtain a first element entity corresponding to each first element;

[0044] The determination module is used to determine whether the content to be processed has passed the review based on the first element entities corresponding to each target element extraction model.

[0045] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0046] The memory stores computer-executable instructions;

[0047] The processor executes the computer-executable instructions stored in the memory to implement the method described in the first aspect above.

[0048] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect above.

[0049] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0050] The present application provides a content review method, device, electronic device, storage medium and program product, the method comprising: obtaining content to be processed; obtaining at least one to-be-extracted element corresponding to the content to be processed; based on the at least one to-be-extracted element, screening out a target element extraction model from at least one element extraction model; wherein the target element extraction model is used to extract element entities corresponding to each first element of at least one first element, wherein the set consisting of at least one first element is a subset of the set consisting of at least one to-be-extracted element; for each target element extraction model, inputting the content to be processed into the target element extraction model to obtain first element entities corresponding to each first element; determining whether the content to be processed has passed the review based on the first element entities respectively corresponding to each target element extraction model, in this way, selecting different target element extraction models based on the elements to be extracted has high flexibility and can adapt to different types of content review requirements, each target element extraction model can run independently, which is convenient for system expansion and maintenance, and new models can be added to the system as needed to enhance the scalability of the system. Moreover, different target factor extraction models conduct audits on different factors. On the one hand, they can analyze the content from different angles, thereby improving the accuracy of the audit. On the other hand, they can reduce the adaptability of malicious content to a certain extent and improve the robustness of the audit. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0052] Figure 1 An application scenario diagram provided for an embodiment of the present application;

[0053] Figure 2 A flowchart of a content review method provided in an embodiment of the present application;

[0054] Figure 3 A schematic diagram of text preprocessing annotation entity arrangement provided in an embodiment of the present application;

[0055] Figure 4 A flowchart of another content review method provided in an embodiment of the present application;

[0056] Figure 5 A schematic diagram of the structure of a content review device provided in an embodiment of the present application;

[0057] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0058] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0059] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0060] Definitions of Key Terms:

[0061] Text annotation: refers to the use of tags or other forms of marking to identify specific elements in the text. Such elements can be names of people, places, proper nouns, verbs, adjectives, etc.

[0062] Word interpretation: refers to the explanation of the meaning of a word, including its semantics, context, usage, etc.

[0063] Text field: refers to a specific position or area in the text, such as a sentence, a word, a character, etc.

[0064] Label review: refers to the process of label review using artificial intelligence technology.

[0065] Parallel orchestration: means that each subtask can be executed independently or depend on each other. When there are dependencies between subtasks, it is necessary to ensure that they are executed in the correct order.

[0066] With the popularization and development of the Internet, a large amount of information is rapidly disseminated on the Internet. The rapid spread of this information has greatly promoted social progress and economic development, and changed people's lifestyles and working methods. However, this has also brought some problems, such as false information, malicious attacks, and privacy violations. False information not only misleads the public, but may also cause social panic; malicious attacks may lead to data leakage, system paralysis, and even pose a threat to national security; the problem of privacy violations is becoming increasingly serious, and the illegal acquisition and abuse of personal information is common.

[0067] In order to maintain the security and order of cyberspace, intelligent audit technology has emerged. Intelligent audit technology can automatically identify and process text, pictures, videos and other content, effectively reducing the cost and workload of manual audits and improving audit efficiency. However, existing intelligent audit methods often rely on a single model, such as keyword-based methods and deep learning-based methods. However, they still have certain limitations when facing complex and changing network environments. For example, in some cases, malicious content cannot be accurately identified, resulting in low accuracy of audit results.

[0068] In view of this, the present application provides a content audit method, firstly obtaining the content to be processed, then obtaining at least one to-be-extracted element corresponding to the content to be processed; according to the at least one to-be-extracted element, a target element extraction model is screened out from at least one element extraction model; wherein the target element extraction model is used to extract the element entities corresponding to each first element in at least one first element, wherein the set consisting of at least one first element is a subset of the set consisting of at least one to-be-extracted element; for each target element extraction model, the content to be processed is input into the target element extraction model to obtain the first element entities corresponding to each first element; according to the first element entities respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the audit, so that different target element extraction models are selected according to the elements to be extracted, which has high flexibility and can adapt to different types of content audit requirements, and each target element extraction model can run independently, which is convenient for the expansion and maintenance of the system, and the new model can be added to the system as needed to enhance the extensibility of the system. And different target element extraction models are audited for different elements, on the one hand, the content can be analyzed from different angles, thereby improving the accuracy of the audit, and on the other hand, the adaptability of malicious content can be reduced to a certain extent, and the robustness of the audit can be improved.

[0069] Figure 1 An application scenario diagram provided for an embodiment of the present application, such as Figure 1As shown, the server first obtains the content to be processed, and then obtains at least one element to be extracted corresponding to the content to be processed. Exemplarily, the at least one element to be extracted includes element 1, element 2, element 3, and element 4, and the at least one element extraction model includes model 1, model 2, and model 3, wherein model 1 is used to extract element entities corresponding to element 1, element 2, and element 3, model 2 is used to extract element entities corresponding to element 3 and element 4, and model 3 is used to extract element 5 and element 6. Therefore, model 1 and model 2 are selected as target element extraction models, and the content to be processed is input into model 1, and element entity 1, element entity 2, and first element entity 3 can be obtained. The content to be processed is input into model 2, and second element entity 3 and element entity 4 can be obtained. Finally, according to element entity 1, element entity 2, first element entity 3, second element entity 3, and element entity 4, it is determined whether the content to be processed has passed the review.

