Text review method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a text review method and device, computer equipment, a storage medium and a program product. The method comprises the following steps: receiving an examination category and a to-be-examined text input by a user terminal; calling a corresponding prompt word according to the review category, inputting the to-be-reviewed text into the large language model, and outputting each review point and a confidence coefficient corresponding to each review point; loading a corresponding regular expression library according to the review category, performing rule matching on the to-be-reviewed text, and outputting hit review points and matching positions corresponding to the review points; performing deduplication merging on the review points to obtain target review points; and sending the target review point and the confidence coefficient and / or the matching position corresponding to the target review point to the user terminal. Therefore, the semantic comprehension ability of the large language model and the rule matching ability of the regular expression library can be combined, the examination texts of different examination categories can be comprehensively and accurately examined without depending on a large amount of standard data, the adaptation range is wide, and the flexibility is high.

Description

Technical Field

[0001] This application relates to the field of big data processing technology, and in particular to a text review method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] Text review is a process of verifying the compliance and legality of multiple elements within a text, and it is a crucial step in ensuring business compliance and preventing legal risks. For example, it includes the review of consumer rights protection (CUP) documents.

[0003] Traditional techniques typically employ machine learning methods to learn censorship-related features from labeled data. However, this approach requires a large amount of labeled data, and the model's interpretability is poor. Summary of the Invention

[0004] Therefore, it is necessary to provide a text review method, apparatus, computer equipment, computer-readable storage medium, and computer program product that does not rely on a large amount of standard data, has a wide range of applicability, and is highly flexible, in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides a text review method, the method comprising:

[0006] Receive the review category and the text to be reviewed from the user terminal;

[0007] The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into the large language model to output each review point and the confidence level corresponding to each review point.

[0008] Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching position corresponding to each review point;

[0009] The checkpoints output by the large language model and the checkpoints matched by the regular expression library are deduplicated, and the deduplicated checkpoints are merged to obtain the target checkpoint.

[0010] The target review point, the confidence level corresponding to the target review point, and / or the matching position are sent to the user terminal.

[0011] In one embodiment, before retrieving the corresponding prompt word based on the review category, the method further includes:

[0012] Predefine review categories and configure corresponding prompts for each review category; wherein, the review categories include: privacy terms, and / or disclaimers;

[0013] A regular expression library is pre-built, which includes regular expressions for various keywords involved in different review categories.

[0014] In one embodiment, before sending the target review point, the confidence level corresponding to the target review point, and / or the matching location to the user terminal, the method further includes:

[0015] The source information for each of the target review points is labeled, and the source information includes: a large language model and / or a regular expression library.

[0016] In one embodiment, after deduplicating the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and merging the deduplicated checkpoints to obtain the target checkpoint, the method further includes:

[0017] Determine the types of scenarios to which the review category applies;

[0018] Based on the scenario type, adjust the weight values ​​corresponding to the review points output by the large language model and the review points hit by the regular expression library to obtain the weight values ​​corresponding to each target review point;

[0019] The priority order of the target review points is determined according to the weight values.

[0020] In one embodiment, sending the target review point, the confidence level corresponding to the target review point, and / or the matching location to the user terminal includes:

[0021] The target review points are sorted according to their priority order.

[0022] The sorted target review points, the confidence level of the target review points, and / or the matching position are sent to the user terminal.

[0023] In one embodiment, when a new review category is added or an existing review category needs to be redefined, a prompt word is configured for the new review category, or a prompt word is reconfigured for the existing review category; and / or,

[0024] When adding a new review category or redefining an existing review category, add the regular expression corresponding to the new review category to the regular expression library, or reconfigure the keywords for the existing review category and add the regular expression corresponding to the reconfigured keywords to the regular expression library.

[0025] Secondly, this application also provides a text review device, the device comprising:

[0026] The receiving module is used to receive the review category and the text to be reviewed input by the user terminal;

[0027] The first review module is used to retrieve the corresponding prompt words according to the review category, input the text to be reviewed into the large language model, and output each review point and the confidence level corresponding to each review point;

[0028] The second review module is used to load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions corresponding to each review point.

