Policy comparison method and system based on large language model

Abstract

The invention provides a policy comparison method and system based on a large language model, and the method comprises the steps: carrying out the policy paragraph splitting of an upper-level policy file and a lower-level policy file according to the title structures of the upper-level policy file and the lower-level policy file, and obtaining an upper-level policy paragraph list and a lower-level policy paragraph list corresponding to the upper-level policy file and the lower-level policy file Performing policy paragraph matching on the lower-level policy paragraph and the upper-level policy paragraph in the upper-level and lower-level policy paragraph list to obtain a Performing policy point splitting on the upper and lower policy paragraphs which are matched with each other to obtain each upper policy point and each lower policy point in the upper and lower policy paragraphs which are matched with each other; performing policy point matching on each superior policy point and each subordinate policy point in the mutually matched superior and subordinate policy paragraphs to obtain a policy point matching result; and analyzing various types of matching conditions in the two matching results by adopting corresponding large language model cue words to obtain a matching analysis result. The method aims to improve analysis and comparison efficiency and reduce cost.

Description

Technical Field

[0001] The present application relates to the field of data analysis technology, and in particular to a policy comparison method and system based on a large language model. Background Art

[0002] Policy documents are formal documents formulated and issued by government departments, and the formulation and issuance of policy documents by lower-level departments are mainly guided by policy documents issued by higher-level departments. Therefore, it is particularly important to analyze and determine whether the policy documents formulated and issued by lower-level departments are guided by policy documents issued by higher-level departments. However, policy documents often record different policy contents and thus carry a large amount of policy information. Manual analysis alone is time-consuming and laborious. Summary of the invention

[0003] In view of this, the present application provides a policy comparison method and system based on a large language model that overcomes the above problems or at least partially solves the above problems.

[0004] In a first aspect of the present application, a policy comparison method based on a large language model is provided, the method comprising:

[0005] According to the title structure of the superior and subordinate policy documents, the superior and subordinate policy documents are split into policy paragraphs to obtain a superior policy paragraph list corresponding to the superior policy document and a subordinate policy paragraph list corresponding to the subordinate policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents;

[0006] Perform policy paragraph matching on the lower-level policy paragraphs in the lower-level policy paragraph list and the upper-level policy paragraphs in the upper-level policy paragraph list to obtain corresponding paragraph matching results;

[0007] Based on the paragraph matching results, policy point splitting is performed on the mutually matched upper-level policy paragraphs and lower-level policy paragraphs respectively, so as to obtain each upper-level policy point in the mutually matched upper-level policy paragraphs and each lower-level policy point in the lower-level policy paragraphs;

[0008] Perform policy point matching on each superior policy point in the mutually matching superior policy paragraphs and each subordinate policy point in the subordinate policy paragraphs to obtain corresponding policy point matching results;

[0009] For various types of matching situations in the paragraph matching results and the policy point matching results, corresponding large language model prompt words are used to analyze the corresponding matching situations to obtain corresponding matching analysis results.

[0010] Optionally, according to the title structure of the superior and subordinate policy files, the superior and subordinate policy files are split into policy paragraphs to obtain a list of superior policy paragraphs corresponding to the superior policy file and a list of subordinate policy paragraphs corresponding to the subordinate policy file, including:

[0011] Determine the titles in the upper and lower policy files by setting rules to match the title serial numbers in the upper and lower policy files;

[0012] Based on the titles of the determined superior and subordinate policy documents, the content between titles and between a title and the end of a policy document is determined as the policy content of the previous title;

[0013] Based on the determined titles in the superior and subordinate policy documents and the policy contents corresponding to the titles, a superior policy paragraph list corresponding to the superior policy document and a subordinate policy paragraph list corresponding to the subordinate policy document are created.