[0070] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0071] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0072] Figure 2 This is a flow chart of a content review method provided in an embodiment of the present application. The execution subject of this embodiment can be any device with data processing function. This application uses the server as the execution subject for specific description. Figure 2 As shown, a content review method provided by an embodiment of the present application may include:

[0073] Step 201: Obtain content to be processed.

[0074] The content to be processed may be any content that can be audited. For example, the content to be processed may be: text, pictures, videos, audios, documents and files, web pages and online content, application data, database records, codes and scripts, sensor data, multimedia combination content, network traffic and logs, etc.

[0075] Among them, multimedia combination content includes interactive content in various media forms such as text, pictures, videos, and audio.

[0076] Specifically, the server obtains content to be processed.

[0077] Step 202: Obtain at least one element to be extracted corresponding to the content to be processed.

[0078] Specifically, the server obtains at least one element to be extracted corresponding to the content to be processed.

[0079] In one example, when the content to be processed is a picture, the review of the picture may include review of image quality, picture content, copyright protection, privacy protection, advertising compliance, etc. Correspondingly, at least one element to be extracted may include: image clarity and resolution, text, copyright logo or watermark, brand logo or trademark, and character posture.

[0080] In another example, when the content to be processed is text, when reviewing whether the user's transaction behavior complies with anti-money laundering regulations, at least one element to be extracted may include: the amount of each transaction of the user within a preset period of time, the average amount of the user's transactions, the transaction frequency of the user within the preset period of time, the user's average transaction frequency, the user's trading counterparty, the type of each transaction of the user within the preset period of time, and the user's regular transaction type.

[0081] Step 203: Filter out a target element extraction model from at least one element extraction model based on the at least one element to be extracted; wherein the target element extraction model is used to extract element entities corresponding to each first element in at least one first element, wherein the set consisting of at least one first element is a subset of the set consisting of at least one element to be extracted.

[0082] Among them, the element entity is a specific example or value corresponding to the element to be extracted. For example, if the element to be extracted is the transaction amount, the corresponding element entity is the numerical value of the amount, such as 5,000 yuan.

[0083] Specifically, the server may select a target element extraction model from at least one element extraction model according to at least one element to be extracted. The target element extraction model is used to extract element entities corresponding to each first element in at least one first element.

[0084] Exemplarily, at least one element to be extracted may include: element 1, element 2...element n. The target element extraction model can extract the element entity corresponding to at least one element among element 1, element 2...element n. Exemplarily, one of them can be extracted, such as the element entity of element 1, or two of them can be extracted, such as the element entities of element 2 and element 4, or the element entities of n elements in total, element 1, element 2...element n, can be extracted.

[0085] Currently, in addition to extracting feature entities corresponding to at least one of feature 1, feature 2...feature n, the target feature extraction model can also extract feature entities corresponding to features other than feature 1, feature 2...feature n, such as feature n+1, feature n+2, etc. This application does not limit this, but the feature entities corresponding to the extracted features n+1 and feature n+2 are not processed.

[0086] Step 204: For each target element extraction model, input the content to be processed into the target element extraction model to obtain the first element entity corresponding to each first element.

[0087] Specifically, for each target element extraction model, the server inputs the content to be processed into the target element extraction model, and the target element extraction model can output the first element entities corresponding to each corresponding first element.

[0088] For example, when the content to be processed is text, a target feature extraction model is a BERT+CRF model. For a given text T, the BERT model first encodes it into a hidden state sequence H: H = BERT(T), where H = (h 1 ,h 2 ,...,h n ), h i is each word x i Next, the CRF model uses these hidden states to predict the label sequence Y:

[0089]

[0090] Where: φ(y i-1 ,y i ,H,i) is the feature function, representing the current label y i With the previous label y i-1 and the relationship between the hidden states H. θ is the weight parameter. Z(X) is the normalization factor. By maximizing P(Y|X), the optimal label sequence Y can be obtained:

[0091] Y * = arg max Y P(Y|X)

[0092] Finally, the label sequence Y * Decoded as the first element entity E B :

[0093] E B =Decode(Y * )

[0094] in It is the set of first-element entities extracted by the BERT+CRF model.

[0095] Step 205: Determine whether the content to be processed has passed the review based on the first element entities corresponding to each target element extraction model.

[0096] Specifically, the server determines whether the content to be processed has passed the review based on the first element entities corresponding to each target element extraction model.

[0097] In one example, at least one element to be extracted includes: image clarity and resolution, text, copyright logo or watermark, brand logo or trademark, and character posture; the element entity corresponding to "image clarity and resolution" can be used to determine whether the image is damaged, blurred or has other quality problems; the element entity corresponding to "text" can be used to detect whether it contains inappropriate language, hate speech, etc.; the element entity corresponding to "copyright logo or watermark" can be used to determine whether there are copyright issues; the element entity corresponding to "brand logo or trademark" can be used to ensure compliance with relevant regulations; the element entity corresponding to "character posture" can be used to determine whether there are dangerous behaviors or inappropriate demonstrations.