[0029] The deduplication and merging module is used to deduplicatize the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and then merge the deduplicated checkpoints to obtain the target checkpoint.

[0030] The output module is used to send the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal.

[0031] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0032] Receive the review category and the text to be reviewed from the user terminal;

[0033] The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into the large language model to output each review point and the confidence level corresponding to each review point.

[0034] Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching position corresponding to each review point;

[0035] The checkpoints output by the large language model and the checkpoints matched by the regular expression library are deduplicated, and the deduplicated checkpoints are merged to obtain the target checkpoint.

[0036] The target review point, the confidence level corresponding to the target review point, and / or the matching position are sent to the user terminal.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0038] Receive the review category and the text to be reviewed from the user terminal;

[0039] The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into the large language model to output each review point and the confidence level corresponding to each review point.

[0040] Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching position corresponding to each review point;

[0041] The checkpoints output by the large language model and the checkpoints matched by the regular expression library are deduplicated, and the deduplicated checkpoints are merged to obtain the target checkpoint.

[0042] The target review point, the confidence level corresponding to the target review point, and / or the matching position are sent to the user terminal.

[0043] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0044] Receive the review category and the text to be reviewed from the user terminal;

[0045] The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into the large language model to output each review point and the confidence level corresponding to each review point.

[0046] Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching position corresponding to each review point;

[0047] The checkpoints output by the large language model and the checkpoints matched by the regular expression library are deduplicated, and the deduplicated checkpoints are merged to obtain the target checkpoint.

[0048] The target review point, the confidence level corresponding to the target review point, and / or the matching position are sent to the user terminal.

[0049] The aforementioned text review method, apparatus, computer equipment, computer-readable storage medium, and computer program product receive the review category and the text to be reviewed from a user terminal. Based on the review category input by the user terminal, they determine the prompt words and regular expression library to be used for the text to be reviewed. The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into a large language model, which outputs each review point and its corresponding confidence level. The corresponding regular expression library is loaded according to the review category, and rule matching is performed on the text to be reviewed, outputting the hit review points and their corresponding matching positions. This combines the semantic understanding capability of the large language model with the rule matching capability of the regular expression library to improve the accuracy of the review. The review points output by the large language model and the hit review points from the regular expression library are deduplicated, and the deduplicated review points are merged to obtain the target review point. This filters out duplicate review points while ensuring the accuracy of the review results, avoiding redundancy. The target review point, its corresponding confidence level, and / or matching position are sent to the user terminal. This allows for comprehensive and accurate review of texts across different review categories without relying on large amounts of standard data, offering broad applicability and high flexibility. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a diagram illustrating the application environment of a text review method in one embodiment;

[0052] Figure 2 This is a flowchart illustrating a text review method in one embodiment;

[0053] Figure 3 This is a flowchart illustrating a text review method in another embodiment;

[0054] Figure 4 This is a flowchart illustrating the text review method in yet another embodiment;

[0055] Figure 5 This is a flowchart of a consumer protection review method in one embodiment;

[0056] Figure 6 This is a structural block diagram of a text review device in one embodiment;

[0057] Figure 7This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] It should be noted that the terms "comprising" and "having," and any variations thereof, as used in this application, are intended to cover non-exclusive inclusion. The term "multiple" as used in this application refers to two or more. The term "and / or" as used in this application refers to one of the solutions, or any combination of multiple solutions.

[0060] The text review method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, user terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Server 104 receives the review category and the text to be reviewed from user terminal 102; it retrieves the corresponding prompt words according to the review category, inputs the text to be reviewed into a large language model, outputs each review point, and the confidence score corresponding to each review point; it loads the corresponding regular expression library according to the review category, performs rule matching on the text to be reviewed, outputs the hit review points, and the matching positions corresponding to each review point; it deduplicates the review points output by the large language model and the review points hit by the regular expression library, and merges the deduplicated review points to obtain the target review point; it sends the target review point, the confidence score corresponding to the target review point, and / or the matching position to user terminal 102. User terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and IoT devices. IoT devices can be smart TVs, projectors, etc. Server 104 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0061] In one exemplary embodiment, such as Figure 2 As shown, a text review method is provided, which can be applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 201 to 205. Wherein:

[0062] Step 201: Receive the review category and the text to be reviewed from the user terminal.