[0014] Optionally, policy paragraph matching is performed on the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list to obtain corresponding paragraph matching results, including:

[0015] Constructing a first sample data set including a first number of policy paragraph matching sample data and policy paragraph non-matching sample data, wherein the policy paragraph matching sample data is sample data consisting of mutually matching upper-level policy paragraphs and lower-level policy paragraphs, and the policy paragraph non-matching sample data is lower-level policy paragraphs without mutually matching upper-level policy paragraphs;

[0016] Mixing the superior policy paragraphs in the superior policy paragraph list with the superior policy paragraphs in the first sample data set to obtain a mixed superior policy paragraph set;

[0017] Numbering the superior policy paragraphs in the mixed superior policy paragraph set to obtain a corresponding first mapping relationship library;

[0018] Converting the sample data in the first sample data set into data suitable for processing by a large language model to obtain a first target sample data set;

[0019] Inputting the mixed upper-level policy paragraph set, the first target sample data set, and the lower-level policy paragraphs in the lower-level policy paragraph list into a large language model for processing, and obtaining numbers corresponding to the lower-level policy paragraphs;

[0020] According to the serial numbers corresponding to the subordinate policy paragraphs in the subordinate policy paragraph list and the first mapping relationship library, a matching result between the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list is determined.

[0021] Optionally, determining, according to the serial numbers corresponding to the subordinate policy paragraphs in the subordinate policy paragraph list and the first mapping relationship library, the matching results between the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list include:

[0022] In the case where the subordinate policy paragraph in the subordinate policy paragraph list has no corresponding number, determining that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph;

[0023] In the case where the subordinate policy paragraph in the subordinate policy paragraph list has a corresponding number, determining a target superior policy paragraph corresponding to the number in the first mapping relationship library;

[0024] In the case where the target superior policy paragraph belongs to the superior policy paragraph in the first target sample data set, determining that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph;

[0025] In the case that the target superior policy paragraph belongs to the superior policy paragraphs in the superior policy paragraph list, it is determined that the subordinate policy paragraph matches the target superior policy paragraph.

[0026] Optionally, the superior policy paragraphs in the superior policy paragraph list are mixed with the superior policy paragraphs in the first sample data set to obtain a mixed superior policy paragraph set, including:

[0027] Determining the similarity between the superior policy paragraphs in the first sample data set and the superior policy paragraphs in the superior policy paragraph list;

[0028] Determine the superior policy paragraphs in the first sample data set whose similarity with the superior policy paragraphs in the superior policy paragraph list exceeds a first threshold as duplicate superior policy paragraphs;

[0029] Filtering repeated superior policy paragraphs in the first sample data set;

[0030] The upper-level policy paragraphs in the filtered first sample data set are mixed with the upper-level policy paragraphs in the upper-level policy paragraph list to obtain a mixed upper-level policy paragraph set.

[0031] Optionally, policy point matching is performed on each superior policy point in the mutually matching superior policy paragraphs and each subordinate policy point in the subordinate policy paragraphs to obtain corresponding policy point matching results, including:

[0032] Constructing a second sample data set including a second number of policy point matching sample data and policy point non-matching sample data, wherein the policy point matching sample data is sample data composed of mutually matching upper-level policy points and lower-level policy points, and the policy point non-matching sample data is lower-level policy points without mutually matching upper-level policy points;

[0033] Mixing the superior policy points in the mutually matching superior policy paragraphs with the superior policy points in the second sample data set to obtain a mixed superior policy point set;

[0034] Numbering the superior policy points in the mixed superior policy point set to obtain a corresponding second mapping relationship library;

[0035] Converting the sample data in the second sample data set into data suitable for processing by a large language model to obtain a second target sample data set;

[0036] Inputting the mixed upper-level policy point set, the second target sample data set, and the lower-level policy points in the mutually matched lower-level policy paragraphs into a large language model for processing to obtain numbers corresponding to the lower-level policy points;

[0037] According to the corresponding numbers of the subordinate policy points in the mutually matching subordinate policy paragraphs and the second mapping relationship library, the matching results between the subordinate policy points in the mutually matching subordinate policy paragraphs and the superior policy points in the mutually matching superior policy paragraphs are determined.

[0038] Optionally, determining the matching result between the subordinate policy points in the mutually matching subordinate policy paragraphs and the superior policy points in the mutually matching superior policy paragraphs according to the respective corresponding serial numbers of the subordinate policy points in the mutually matching subordinate policy paragraphs and the second mapping relationship library includes:

[0039] In the case where the subordinate policy point in the mutually matching subordinate policy paragraph has no corresponding number, determining that there is no superior policy point in the mutually matching superior policy paragraph that matches the subordinate policy point;

[0040] In the case where the subordinate policy points in the mutually matching subordinate policy paragraphs have corresponding numbers, determining a target superior policy point corresponding to the number in the second mapping relationship library;

[0041] In the case where the target superior policy point belongs to the superior policy point in the second target sample data set, determining that there is no superior policy point in the mutually matched superior policy paragraphs that matches the subordinate policy point;

[0042] In the case where the target superior policy point belongs to the superior policy point in the mutually matching superior policy paragraphs, it is determined that the subordinate policy point and the target superior policy point match each other.