[0098] In another example, at least one element to be extracted includes: the amount of each transaction of the user within a preset period, the average amount of the user's transactions, the transaction frequency of the user within the preset period, the average transaction frequency of the user, the user's trading counterparty, the type of each transaction of the user within the preset period, and the user's regular transaction type.

[0099] When the number of factor entities corresponding to "the amount of a single transaction of the user within the preset period" exceeds the number of times corresponding to the factor entity of "the average amount of the user's transactions" exceeds the preset number of times, the user may be at risk of money laundering; when the factor entity corresponding to "the transaction frequency of the user within the preset period" is significantly higher than the factor entity corresponding to "the average transaction frequency of the user", the user may be at risk of money laundering; when the factor entity corresponding to "the type of user's transactions within the preset period" does not conform to the factor entity corresponding to "the user's regular transaction type" more than the preset number of times, the user may be at risk of money laundering; when the factor entity corresponding to "the user's counterparty" is a high-risk country, region or individual, it means that the user may be at risk of money laundering.

[0100] For the extracted first element entity, the following operations can be performed:

[0101] 1. Digital operations: for the extracted numerical type element entities Common mathematical operations (such as addition, subtraction, comparison) can be performed:

[0102] A. Addition operation: where αi Is a feature entity The weight of .

[0103] B. Difference calculation:

[0104] C. Numerical comparison:

[0105] 2. Date operation: for date type feature entities You can calculate the difference between dates and compare dates:

[0106] A. Date difference:

[0107] B. Date comparison:

[0108] 3. Logical operations: For other types of element entities (such as string and Boolean types), logical AND or OR operations can be performed:

[0109] A. Logical AND:

[0110] B. Logical OR:

[0111] The content audit method provided by the present application can obtain the content to be processed; obtain at least one element to be extracted corresponding to the content to be processed; according to the at least one element to be extracted, select the target element extraction model from at least one element extraction model; wherein the target element extraction model is used to extract the element entities corresponding to each first element in at least one first element, wherein the set composed of at least one first element is a subset of the set composed of at least one element to be extracted; for each target element extraction model, the content to be processed is input into the target element extraction model to obtain the first element entities corresponding to each first element; according to the first element entities respectively corresponding to each target element extraction model, determine whether the content to be processed has passed the audit, so that different target element extraction models are selected according to the elements to be extracted, which has high flexibility and can adapt to different types of content audit requirements. Each target element extraction model can run independently, which is convenient for system expansion and maintenance. New models can be added to the system as needed to enhance the extensibility of the system. And different target element extraction models audit different elements. On the one hand, they can analyze the content from different angles, thereby improving the accuracy of the audit, and on the other hand, they can reduce the adaptability of malicious content to a certain extent and improve the robustness of the audit.

[0112] Optionally, the content to be processed is structured text, and the content review method provided by the present application further includes:

[0113] Determining the type of the content to be processed;

[0114] Acquire at least one second element, wherein the at least one second element is selected from at least one element to be extracted;

[0115] Determining a target regular expression corresponding to the content to be processed according to the type and at least one second element;

[0116] Using the target regular expression, extracting second element entities corresponding to each second element from the content to be processed;

[0117] Correspondingly, according to the first element entities respectively corresponding to each target element extraction model, determining whether the content to be processed has passed the review includes:

[0118] According to the first element entity and at least one second element entity respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the review.

[0119] Specifically, when the content to be processed is structured text, the server can determine the type of the content to be processed and obtain at least one second element. The at least one second element is screened out from at least one element to be extracted and is an element of at least one element to be extracted whose corresponding element entity can be obtained through a regular expression.

[0120] According to the type of the content to be processed and at least one second element, a target regular expression corresponding to the content to be processed can be determined, and the second element entities corresponding to each second element can be extracted from the content to be processed by using the target regular expression;

[0121] For example, when the content to be processed is a SWIFT message, regular expressions can be used to match specific patterns in the SWIFT message to extract key information, such as transaction reference number, sender information, confirmation status, etc.

[0122] Table 1 is a mapping table of labels, specific contents and meanings of some regular expressions provided in the embodiments of the present application, as shown in Table 1,

[0123] “(41A|78|49H|72Z|72|47A|57A|57D)\s?:\s*.*?\s?(\d+(\.\d+)?\s*(CNY|USD|RMB))”

[0124] "(41A|78|49H|72Z|72|47A|57A|57D)\s?:\s*.*?\s?((USD|CNY|RMB)\s*((\d+(\.\d+)?)|\d+))" are two ways to obtain the amount by matching different types of structured text.

[0125] Table 1

[0126]

[0127] Set the content to be processed to T and the regular expression to R i It is used to match a specific pattern from the content to be processed. The extracted second element entity can be expressed as: For multiple regular expressions, it is represented as: Where E R It is a collection of second-factor entities extracted by regular expressions.

[0128] Finally, the server determines whether the content to be processed has passed the review based on the first element entity and at least one second element entity corresponding to each target element extraction model.