[0063] In this embodiment, the user inputs the review category and the text to be reviewed on the user terminal's display interface. The review category and the text to be reviewed have a corresponding relationship. Taking consumer rights protection review as an example, when the input text to be reviewed is related to consumer rights protection, its review category can include: privacy terms, disclaimers, etc.

[0064] It should be understood that the embodiments of this application do not limit the specific type or content of the uploaded text to be reviewed, nor do they limit the review category corresponding to each text to be reviewed. Differences in the text to be reviewed and the review category will not affect the specific implementation process of the method of this application.

[0065] Step 202: Retrieve the corresponding prompt words according to the review category, input the text to be reviewed into the large language model, and output each review point and the confidence level corresponding to each review point.

[0066] In this embodiment, each review category will have a corresponding prompt. For example, when the review category is "Privacy Terms", the corresponding prompt could be: "Please identify all terms involving user privacy from the following text".

[0067] In this embodiment, a pre-trained large language model can be used to parse the text to be reviewed and, based on prompt words, find various review points from the text. Each review point corresponds to a confidence level. The confidence level can be used to characterize the reliability of the review point, i.e., a probabilistic representation.

[0068] Optionally, when there are many review points in the output, these review points can be filtered based on their corresponding confidence levels, removing review points with confidence levels below a preset threshold. This removes less reliable review points.

[0069] Step 203: Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions corresponding to each review point.

[0070] In this embodiment, each review category corresponds to a regular expression library. Rules matching the text to be reviewed, as well as the matching positions, can be found in the regular expression library. Optionally, the regular expressions in the library are related to keywords; therefore, when matching the text to be reviewed, the corresponding regular expression can be determined by the keywords. For example, if the keyword is "personal information," then regular expressions related to personal information can be found. Similarly, if the keyword is "data collection," then regular expressions related to data collection can be found.

[0071] For example, input the prompt and the text to be reviewed into a large language model to obtain the review results (e.g., "Privacy Terms: Paragraph 3 mentions that data collection was not clearly communicated to users"). Use regular expressions to match the text and output the locations where the rules are matched (e.g., "Line 5: Matched the keyword 'personal information'").

[0072] Step 204: Deduplicate the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and merge the deduplicated checkpoints to obtain the target checkpoint.

[0073] In this embodiment, since the checkpoints output by the large language model and the regular expression library may overlap, deduplication is required before sending the target checkpoint to the user terminal to remove duplicate checkpoints. Then, the deduplicated checkpoints are merged, which includes the integration of all information. For example, checkpoint 1 is a checkpoint output by the large language model with a confidence level of 95%; checkpoint 1 is also a checkpoint matched by the regular expression library, with the matching position being "personal information". Therefore, the integrated information is: checkpoint 1, confidence level 95%, matching position "personal information".

[0074] Step 205: Send the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal.

[0075] In this embodiment, the server sends the target review points, their corresponding confidence levels, and / or matching positions to the user terminal. For example, all target review points, their corresponding confidence levels, and / or matching positions can be displayed in a list on the user terminal's display interface.

[0076] In the aforementioned text review method, the review category and the text to be reviewed are received from the user terminal. Based on the review category, the appropriate prompt words and regular expression library should be used. The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into a large language model, which outputs each review point and its corresponding confidence level. The corresponding regular expression library is loaded according to the review category, and rule matching is performed on the text to be reviewed, outputting the matched review points and their corresponding matching positions. This combines the semantic understanding capability of the large language model with the rule matching capability of the regular expression library, improving the accuracy of the review. The review points output by the large language model and the matched review points from the regular expression library are deduplicated, and the deduplicated review points are merged to obtain the target review point. This filters out duplicate review points while ensuring the accuracy of the review results, avoiding redundancy. The target review point, its corresponding confidence level, and / or matching position are sent to the user terminal. Therefore, it can comprehensively and accurately review texts of different review categories without relying on a large amount of standard data, offering wide adaptability and high flexibility.