[0043] Optionally, the method further includes:

[0044] Determine whether there are multiple subordinate policy points that match each other and the target superior policy point that match the same superior policy point;

[0045] The multiple subordinate policy points that match the same superior policy point are merged to obtain a merged single subordinate policy point that matches the same superior policy point.

[0046] Optionally, the upper-level policy points in the mutually matching upper-level policy paragraphs are mixed with the upper-level policy points in the second sample data set to obtain a mixed upper-level policy point set, including:

[0047] Determining the similarity between the superior policy point in the second sample data set and the superior policy point in the mutually matching superior policy paragraph;

[0048] Determine the superior policy points in the second sample data set whose similarity with the superior policy points in the mutually matching superior policy paragraph exceeds a second threshold as duplicate superior policy points;

[0049] filtering repeated superior policy points in the second sample data set;

[0050] The upper-level policy points in the filtered second sample data set are mixed with the upper-level policy points in the mutually matching upper-level policy paragraphs to obtain a mixed upper-level policy point set.

[0051] A second aspect of the present application provides a policy comparison system based on a large language model, the system comprising:

[0052] A policy paragraph splitting module is used to split the policy paragraphs of the upper and lower policy documents according to their title structures, and obtain a list of upper-level policy paragraphs corresponding to the upper-level policy document and a list of lower-level policy paragraphs corresponding to the lower-level policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents;

[0053] A policy paragraph matching module, used to perform policy paragraph matching on the lower-level policy paragraphs in the lower-level policy paragraph list and the upper-level policy paragraphs in the upper-level policy paragraph list to obtain corresponding paragraph matching results;

[0054] A policy point splitting module is used to split the policy points of the mutually matched upper policy paragraphs and lower policy paragraphs respectively based on the paragraph matching results, so as to obtain each upper policy point in the mutually matched upper policy paragraphs and each lower policy point in the lower policy paragraphs;

[0055] A policy point matching module is used to match each upper policy point in the upper policy paragraphs and each lower policy point in the lower policy paragraphs that match each other, and obtain corresponding policy point matching results;

[0056] The matching analysis module is used to analyze the corresponding matching situations of the paragraph matching results and the policy point matching results using the corresponding large language model prompt words to obtain the corresponding matching analysis results.

[0057] Compared with the prior art, this application has the following advantages:

[0058] The present application provides a policy comparison method based on a large language model. First, according to the title structure of the upper and lower policy documents, the upper and lower policy documents are split into policy paragraphs to obtain a list of upper policy paragraphs corresponding to the upper policy document and a list of lower policy paragraphs corresponding to the lower policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents; the lower policy paragraphs in the lower policy paragraph list are matched with the upper policy paragraphs in the upper policy paragraph list to obtain corresponding paragraph matching results; based on the paragraph matching results, the upper policy paragraphs and lower policy paragraphs that match each other are split into policy points to obtain each upper policy point in the upper policy paragraph that matches each other and each lower policy point in the lower policy paragraph; the upper policy points in the upper policy paragraph that matches each other and each lower policy point in the lower policy paragraph are matched to obtain corresponding policy point matching results; for various types of matching situations in the paragraph matching results and the policy point matching results, the corresponding matching situations are analyzed using corresponding large language model prompt words to obtain corresponding matching analysis results. Therefore, the present application first divides and matches the policy paragraphs of the superior and subordinate policy documents in a segmented manner through a large language model, and then further divides and matches the policy points of the matched superior and subordinate policy paragraphs, and finally analyzes the matching results based on the obtained policy paragraph matching results and policy point matching results, so as to obtain the policy consistency and adaptability between the formulated subordinate policy documents and the superior policy documents. For example, the subordinate policy document is well formulated under the guidance of the superior policy document, indicating that the policy consistency between the subordinate policy document and the superior policy document is relatively good; for example, the subordinate policy document is formulated based on the guidance of the superior policy document, and at the same time, the subordinate policy document is flexibly adjusted according to the actual situation, so that the subordinate policy document can better adapt to the development needs of the implementation area and department of the subordinate policy document.