[0129] In this way, the target regular expression is used to quickly extract the entities corresponding to each second element from the structured text, which greatly improves the speed and efficiency of information extraction. In addition, the target regular expression and audit model are dynamically determined according to the type of content to be processed and the screened elements. It has high flexibility and can adapt to different types of text audit needs. Finally, a comprehensive evaluation is performed by combining the first element entity and the second element entity to ensure the comprehensiveness and accuracy of the audit results.

[0130] Optionally, obtain the content to be processed, including:

[0131] Acquire original content, wherein the original content includes at least one column and content corresponding to each column;

[0132] Determine a first field from at least one field, divide the first field into a plurality of sub-fields for each first field, and determine content corresponding to each of the plurality of sub-fields;

[0133] The content to be processed is determined according to the content corresponding to each sub-field in the plurality of sub-fields.

[0134] Specifically, the server obtains the original content, which is also in a structured form, including at least one field and the content corresponding to each field. The original content is processed according to the division of the fields in the original content, and the line breaks in the content corresponding to each field are removed. After processing, the content corresponding to one field is a line. Secondly, special symbols in the original content, such as &, <, etc., need to be removed.

[0135] Then, a first column is determined from at least one column, wherein the first column can be determined manually or automatically by a program, and this application does not limit this. The information corresponding to the first column is relatively complex and comprehensive, and the first column can be divided into multiple sub-columns, and the information corresponding to the sub-columns is more single and detailed information.

[0136] For example, a letter of credit is a payment guarantee issued by a bank, which contains detailed transaction conditions and requirements, such as description of goods, shipping date, document requirements, etc. The complex or comprehensive terms or information in the letter of credit can be broken down into more specific and detailed information units.

[0137] The content to be processed is determined according to the content corresponding to each sub-field in the plurality of sub-fields.

[0138] Table 2 is a schematic structural table of an original content provided by an embodiment of the present application, as shown in Table 2. Each column represents a field, and Table 2 includes 5 fields in total, namely field 1, field 2, field 3, field 4 and field 5.

[0139] Table 2

[0140] Field 1 Field 2 Field 3 Field 4 Column 5 Content 1 Content 2 Content 3 Content 4 Content 5

[0141] Among them, field 3 is the first field, field 3 is divided into multiple sub-fields, the multiple sub-fields are sub-field 31, sub-field 32, and sub-field 33, and the content corresponding to each sub-field is determined, which are content 31, content 32, and content 33 respectively. According to the content corresponding to each sub-field in the multiple sub-fields, the content to be processed is determined. Table 3 is a structural schematic table of the content to be processed provided in an embodiment of the present application.

[0142] Table 3

[0143] Field 1 Field 2 Field 31 Field 32 Field 33 Field 4 Column 5 Content 1 Content 2 Content 31 Content 32 Content 33 Content 4 Content 5

[0144] In this way, by dividing the complex first column into multiple sub-columns, the information can be further refined, so that the information corresponding to each sub-column is more unified and clear, thereby improving the precision and accuracy of data processing.

[0145] Optionally, determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes:

[0146] Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element;

[0147] For any audit indicator of any target element extraction model, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to the target element extraction model, determine the audit result of the audit indicator according to the constraint conditions corresponding to the audit indicator and the first target element entity, and determine the audit result corresponding to the target element extraction model according to the audit results corresponding to each audit indicator;

[0148] According to the audit results corresponding to each target element extraction model, it is determined whether the content to be processed has passed the audit.

[0149] Among them, the audit indicators are business logic conditions predefined according to the business.

[0150] Exemplarily, referring to the above embodiment of determining whether a user has money laundering risk, the audit indicators may include 4: the first audit indicator is the transaction amount indicator, and the constraint condition corresponding to the amount indicator is: the number of times the amount of a single transaction of a user is less than the average amount of the user's transaction within a preset period does not exceed the preset number of times; the second audit indicator is the transaction frequency indicator, and the constraint condition corresponding to the transaction frequency indicator is that the transaction frequency of the user within the preset period is significantly greater than the average transaction frequency of the user, where the significant standard can be freely set according to demand; the third audit indicator is the transaction type indicator, and the constraint condition corresponding to the transaction type indicator is: the number of times the type of the user's transaction within the preset period is not the user's regular transaction type does not exceed the preset number of times, where the regular transaction type can be obtained by counting the user's historical transactions, and the regular transaction type can be one or more. The fourth audit indicator is the counterparty indicator, and the constraint condition corresponding to the counterparty indicator is: the counterparty is not a high-risk country, region or individual.

[0151] Specifically, in an optional implementation, determining the audit result of the audit indicator according to the constraint condition corresponding to the audit indicator and the first target element entity, and determining the audit result corresponding to the target element extraction model according to the audit result corresponding to each audit indicator, including:

[0152] When the first target element entity corresponding to the audit indicator meets the corresponding constraint conditions, the audit result of the audit indicator is determined to be passed; when the first target element entity corresponding to the audit indicator does not meet the corresponding constraint conditions, the audit result of the audit indicator is determined to be failed. The audit results corresponding to each audit indicator are subjected to logical operations, usually AND or OR operations, and the audit results corresponding to the target element extraction model are determined according to the operation results.

[0153] Accordingly, in an optional implementation, determining whether the content to be processed has passed the review according to the review results corresponding to each target element extraction model includes:

[0154] When the audit result corresponding to any target factor extraction model is failed, it is determined that the content to be processed has failed the audit. When the audit results corresponding to each target factor extraction model are passed, it is determined that the content to be processed has passed the audit.