[0077] In another exemplary embodiment, such as Figure 3 As shown, a text review method is provided, which can be applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 301 to 306. Wherein:

[0078] Step 301: Receive the review category and the text to be reviewed from the user terminal.

[0079] Step 302: Retrieve the corresponding prompt words according to the review category, input the text to be reviewed into the large language model, and output each review point and the confidence level corresponding to each review point.

[0080] Optionally, before performing step 302, it is necessary to predefine review categories and configure corresponding prompts for each review category; wherein, the review categories include: privacy terms, and / or disclaimers.

[0081] Optionally, before performing step 302, it is also necessary to build a regular expression library in advance, which includes regular expressions for various keywords involved in different review categories.

[0082] It should be understood that there is a correspondence between review categories and review texts. Therefore, once the text to be reviewed is identified, its corresponding review category is known. The prompt words and regular expression library corresponding to each review category are predetermined. Thus, after the user inputs the review category and the text to be reviewed on the terminal, the server can automatically retrieve the prompt words corresponding to the review category, facilitating the large language model to output review points based on the prompt words. The server can automatically load the corresponding regular expression library based on the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions of each review point.

[0083] Step 303: Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions corresponding to each review point.

[0084] Step 304: Deduplicate the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and merge the deduplicated checkpoints to obtain the target checkpoint.

[0085] For the specific implementation process and technical effects of steps 301 to 304 in the embodiments of this application, please refer to [link to relevant documentation]. Figure 2 The relevant descriptions of steps 201 to 204 in the method embodiment shown will not be repeated here.

[0086] Step 305: Label the source information for each target review point. The source information includes: large language model and / or regular expression library.

[0087] In this embodiment, after deduplicating the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and merging the deduplicated checkpoints, the source information of each target checkpoint is also marked. The source information includes: the large language model and / or the regular expression library.

[0088] For example, if review point 1 is both a review point output by the large language model and a review point matched by the regular expression library, then review point 1 is marked with "from the large language model and the regular expression library". Similarly, if review point 2 is a review point output by the language model, then review point 2 is marked with "from the large language model". And again, if review point 3 is a review point matched by the regular expression library, then review point 3 is marked with "from the regular expression library".

[0089] Step 306: Send the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal.

[0090] For the specific implementation process and technical effects of step 306 in the embodiments of this application, please refer to [link to relevant documentation]. Figure 2 The relevant description of step 205 in the method embodiment shown will not be repeated here.

[0091] In this embodiment, the source information of each target review point is labeled, including: a large language model and / or a regular expression library. This allows users to know the source of the review point pair, facilitating the subsequent determination of the modification method.

[0092] In yet another exemplary embodiment, such as Figure 4 As shown, a text review method is provided, which can be applied to... Figure 1 Taking the server in the example, the explanation includes steps 401 to 409. Wherein:

[0093] Step 401: Receive the review category and the text to be reviewed from the user terminal.

[0094] Step 402: Retrieve the corresponding prompt words according to the review category, input the text to be reviewed into the large language model, and output each review point and the confidence level corresponding to each review point.

[0095] Optionally, before performing step 402, it is necessary to predefine review categories and configure corresponding prompts for each review category; wherein, the review categories include: privacy terms, and / or disclaimers.

[0096] Optionally, before performing step 402, it is also necessary to build a regular expression library in advance, which includes regular expressions for each keyword involved in different review categories.

[0097] It should be understood that there is a correspondence between review categories and review texts. Therefore, once the text to be reviewed is identified, its corresponding review category is known. The prompt words and regular expression library corresponding to each review category are predetermined. Thus, after the user inputs the review category and the text to be reviewed on the terminal, the server can automatically retrieve the prompt words corresponding to the review category, facilitating the large language model to output review points based on the prompt words. The server can automatically load the corresponding regular expression library based on the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions of each review point.