[0059] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0061] Figure 1 A flowchart of a policy comparison method based on a large language model provided in an embodiment of the present application;

[0062] Figure 2A schematic diagram of a policy comparison system based on a large language model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings.

[0064] Figure 1 A flowchart of a policy comparison method based on a large language model provided in an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0065] Step S1: According to the title structure of the superior and subordinate policy documents, the superior and subordinate policy documents are split into policy paragraphs to obtain a superior policy paragraph list corresponding to the superior policy document and a subordinate policy paragraph list corresponding to the subordinate policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents.

[0066] In this embodiment, with respect to a superior policy document and a subordinate policy document obtained for policy comparison, the text content in the superior policy document and the text content in the subordinate policy document are identified through document parsing methods such as PDF parsing and Word parsing.

[0067] In the present embodiment, the text content of the superior policy document is divided into paragraphs based on the title in the text content of the superior policy document, and each paragraph obtained by the division has a corresponding title. The paragraph content in a paragraph and the title corresponding to the paragraph constitute a corresponding superior policy paragraph, the paragraph content in the superior policy paragraph is the policy content of the superior policy paragraph, and the title in the superior policy paragraph is the policy title of the superior policy paragraph. Based on the same implementation method, a superior policy paragraph corresponding to each paragraph obtained by the division will be obtained. All superior policy paragraphs obtained based on the text content of the superior policy document are recorded to obtain a superior policy paragraph list, and the elements constituting the superior policy paragraph list are superior policy paragraphs including corresponding policy titles and policy contents.

[0068] For example, the title in the text content of the superior policy document is used as the basis for division, and the text content of the superior policy document is divided into paragraphs to obtain paragraphs A1, paragraph A2, and paragraph A3. The title corresponding to paragraph A1 is a1, the title corresponding to paragraph A2 is a2, and the title corresponding to paragraph A3 is a3. Paragraph A1 and its corresponding title a1 form a superior policy paragraph X1, paragraph A2 and its corresponding title a2 form a superior policy paragraph X2, and paragraph A3 and its corresponding title a3 form a superior policy paragraph X3. All superior policy paragraphs X1, X2, and X3 obtained based on the text content of the superior policy document are recorded to obtain a superior policy paragraph list consisting of all superior policy paragraphs X1, X2, and X3.

[0069] In this embodiment, the text content of the subordinate policy document is divided into paragraphs based on the title in the text content of the subordinate policy document, and each paragraph obtained by the division has a corresponding title. The paragraph content in a paragraph and the title corresponding to the paragraph constitute a corresponding subordinate policy paragraph. The paragraph content in the subordinate policy paragraph is the policy content of the subordinate policy paragraph, and the title in the subordinate policy paragraph is the policy title of the subordinate policy paragraph. Based on the same implementation method, a corresponding subordinate policy paragraph will be obtained for each paragraph obtained by the division. All subordinate policy paragraphs obtained based on the text content of the subordinate policy document are recorded to obtain a subordinate policy paragraph list, and the elements constituting the subordinate policy paragraph list are subordinate policy paragraphs including corresponding policy titles and policy contents.

[0070] Step S2: performing policy paragraph matching on the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list to obtain corresponding paragraph matching results.

[0071] In this embodiment, after obtaining a list of superior policy paragraphs corresponding to a superior policy file and a list of subordinate policy paragraphs corresponding to a subordinate policy file for policy comparison in step S1, each subordinate policy paragraph in the subordinate policy paragraph list is matched with all superior policy paragraphs in the superior policy paragraph list, and it is determined which superior policy paragraph in the superior policy paragraph list each subordinate policy paragraph in the subordinate policy paragraph list matches with, thereby obtaining corresponding paragraph matching results. When a subordinate policy paragraph matches an superior policy paragraph, it indicates that the subordinate policy paragraph belongs to the derived policy paragraph of the superior policy paragraph, that is, the subordinate policy paragraph is formulated based on the guidance of the superior policy paragraph. In the obtained paragraph matching results, possible matching situations include: some subordinate policy paragraphs in the subordinate policy paragraph list have no matching superior policy paragraphs in the superior policy paragraph list; some subordinate policy paragraphs in the subordinate policy paragraph list have matching superior policy paragraphs in the superior policy paragraph list; and also involve a situation in which a subordinate policy paragraph not in the subordinate policy paragraph list matches a superior policy paragraph in the superior policy paragraph list (which may be one superior policy paragraph in the superior policy paragraph list, or may be multiple superior policy paragraphs).