[0155] In another optional implementation, determining whether the content to be processed has passed the review according to the review results corresponding to each target element extraction model includes:

[0156] When the corresponding audit result is that the proportion of the passed target element extraction models to all target element extraction models is greater than the preset proportion, it is determined that the content to be processed has passed the audit. When the corresponding audit result is that the proportion of the passed target element extraction models to all target element extraction models is less than the preset proportion, it is determined that the content to be processed has not passed the audit.

[0157] Optionally, when the content to be processed is structured text, the second element entities extracted by the target regular expression can also be used to determine the audit results corresponding to each audit indicator, and then determine the audit results corresponding to the regular expression based on the audit results corresponding to each audit indicator. The specific determination method refers to the above-mentioned method of determining the audit results corresponding to each audit indicator based on each first element entity, and determining the audit results corresponding to the target element extraction model based on the audit results corresponding to each audit indicator, which will not be repeated here.

[0158] Finally, whether the content to be processed has passed the review is determined based on the review results corresponding to each target element extraction model and the review results corresponding to the regular expression.

[0159] In this way, by obtaining multiple audit indicators corresponding to the content to be processed and determining the specific first target element and constraint conditions for each audit indicator, the audit process can be refined to ensure that each audit indicator is accurately evaluated. By auditing different target element extraction models and audit indicators, the content to be processed can be comprehensively evaluated from multiple dimensions to ensure the comprehensiveness of the audit.

[0160] Optionally, determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes:

[0161] Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element;

[0162] For any audit indicator, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to each target element extraction model; and determine the audit result corresponding to the audit indicator according to the constraint condition corresponding to the audit indicator and the first target element entity;

[0163] According to the audit results corresponding to each audit indicator, it is determined whether the content to be processed has passed the audit.

[0164] Optionally, when the target element extraction model outputs the first element entity, each element entity has its corresponding confidence level, and the audit result corresponding to the audit indicator is determined according to the constraint condition corresponding to the audit indicator and the first target element entity, including:

[0165] For at least one first target element entity corresponding to the same element, determine the element entity with the highest confidence among the first target element entities as the second target element entity;

[0166] According to the multiple second target element entities, the audit results corresponding to the audit indicators are determined.

[0167] Specifically, when multiple second target element entities meet the constraint conditions corresponding to the audit indicator, the audit result corresponding to the audit indicator is determined to be passed; when multiple second target element entities do not meet the constraint conditions corresponding to the audit indicator, the audit result corresponding to the audit indicator is determined to be failed.

[0168] Accordingly, according to the audit results corresponding to each audit indicator, determining whether the content to be processed has passed the audit includes:

[0169] The audit results corresponding to each audit indicator are subjected to logical operations, usually AND or OR operations, and whether the content to be processed has passed the audit is determined based on the operation results.

[0170] In this way, the audit results corresponding to each audit indicator are determined by using the first target element entity output by multiple target element extraction models, which can further improve the accuracy of the determined audit results.

[0171] Optionally, determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes:

[0172] Acquire at least one audit indicator corresponding to the content to be processed, and select a target audit indicator from the at least one audit indicator;

[0173] Determine the second target elements and constraints corresponding to each target audit indicator, wherein the constraints are used to indicate the rules satisfied by the second target element entity corresponding to the second target element;

[0174] For any target audit indicator, perform the following audit operations until the audit result of the target audit indicator is not passed;

[0175] When the audit result of any target audit indicator is not passed, it is determined that the pending content has not passed the audit;

[0176] The audit operation includes:

[0177] Select the second target element entity corresponding to the current target audit indicator from the first element entities corresponding to each target element extraction model;

[0178] When the second target element entity satisfies the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be passed, the next-order target audit indicator is used as the current target audit indicator, and the second target element entity is continuously extracted;

[0179] When the second target element entity does not satisfy the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be failed.

[0180] Exemplarily, at least one audit indicator includes audit indicator 1, audit indicator 2, audit indicator 3 and audit indicator 4, wherein the target audit indicators screened manually or automatically are audit indicator 2 and audit indicator 3.

[0181] For audit indicator 2, the second target element entity corresponding to audit indicator 2 is determined. When the second target element entity corresponding to audit indicator 2 does not satisfy the constraint condition corresponding to audit indicator 2, it is determined that the content to be processed has failed the audit.

[0182] When the second target element entity corresponding to audit indicator 2 satisfies the constraint condition corresponding to audit indicator 2 , it is further determined whether the second target element entity corresponding to audit indicator 3 satisfies the constraint condition corresponding to audit indicator 3 .

[0183] That is, among the target audit indicators, as long as it can be determined that the second target element entity corresponding to one of the target audit indicators does not meet the constraint conditions, it can be determined that the content to be processed has not passed the audit.

[0184] In this way, by gradually reviewing the target audit indicators until the audit result of a certain indicator is failed, it is possible to quickly determine whether the content to be processed has passed the audit, avoid unnecessary repeated audits, and improve audit efficiency.

[0185] Figure 3 A schematic diagram of text preprocessing annotation entity arrangement provided in an embodiment of the present application, such as Figure 3 As shown, the main steps are divided into entity splitting, label classification, and annotation consistency.