[0098] Step 403: Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions corresponding to each review point.

[0099] Step 404: Deduplicate the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and merge the deduplicated checkpoints to obtain the target checkpoint.

[0100] For the specific implementation process and technical effects of steps 401 to 304 in the embodiments of this application, please refer to [link to relevant documentation]. Figure 2 The relevant descriptions of steps 201 to 204 in the method embodiment shown will not be repeated here.

[0101] Step 405: Determine the scenario type to which the review category applies.

[0102] In this embodiment, different review categories correspond to different use cases, such as high-precision scenarios and high-generalization scenarios. Scenario examples corresponding to each review category can be added to the model indicator library, and the applicable scenario type for each review category can be determined based on requirements.

[0103] Step 406: Based on the scenario type, adjust the weight values ​​corresponding to the review points output by the large language model and the review points hit by the regular expression library to obtain the weight values ​​corresponding to each target review point.

[0104] For example, in a high-precision scenario: rule weight 70%, model weight 30%. In a high-generalization scenario: model weight 70%, rule weight 30%. Optionally, assuming that review point 1 comes from a large language model and a regular expression library, review point 2 comes from a large language model, and review point 3 comes from a regular expression library; then in the high-precision scenario, the weight of review point 1 is 70% + 30%, the weight of review point 2 is 30%, and the weight of review point 3 is 70%. The calculation method in the high-generalization scenario is similar and will not be repeated here.

[0105] Step 407: Determine the priority order of the target review points according to the weight values.

[0106] In this embodiment, the weight value corresponding to each target review point can be calculated according to step 406, and then the target review points can be sorted according to the order of the weight values.

[0107] Optionally, when multiple review points have the same weight value, they can be sorted from high to low according to the confidence level corresponding to the review points.

[0108] Step 408: Sort the target review points according to their priority order.

[0109] Step 409: Send the sorted target review points, the confidence scores corresponding to the target review points, and / or the matching positions to the user terminal.

[0110] In this embodiment, the applicable scenario type for the review category is determined; based on the scenario type, the weight values ​​corresponding to the review points output by the large language model and the review points matched by the regular expression library are adjusted to obtain the weight values ​​corresponding to each target review point; the priority order of the target review points is determined according to the weight values; the target review points are sorted according to their priority order; and the sorted target review points, their corresponding confidence scores, and / or matching positions are sent to the user terminal. This allows the target review points to be sorted according to different priorities, enabling end users to intuitively determine the importance of each target review point and facilitating user review.

[0111] In one alternative implementation, after performing the above... Figure 2 Step 201 Figure 3 Step 301, and Figure 4 Before step 401, the following steps may also be included:

[0112] In cases where a new review category is added or an existing review category needs to be redefined, configure prompt words for the new review category, or reconfigure prompt words for the existing review category; and / or,

[0113] When adding a new review category or redefining an existing review category, add the regular expression corresponding to the new review category to the regular expression library, or reconfigure keywords for an existing review category and add the regular expression corresponding to the reconfigured keywords to the regular expression library.

[0114] For example, this application also provides a flowchart of a consumer protection review method, such as... Figure 5 As shown, this embodiment takes the review of text related to consumer rights protection (hereinafter referred to as consumer protection) as an example. First, the review category corresponding to the consumer protection review is processed, and the prompt words and review rules corresponding to different review categories are retrieved. Then, a large language model (hereinafter referred to as the large model) and a regular expression library (hereinafter referred to as the rules) are constructed. The corresponding prompt words are retrieved according to the review category, and the prompt words and the text to be reviewed are input into the large language model, which outputs each review point and the confidence level corresponding to each review point. The corresponding regular expression library is loaded according to the review category, and the text to be reviewed is matched according to the rules, which outputs the hit review points and the matching position of each review point. The review points output by the large language model and the review points hit by the regular expression library are deduplicated and merged to obtain the target review points. Finally, the target review points are arranged according to the weight, and the corresponding label information is output (which may include: source information, confidence level, matching position, etc. of the target review points).