[0072] For example, the list of superior policy paragraphs for comparison includes superior policy paragraphs X1, X2, and X3, and the list of subordinate policy paragraphs for comparison includes subordinate policy paragraphs Y1, Y2, and Y3. Through policy paragraph matching, the corresponding matching results are that subordinate policy paragraph Y1 matches superior policy paragraph X2 (the matching situation is that a subordinate policy paragraph in the subordinate policy paragraph list has a matching superior policy paragraph in the superior policy paragraph list), subordinate policy paragraph Y2 matches superior policy paragraph X3, subordinate policy paragraph Y3 has no matching superior policy paragraph (the matching situation is that a subordinate policy paragraph in the subordinate policy paragraph list has no matching superior policy paragraph in the superior policy paragraph list), and superior policy paragraph X1 has no matching subordinate policy paragraph (the matching situation is that no subordinate policy paragraph in the subordinate policy paragraph list matches a superior policy paragraph in the superior policy paragraph list).

[0073] Step S3: Based on the paragraph matching results, policy point splitting is performed on the mutually matching upper-level policy paragraphs and lower-level policy paragraphs respectively to obtain each upper-level policy point in the mutually matching upper-level policy paragraphs and each lower-level policy point in the lower-level policy paragraphs.

[0074] In this embodiment, after the corresponding paragraph matching result is obtained in step S2, a further matching analysis is performed on the upper-level policy paragraph and the lower-level policy paragraph that form a matching relationship with each other in the matching result.

[0075] Specifically: a policy paragraph generally involves multiple policy points, which refer to specific and clear policy requirements or measures, etc. The policy points of the upper-level policy paragraph and the lower-level policy paragraph that form a matching relationship are split respectively, so as to obtain multiple upper-level policy points (it should be understood that there may be only one upper-level policy point) and multiple lower-level policy points (it should be understood that there may be only one lower-level policy point) in the upper-level policy paragraph and the lower-level policy paragraph that form a matching relationship. Among them, the upper-level policy point is a small section of policy content obtained by splitting the policy points of the upper-level policy paragraph that forms a matching relationship, and the lower-level policy point is a small section of policy content obtained by splitting the policy points of the lower-level policy paragraph that forms a matching relationship.

[0076] Step S4: policy point matching is performed on each superior policy point in the superior policy paragraphs and each subordinate policy point in the subordinate policy paragraphs that match each other to obtain corresponding policy point matching results.

[0077] In this embodiment, after obtaining the multiple upper policy points and multiple lower policy points in the upper policy paragraph and the lower policy paragraph that form a matching relationship with each other through step S3, the multiple upper policy points and the multiple lower policy points are further matched with each other to determine which upper policy point in the multiple upper policy points each lower policy point matches with, thereby obtaining the corresponding policy point matching results. When a lower policy point matches an upper policy point, it indicates that the lower policy point belongs to the derived policy point of the upper policy point, that is, the lower policy point is formulated based on the guidance of the upper policy point.

[0078] In the obtained policy point matching results, possible matching situations include: some subordinate policy points in the subordinate policy paragraphs that form a matching relationship have no matching superior policy points in the superior policy paragraphs that form a matching relationship; some subordinate policy points in the subordinate policy paragraphs that form a matching relationship have matching superior policy points in the superior policy paragraphs that form a matching relationship; and also involve a situation in which no subordinate policy point in the subordinate policy paragraphs that form a matching relationship matches a superior policy point (which may be one superior policy point in the superior policy paragraph, or may be multiple superior policy points) in the superior policy paragraphs that form a matching relationship.

[0079] For example, in the upper policy paragraph X2 and the lower policy paragraph Y1 that form a matching relationship with each other, the upper policy paragraph X2 includes multiple upper policy points x21, x22, and x23, and the lower policy paragraph Y1 includes multiple lower policy points y11, y12, and y13. Through policy point matching, the corresponding matching results are obtained as follows: the subordinate policy point y11 matches the superior policy point x22 (this matching situation is that some subordinate policy points in the subordinate policy paragraphs that form a matching relationship have matching superior policy points in the superior policy paragraphs that form a matching relationship), the subordinate policy point y12 matches the superior policy point c23, the subordinate policy point y13 has no matching superior policy paragraph (this matching situation is that some subordinate policy points in the subordinate policy paragraphs that form a matching relationship have no matching superior policy points in the superior policy paragraph), and the superior policy point x11 has no matching subordinate policy point (this matching situation is that no subordinate policy point in the subordinate policy paragraphs that form a matching relationship matches with some superior policy points in the superior policy paragraphs that form a matching relationship).