[0186] Message input: Different message text inputs carry specific text features. Depending on the number of classification results, the model training data volume needs to reach about 300 to 500 classification label data volumes, and the amount of each label data volume invested in model training needs to be balanced.

[0187] Text preprocessing: mainly for data cleaning, 1) Remove duplicate data: This is to ensure the independence and consistency of the data and avoid analysis bias caused by duplicate data. 2) Fill missing values: For null values ​​or missing data in the data, you can choose to delete, fill or interpolate. 3) Process outliers: Detect and process outliers or error values ​​in the data, including but not limited to irrelevant URLs, spaces, letters, special symbols, etc. 4) Convert data formats: This is to make the data meet specific requirements or specifications, such as date format conversion, unit conversion, etc.

[0188] Splitting of labeled entities: Before data labeling, formulating standardized and accurate data specifications is a crucial part of ensuring the quality of labeling. The main function of data specifications for text classification is to clarify the boundaries between the classification results, minimize data entanglement problems, align the labeling standards of different labelers for labeling tasks, and ensure the consistency and accuracy of labeling. The following example is: In the intelligent audit element project, the MTXXX message is a multi-label classification, and a total of four intentions are required: ① "confirmation of receipt", ② "confirmation added", ③ "deletion of sanctions clauses", and ④ "clarification of clause content". In addition to the above four labels, ⑤ "non-deletion of sanctions clauses" and ⑥ "non-clarification of clause content" are also required. Among them, ①② are mutually exclusive, and one of them must be selected; to identify ③, for this classification, one of ③⑤ must be selected; to identify ④, for this classification, one of ④⑥ must be selected.

[0189] Entity label classification: 1) Label distribution in the annotated text. The amount of data for label classification needs to be evenly distributed. If there is no data to supplement in the actual business, the data can be prepared in batches by simulating data features. 2) "The two labels in the training set are entangled." First, check whether the classification boundaries of these data are clearly defined in the specification. If the classification boundaries are vague, the specification needs to be optimized and reviewed again. If the classification boundaries are clearly defined in the annotation specification, the original data or the annotated data needs to be quality checked for deviation errors. If there are errors, the entire amount needs to be corrected before optimization training.

[0190] Annotation text specifications: 1) Business logic boundaries: The main function of text classification data specifications is to clarify the boundaries between classification results, minimize data entanglement problems, and align the annotation standards of different annotators for annotation tasks to ensure annotation consistency and accuracy. 2) Annotation accuracy: In the case of complex requirements, even with the guidance of annotation specifications, there is still the possibility that the annotator will mislabel or miss a label. Therefore, for complex labeling situations, the annotation entity splitting and entity label classification are performed again.

[0191] Figure 4 A flowchart of another content review method provided in an embodiment of the present application is shown as follows: Figure 4 As shown, the engine processes: 1) Preprocessing: For example, the content of the message in the case is relatively structured, in the format of field + content. If there is cross-line annotation during annotation, the message should be processed according to the division of the fields in the message, and the line breaks in the message data should be removed. After processing, one field is a line, and the special symbols &< in the message need to be removed. 2) Element field splitting: refers to splitting the complex or comprehensive terms or information in the letter of credit into more specific and detailed information units. A letter of credit is a payment guarantee issued by a bank, which contains detailed transaction conditions and requirements, such as description of goods, shipping date, document requirements, etc.

[0192] Operational services: These rules use regular expressions to match specific patterns in SWIFT messages to extract key information. For example, banks or financial institutions can use these rules to automatically parse and process SWIFT messages to extract important data such as transaction reference numbers, sender information, confirmation status, etc.

[0193] 1. Hard entity extraction: Set the input text to T and the regular expression to R i Used to match specific patterns from text, the extracted elements can be expressed as: For multiple regular expressions, it is represented as: Where E R It is a collection of features extracted by regular expressions.

[0194] Soft entity extraction: For a given text T, the BERT model first encodes it into a hidden state sequence H: H = BERT(T), where H = (h 1 ,h 2 ,...,h n ), h i is each word x i Next, the CRF model uses these hidden states to predict the label sequence Y:

[0195]

[0196] Where: φ(y i-1 ,y i ,H,i) is the feature function, representing the current label y i With the previous label y i-1 and the relationship between the hidden states H. θ is the weight parameter. Z(X) is the normalization factor. By maximizing P(Y|X), the optimal label sequence Y can be obtained:

[0197] Y * = arg max Y P(Y|X)

[0198] Finally, the label sequence Y * Decoded into specific element E B :

[0199] E B =Decode(Y * )

[0200] in It is a set of features extracted by the BERT+CRF model.

[0201] 2. Entity Orchestration

[0202] For the feature E extracted by hard entity R and soft entity extracted element E B , according to the predefined business logic conditions L j Logical combinations can be expressed as Boolean functions:

[0203]

[0204] where f j is a Boolean function used to define a specific business logic condition. Assume that a business logic condition L j The specific form is:

[0205]

[0206] Among them, α i and β j is the importance weight of the corresponding element, and ^ represents the logical AND operation.