[0115] In this embodiment, by dynamically configuring the review categories, the language understanding capabilities of the large model are combined with dynamic prompt words to complete text review, and combined with the precise matching of regular expressions to achieve complementary advantages.

[0116] In this embodiment, the semantic understanding capabilities of a large language model and the accuracy of rule matching can be combined to improve the accuracy of text review. By dynamically configuring review categories and prompts, it is possible to flexibly adapt to the text review needs of different scenarios. Furthermore, multimodal review (model + rules) can output more comprehensive (covering various implicit review points) and interpretable review results.

[0117] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0118] Based on the same inventive concept, this application also provides a text review apparatus for implementing the text review method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more text review apparatus embodiments provided below can be found in the limitations of the text review method described above, and will not be repeated here.

[0119] In one exemplary embodiment, such as Figure 6 As shown, a text review device is provided, including: a receiving module 601, a first review module 602, a second review module 603, a deduplication and merging module 604, and an output module 605, wherein:

[0120] The receiving module 601 is used to receive the review category and the text to be reviewed input by the user terminal;

[0121] The first review module 602 is used to retrieve the corresponding prompt words according to the review category, input the text to be reviewed into the large language model, and output each review point and the confidence level corresponding to each review point;

[0122] The second review module 603 is used to load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions corresponding to each review point.

[0123] The deduplication and merging module 604 is used to deduplicatize the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and then merge the deduplicated checkpoints to obtain the target checkpoint.

[0124] The output module 605 is used to send the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal.

[0125] For example, the above apparatus may further include: a preprocessing module 606, the preprocessing module 606 being used for:

[0126] Predefine review categories and configure corresponding prompts for each review category; the review categories include: privacy terms and / or disclaimers; pre-build a regular expression library, which includes regular expressions for each keyword involved in different review categories.

[0127] For example, the above-described apparatus may further include: a labeling module 607, which is used for:

[0128] The source information for each target review point is labeled. The source information includes: large language model and / or regular expression library.

[0129] For example, the above-described apparatus may further include: a sorting module 608, which is used for:

[0130] Determine the applicable scenario type for the review category; based on the scenario type, adjust the weight values ​​corresponding to the review points output by the large language model and the review points hit by the regular expression library to obtain the weight values ​​corresponding to each target review point; determine the priority order of the target review points according to the weight values.

[0131] For example, the output module 605 is further configured to: sort the target review points according to their priority order; and send the sorted target review points, the confidence level of the target review points and / or the matching position to the user terminal.

[0132] For example, the above-described apparatus may further include: a dynamic configuration module 609, which is configured to: configure prompt words for a newly added review category, or reconfigure prompt words for an existing review category, when a new review category is added or an existing review category needs to be redefined; and / or,

[0133] When adding a new review category or redefining an existing review category, add the regular expression corresponding to the new review category to the regular expression library, or reconfigure keywords for an existing review category and add the regular expression corresponding to the reconfigured keywords to the regular expression library.

[0134] Each module in the aforementioned text review device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can invoke and execute the corresponding operations of each module.

[0135] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a text review method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0136] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0137] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0138] The system receives the review category and the text to be reviewed from the user terminal; retrieves the corresponding prompt words according to the review category, inputs the text to be reviewed into the large language model, outputs each review point and the confidence score corresponding to each review point; loads the corresponding regular expression library according to the review category, performs rule matching on the text to be reviewed, outputs the hit review points and the matching positions corresponding to each review point; deduplicates the review points output by the large language model and the review points hit by the regular expression library, and merges the deduplicated review points to obtain the target review point; and sends the target review point, the confidence score corresponding to the target review point, and / or the matching position to the user terminal.

[0139] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0140] Predefine review categories and configure corresponding prompts for each review category; the review categories include: privacy terms and / or disclaimers; pre-build a regular expression library, which includes regular expressions for each keyword involved in different review categories.