[0080] Step S5: For various types of matching situations in the paragraph matching results and the policy point matching results, corresponding large language model prompt words are used to analyze the corresponding matching situations to obtain corresponding matching analysis results.

[0081] In this embodiment, after the paragraph matching results and policy point matching results are obtained through the segmented matching of the above steps S1 to S4, the final matching situation includes at least one of the following five types, namely: 1. In the superior policy paragraph list and the subordinate policy paragraph list corresponding to the superior and subordinate policy documents for policy comparison, there are subordinate policy paragraphs in the subordinate policy paragraph list that do not have matching superior policy paragraphs in the superior policy paragraph list; 2. There is no subordinate policy paragraph in the subordinate policy paragraph list that matches with the superior policy paragraph in the superior policy paragraph list. ; 3. For the upper-level policy paragraph and the lower-level policy paragraph that form a matching relationship, some lower-level policy points in the lower-level policy paragraph that forms a matching relationship do not have matching upper-level policy points in the upper-level policy paragraph that forms a matching relationship; 4. Some lower-level policy points in the lower-level policy paragraph that forms a matching relationship have matching upper-level policy points in the upper-level policy paragraph that forms a matching relationship; 5. No lower-level policy point in the lower-level policy paragraph that forms a matching relationship matches some upper-level policy point in the upper-level policy paragraph that forms a matching relationship. Among them, the above-mentioned third to fifth matching situations are three matching situations derived from the lower-level policy paragraphs in the lower-level policy paragraph list and the upper-level policy paragraphs in the upper-level policy paragraph list that match them. For each type of matching situation, this application presets the corresponding large language model prompt words in advance, and then for each type of matching situation in the final matching situation, the corresponding large language model prompt words are used to analyze the matching situation through the large language model to obtain the corresponding matching analysis results.

[0082] Specifically, in the matching situation of the superior policy paragraph list and the subordinate policy paragraph list corresponding to the superior and subordinate policy documents for policy comparison, when there is a subordinate policy paragraph in the subordinate policy paragraph list that has no superior policy paragraph matching it in the superior policy paragraph list, the reason for the matching situation is that the subordinate policy paragraph is an extended policy paragraph, and the reason for the mismatch is "policy extension". At this time, the large language model prompt word corresponding to the matching situation is used to summarize the subordinate policy paragraph that has no matching superior policy paragraph to obtain the corresponding matching analysis result. The large language model prompt word corresponding to the matching situation is preferably:

[0083]

[0084]

[0085] For example, in the superior policy paragraph list and the subordinate policy paragraph list corresponding to the superior and subordinate policy documents for policy comparison, when a subordinate policy paragraph Y3 in the subordinate policy paragraph list has no superior policy paragraph matching it in the superior policy paragraph list, the large language model prompt words corresponding to the matching situation are used to summarize the subordinate policy paragraph Y3 to obtain the corresponding matching analysis result.

[0086] When there is no subordinate policy paragraph in the subordinate policy paragraph list and the superior policy paragraph list corresponding to the superior and inferior policy documents for policy comparison, and there is a superior policy paragraph in the superior policy paragraph list that matches the superior policy paragraph in the superior policy paragraph list, there are three preset reasons for the matching situation, namely: the functions of the superior and inferior governments are different; the superior policy document specifies a certain city for execution, but does not involve the city where the subordinate policy document is implemented; the subordinate policy document does lack a subordinate policy paragraph that matches the superior policy paragraph in the superior policy paragraph list (that is, the subordinate policy paragraph corresponding to the superior policy paragraph originally needed to be recorded in the subordinate policy document, but it is not actually in the subordinate policy document). At this time, the large language model prompt word corresponding to the matching situation is used to summarize the superior policy paragraphs that do not have a matching subordinate policy paragraph in the superior policy paragraph list to obtain the corresponding matching analysis result. The large language model prompt word corresponding to the matching situation is preferably:

[0087]

[0088] For example, in the superior policy paragraph list and the subordinate policy paragraph list corresponding to the superior and subordinate policy documents for policy comparison, when a superior policy paragraph X1 in the superior policy paragraph list has no matching subordinate policy paragraph in the subordinate policy paragraph list, the large language model prompt words corresponding to the matching situation are used to summarize the superior policy paragraph X1 to obtain the corresponding matching analysis result.