[0207] 3. Ability level expression:

[0208] Taking the scenario of intelligent audit as an example, setting the intelligent audit target D is based on all logical conditions L j The combined result is expressed as:

[0209] D=g(L 1 (E R ,E B ),L2 (E R ,E B ),...,L n (E R ,E B ))

[0210] Among them, g is a logical operation function, usually AND or OR operation. If D is true, it means the audit is passed; otherwise, it means the audit failed. We can further expand the specific form of logical combination, for example:

[0211]

[0212] Combining the above steps, the whole process can be summarized into a unified formula:

[0213]

[0214] This formula describes the entire process from information extraction, entity arrangement, and capability layer expression. Each step is based on the results of the previous step, forming a closed-loop element extraction process. In practical applications, it is not limited to intelligent auditing, but can be an algorithm or workflow in risk warning and intent recognition of banks or financial institutions.

[0215] The following is the confidence weight α for soft and hard entities i and β j Pseudo algorithm flow for updating:

[0216] Step 1: Initialize the soft and hard entity confidence weights α through prior knowledge setting i and β j

[0217] Step 2: Calculate the loss, the objective function is

[0218]

[0219] where λ is the regularization coefficient. k is the true value of the kth sample. is the predicted value of the logistic combination model for the kth sample, calculated by the following formula: Here x i,k and x j,k are the i-th and j-th elements in the input features, respectively, and the weight α i and β j Indicate the importance of these elements.

[0220] Step 3: Calculate the gradient and calculate the loss function for each weight α i and β j The partial derivative of is the gradient.

[0221] in is the momentum term and γ is the momentum coefficient.

[0222] Step 4: Update the weights. Update the weights according to the gradient and iterate using the weight update formula.

[0223] Step 5: Convergence detection. After the weights are updated several times, detect the change in the loss function. If the change is small or reaches the preset convergence condition, stop the iteration.

[0224] Corresponding to the above content review method, the embodiment of the present application also provides a content review device, Figure 5 A schematic diagram of the structure of a content review device provided in an embodiment of the present application. Figure 5 As shown, the device comprises:

[0225] A first acquisition module 501 is used to acquire content to be processed;

[0226] A second acquisition module 502 is used to acquire at least one element to be extracted corresponding to the content to be processed;

[0227] A screening module 503 is used to screen out a target element extraction model from at least one element extraction model according to the at least one element to be extracted; wherein the target element extraction model is used to extract element entities corresponding to each first element in at least one first element, wherein the set consisting of at least one first element is a subset of the set consisting of at least one element to be extracted;

[0228] An input module 504, for inputting the to-be-processed content into each target element extraction model to obtain a first element entity corresponding to each first element;

[0229] The determination module 505 is used to determine whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model.

[0230] Optionally, the content to be processed is structured text, and the screening module 503 is further used to:

[0231] Determining the type of the content to be processed;

[0232] Acquire at least one second element, wherein the at least one second element is selected from at least one element to be extracted;

[0233] Determine, according to the type and at least one second element, a target regular expression corresponding to the content to be processed;

[0234] Using the target regular expression, extracting second element entities corresponding to each second element from the content to be processed;

[0235] Accordingly, the determination module 505 is specifically configured to:

[0236] According to the first element entity and at least one second element entity respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the review.

[0237] Optionally, the first acquisition module 501 is specifically configured to:

[0238] Acquire original content, wherein the original content includes at least one column and content corresponding to each column;

[0239] Determine a first field from at least one field, divide the first field into a plurality of sub-fields for each first field, and determine content corresponding to each of the plurality of sub-fields;

[0240] The content to be processed is determined according to the content corresponding to each sub-field in the plurality of sub-fields.

[0241] Optionally, the determination module 505 is specifically configured to:

[0242] Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element;

[0243] For any audit indicator of any target element extraction model, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to the target element extraction model, determine the audit result of the audit indicator according to the constraint conditions corresponding to the audit indicator and the first target element entity, and determine the audit result corresponding to the target element extraction model according to the audit results corresponding to each audit indicator;

[0244] According to the audit results corresponding to each target element extraction model, it is determined whether the content to be processed has passed the audit.

[0245] Optionally, the determination module 505 is specifically configured to:

[0246] Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element;

[0247] For any audit indicator, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to each target element extraction model; and determine the audit result corresponding to the audit indicator according to the constraint condition corresponding to the audit indicator and the first target element entity;

[0248] According to the audit results corresponding to each audit indicator, it is determined whether the content to be processed has passed the audit.

[0249] Optionally, the determination module 505 is specifically configured to:

[0250] Acquire at least one audit indicator corresponding to the content to be processed, and select a target audit indicator from the at least one audit indicator;

[0251] Determine the second target elements and constraints corresponding to each target audit indicator, wherein the constraints are used to indicate the rules satisfied by the second target element entity corresponding to the second target element;

[0252] For any target audit indicator, perform the following audit operations until the audit result of the target audit indicator is not passed;

[0253] When the audit result of any target audit indicator is not passed, it is determined that the pending content has not passed the audit;

[0254] The audit operation includes:

[0255] Select the second target element entity corresponding to the current target audit indicator from the first element entities corresponding to each target element extraction model;

[0256] When the second target element entity satisfies the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be passed, the next-order target audit indicator is used as the current target audit indicator, and the second target element entity is continuously extracted;

[0257] When the second target element entity does not satisfy the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be failed.

[0258] The specific implementation principles and effects of the content review device provided in the embodiment of the present application can be found in the aforementioned embodiment and will not be repeated here.