[0141] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0142] The source information for each target review point is labeled. The source information includes: large language model and / or regular expression library.

[0143] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0144] Determine the applicable scenario type for the review category; based on the scenario type, adjust the weight values ​​corresponding to the review points output by the large language model and the review points hit by the regular expression library to obtain the weight values ​​corresponding to each target review point; determine the priority order of the target review points according to the weight values.

[0145] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0146] The target review points are sorted according to their priority order; the sorted target review points, their corresponding confidence scores, and / or matching positions are sent to the user terminal.

[0147] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0148] In cases where a new review category is added or an existing review category needs to be redefined, configure prompt words for the new review category, or reconfigure prompt words for the existing review category; and / or,

[0149] When adding a new review category or redefining an existing review category, add the regular expression corresponding to the new review category to the regular expression library, or reconfigure keywords for an existing review category and add the regular expression corresponding to the reconfigured keywords to the regular expression library.

[0150] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method steps of the various embodiments described above.

[0151] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the method steps of the various embodiments described above.

[0152] 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, data stored, data displayed, 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 the relevant data must comply with relevant regulations.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0154] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0155] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A text review method, characterized in that, The method includes: Receive the review category and the text to be reviewed from the user terminal; The corresponding prompt words are retrieved according to the review category, and the text to be reviewed is input into the large language model to output each review point and the confidence level corresponding to each review point. Load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching position corresponding to each review point; The checkpoints output by the large language model and the checkpoints matched by the regular expression library are deduplicated, and the deduplicated checkpoints are merged to obtain the target checkpoint. The target review point, the confidence level corresponding to the target review point, and / or the matching position are sent to the user terminal.

2. The method according to claim 1, characterized in that, Before retrieving the corresponding prompt words based on the review category, the method further includes: Predefine review categories and configure corresponding prompts for each review category; wherein, the review categories include: privacy terms, and / or disclaimers; A regular expression library is pre-built, which includes regular expressions for various keywords involved in different review categories.

3. The method according to claim 1, characterized in that, Before sending the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal, the method further includes: The source information for each of the target review points is labeled, and the source information includes: a large language model and / or a regular expression library.

4. The method according to any one of claims 1 to 3, characterized in that, After deduplicating the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and merging the deduplicated checkpoints to obtain the target checkpoint, the method further includes: Determine the types of scenarios to which the review category applies; Based on the scenario type, adjust the weight values ​​corresponding to the review points output by the large language model and the review points hit by the regular expression library to obtain the weight values ​​corresponding to each target review point; The priority order of the target review points is determined according to the weight values.

5. The method according to claim 4, characterized in that, Sending the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal includes: The target review points are sorted according to their priority order. The sorted target review points, the confidence level of the target review points, and / or the matching position are sent to the user terminal.

6. The method according to any one of claims 1 to 3, characterized in that, Before retrieving the corresponding prompt words based on the review category, the method further includes: In cases where a new review category is added or an existing review category needs to be redefined, a prompt word is configured for the new review category, or a prompt word is reconfigured for the existing review category; and / or, When adding a new review category or redefining an existing review category, add the regular expression corresponding to the new review category to the regular expression library, or reconfigure the keywords for the existing review category and add the regular expression corresponding to the reconfigured keywords to the regular expression library.

7. A text review device, characterized in that, The device includes: The receiving module is used to receive the review category and the text to be reviewed input by the user terminal; The first review module is used to retrieve the corresponding prompt words according to the review category, input the text to be reviewed into the large language model, and output each review point and the confidence level corresponding to each review point; The second review module is used to load the corresponding regular expression library according to the review category, perform rule matching on the text to be reviewed, and output the hit review points and the matching positions corresponding to each review point. The deduplication and merging module is used to deduplicatize the checkpoints output by the large language model and the checkpoints matched by the regular expression library, and then merge the deduplicated checkpoints to obtain the target checkpoint. The output module is used to send the target review point, the confidence level corresponding to the target review point, and / or the matching position to the user terminal.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

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

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

Images