[0089] In the matching situation of the upper policy paragraph and the lower policy paragraph that form a matching relationship, if there is a lower policy point in the lower policy paragraph that forms a matching relationship, and there is no upper policy point that matches it in the upper policy paragraph that forms a matching relationship, the reason for the matching situation is that the lower policy point is an extended policy point, and the reason for the mismatch is "policy extension". At this time, the large language model prompt word corresponding to the matching situation is used to summarize the lower policy point that does not match the upper policy point, and obtain the corresponding matching analysis result. The large language model prompt word corresponding to the matching situation is preferably:

[0090]

[0091] For example, in an upper-level policy paragraph and a lower-level policy paragraph that form a matching relationship with each other, when the lower-level policy point y13 in the lower-level policy paragraph that forms a matching relationship has no matching upper-level policy point in the upper-level policy paragraph that forms a matching relationship, the large language model prompt word corresponding to the matching situation is used to summarize the lower-level policy point y13 to obtain the corresponding matching analysis result.

[0092] In the matching situation of a superior policy paragraph and a subordinate policy paragraph that form a matching relationship with each other, when there is no subordinate policy point in the subordinate policy paragraph that forms a matching relationship that matches the superior policy point in the superior policy paragraph that forms a matching relationship, there are three preset reasons for the occurrence of this matching situation, namely: the functions of the superior and subordinate governments are different; the superior policy document specifies a certain city for execution, but does not involve the city where the subordinate policy document is implemented; the subordinate policy paragraph does lack subordinate policy points that match the superior policy points in the superior policy paragraph (that is, the subordinate policy points that originally needed to be recorded in the subordinate policy paragraph to match the superior policy points in the superior policy paragraph, but there are actually none in the subordinate policy paragraph). At this time, the large language model prompt words corresponding to the matching situation are used to summarize the superior policy points in the superior policy paragraph that do not have matching subordinate policy points to obtain the corresponding matching analysis results. The large language model prompt words corresponding to the matching situation are preferably:

[0093]

[0094] For example, in a superior policy paragraph and a subordinate policy paragraph that form a matching relationship with each other, if there is a superior policy point x11 in the superior policy paragraph and there is no subordinate policy point matching it in the subordinate policy paragraph, the large language model prompt word corresponding to the matching situation is used to summarize the superior policy point x11 to obtain the corresponding matching analysis result.

[0095] In the matching situation of a superior policy paragraph and a subordinate policy paragraph that form a matching relationship with each other, when there are subordinate policy points in the subordinate policy paragraph that form a matching relationship with each other, and there are superior policy points that match with them in the superior policy paragraph that forms a matching relationship with each other, although the subordinate policy points are formulated based on the guidance of the superior policy points that match with each other, whether there is good consistency and / or adaptability between the two needs further analysis. At this time, the corresponding large language model prompt words of the matching situation are used to summarize the subordinate policy points and superior policy points that match with each other, and obtain the corresponding matching analysis results. The corresponding large language model prompt words of the matching situation are as follows:

[0096]

[0097] The present application provides a policy comparison method based on a large language model. First, according to the title structure of the upper and lower policy documents, the upper and lower policy documents are split into policy paragraphs to obtain a list of upper policy paragraphs corresponding to the upper policy document and a list of lower policy paragraphs corresponding to the lower policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents; the lower policy paragraphs in the lower policy paragraph list are matched with the upper policy paragraphs in the upper policy paragraph list to obtain corresponding paragraph matching results; based on the paragraph matching results, the upper policy paragraphs and lower policy paragraphs that match each other are split into policy points to obtain each upper policy point in the upper policy paragraph that matches each other and each lower policy point in the lower policy paragraph; the upper policy points in the upper policy paragraph that matches each other and each lower policy point in the lower policy paragraph are matched to obtain corresponding policy point matching results; for various types of matching situations in the paragraph matching results and the policy point matching results, the corresponding matching situations are analyzed using corresponding large language model prompt words to obtain corresponding matching analysis results. Therefore, the present application first divides and matches the policy paragraphs of the superior and subordinate policy documents in a segmented manner through a large language model, and then further divides and matches the policy points of the matched superior and subordinate policy paragraphs, and finally analyzes the matching results based on the obtained policy paragraph matching results and policy point matching results, so as to obtain the policy consistency and adaptability between the formulated subordinate policy documents and the superior policy documents. For example, the subordinate policy document is well formulated under the guidance of the superior policy document, indicating that the policy consistency between the subordinate policy document and the superior policy document is relatively good; for example, the subordinate policy document is formulated based on the guidance of the superior policy document, and at the same time, the subordinate policy document is flexibly adjusted according to the actual situation, so that the subordinate policy document can better adapt to the development needs of the implementation area and department of the subordinate policy document.