[0259] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device of this embodiment may include:

[0260] at least one processor 601; and

[0261] A memory 602 communicatively connected to the at least one processor;

[0262] The memory 602 stores instructions that can be executed by the at least one processor 601, and the instructions are executed by the at least one processor 601 to enable the electronic device to execute the method described in any of the above embodiments.

[0263] Optionally, the memory 602 may be independent or integrated with the processor 601 .

[0264] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the aforementioned embodiments and will not be described in detail here.

[0265] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method described in any of the above embodiments is implemented.

[0266] An embodiment of the present application further provides a computer program product, including a computer program, which implements the method described in any of the above embodiments when executed by a processor.

[0267] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0268] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the method described in each embodiment of the present application.

[0269] It should be understood that the above-mentioned processor can be a processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the application can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor. The memory may include a high-speed random access memory (Random Access Memory, RAM), and may also include non-volatile memory (Non-Volatile Memory, NVM), such as at least one disk storage, and can also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0270] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0271] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0272] It should be noted that, in this article, the terms "include", "comprises" or any other variations 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, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0273] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0274] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0275] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A content review method, characterized in that: include: Get the content to be processed; Acquire at least one element to be extracted corresponding to the content to be processed; According to the at least one element to be extracted, a target element extraction model is screened out from at least one element extraction model; wherein the target element extraction model is used to extract element entities corresponding to each first element in at least one first element, wherein the set formed by the at least one first element is a subset of the set formed by the at least one element to be extracted; For each target element extraction model, input the content to be processed into the target element extraction model to obtain the first element entity corresponding to each first element; According to the first element entities respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the review.

2. The method according to claim 1, characterized in that: The content to be processed is structured text, and the method further includes: Determining the type of the content to be processed; Acquire at least one second element, wherein the at least one second element is selected from at least one element to be extracted; Determining a target regular expression corresponding to the content to be processed according to the type and at least one second element; Using the target regular expression, extracting second element entities corresponding to each second element from the content to be processed; Correspondingly, according to the first element entities respectively corresponding to each target element extraction model, determining whether the content to be processed has passed the review includes: According to the first element entity and at least one second element entity respectively corresponding to each target element extraction model, it is determined whether the content to be processed has passed the review.

3. The method according to claim 2, characterized in that Get the content to be processed, including: Acquire original content, wherein the original content includes at least one column and content corresponding to each column; Determine a first field from at least one field, divide the first field into a plurality of sub-fields for each first field, and determine content corresponding to each of the plurality of sub-fields; The content to be processed is determined according to the content corresponding to each sub-field in the plurality of sub-fields.

4. The method according to claim 1, characterized in that: Determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes: Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element; For any audit indicator of any target element extraction model, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to the target element extraction model, determine the audit result of the audit indicator according to the constraint conditions corresponding to the audit indicator and the first target element entity, and determine the audit result corresponding to the target element extraction model according to the audit results corresponding to each audit indicator; According to the audit results corresponding to each target element extraction model, it is determined whether the content to be processed has passed the audit.

5. The method according to claim 1, characterized in that Determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes: Acquire at least one audit indicator corresponding to the content to be processed, and determine the first target element and constraint conditions corresponding to each audit indicator, wherein the constraint conditions are used to indicate the rules satisfied by the first target element entity corresponding to the first target element; For any audit indicator, select the first target element entity corresponding to the audit indicator from the first element entities corresponding to each target element extraction model; and determine the audit result corresponding to the audit indicator according to the constraint condition corresponding to the audit indicator and the first target element entity; According to the audit results corresponding to each audit indicator, it is determined whether the content to be processed has passed the audit.

6. The method according to claim 1, characterized in that Determining whether the content to be processed has passed the review according to the first element entities respectively corresponding to each target element extraction model includes: Acquire at least one audit indicator corresponding to the content to be processed, and select a target audit indicator from the at least one audit indicator; Determine the second target elements and constraints corresponding to each target audit indicator, wherein the constraints are used to indicate the rules satisfied by the second target element entity corresponding to the second target element; For any target audit indicator, perform the following audit operations until the audit result of the target audit indicator is not passed; When the audit result of any target audit indicator is not passed, it is determined that the pending content has not passed the audit; The audit operation includes: Select the second target element entity corresponding to the current target audit indicator from the first element entities corresponding to each target element extraction model; When the second target element entity satisfies the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be passed, the next-order target audit indicator is used as the current target audit indicator, and the second target element entity is continuously extracted; When the second target element entity does not satisfy the constraint condition corresponding to the current target audit indicator, the audit result of the current target audit indicator is determined to be failed.

7. A content review device, characterized in that: include: A first acquisition module, used to acquire content to be processed; A second acquisition module, used to acquire at least one element to be extracted corresponding to the content to be processed; A screening module, used to screen out a target element extraction model from at least one element extraction model according to the at least one element to be extracted; wherein the target element extraction model is used to extract element entities corresponding to each first element in at least one first element, wherein the set formed by the at least one first element is a subset of the set formed by the at least one element to be extracted; An input module, for inputting the to-be-processed content into each target element extraction model to obtain a first element entity corresponding to each first element; The determination module is used to determine whether the content to be processed has passed the review based on the first element entities corresponding to each target element extraction model.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.

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

Images