[0098] In combination with the above embodiments, in one implementation, the embodiment of the present application further provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S1 may include steps S11 to S13:

[0099] Step S11: determining the titles in the upper and lower policy files by matching the title serial numbers in the upper and lower policy files through setting rules.

[0100] In this embodiment, the policy is expressed in the form of paragraphs in the policy document, and each policy paragraph has its corresponding title and content. Therefore, this application matches the title sequence numbers in the text content of the upper and lower policy documents through the set rules to obtain each title in the text content.

[0101] Step S12: Based on the determined titles in the superior and subordinate policy documents, the content between titles and between a title and the end of the policy document is determined as the policy content of the previous title.

[0102] In this embodiment, after obtaining the titles in the text content of the superior and subordinate policy documents through step S11, the content between every two titles in the text content of the superior policy document is determined as the policy content of the previous title, and the content between the last title and the end of the policy document is determined as the policy content of the last title. Similarly, the content between every two titles in the text content of the subordinate policy document is determined as the policy content of the previous title, and the content between the last title and the end of the policy document is determined as the policy content of the last title.

[0103] Step S13: creating a list of superior policy paragraphs corresponding to the superior policy file and a list of subordinate policy paragraphs corresponding to the subordinate policy file according to the determined titles in the superior and subordinate policy files and the policy contents corresponding to the titles.

[0104] In this embodiment, after obtaining the various titles in the superior and subordinate policy files and the policy contents corresponding to each title through the above steps S11 to S12, a corresponding superior policy paragraph is formed by the title in the superior policy file and the corresponding policy content, and a superior policy paragraph list is formed based on all the superior policy paragraphs composed of the superior policy file; at the same time, a corresponding subordinate policy paragraph is formed by the title in the subordinate policy file and the corresponding policy content, and a subordinate policy paragraph list is formed based on all the subordinate policy paragraphs composed of the subordinate policy files.

[0105] In combination with the above embodiments, in one implementation, the embodiment of the present application further provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S2 may include steps S21 to S26:

[0106] Step S21: construct a first sample data set including a first number of policy paragraph matching sample data and policy paragraph non-matching sample data, wherein the policy paragraph matching sample data is sample data composed of mutually matching upper-level policy paragraphs and lower-level policy paragraphs, and the policy paragraph non-matching sample data is lower-level policy paragraphs without mutually matching upper-level policy paragraphs.

[0107] In this embodiment, the present application first constructs a first sample data set including a small amount of sample data, the first sample data set includes some policy paragraph matching sample data and some policy paragraph non-matching sample data, the number of sample data included in the first sample data set is a first number, the first number can be set according to the actual application scenario, and is not specifically limited here. Among them, the policy paragraph matching sample data refers to the sample data composed of the matching upper-level policy paragraph and the lower-level policy paragraph, and the policy paragraph mentioned here is also composed of policy content and policy title; the policy paragraph non-matching sample data refers to the lower-level policy paragraph without the matching upper-level policy paragraph, that is, the policy paragraph non-matching sample data only has the lower-level policy paragraph, and the policy paragraph mentioned here is also composed of policy content and policy title.

[0108] Step S22: Mix the superior policy paragraphs in the superior policy paragraph list with the superior policy paragraphs in the first sample data set to obtain a mixed superior policy paragraph set.

[0109] In this embodiment, the superior policy paragraphs in the first sample data set obtained in step S21 are shuffled and mixed with all superior policy paragraphs in the superior policy paragraph list corresponding to the superior policy file for policy comparison to obtain a mixed superior policy paragraph set.

[0110] Step S23: numbering the superior policy paragraphs in the mixed superior policy paragraph set to obtain a corresponding first mapping relationship library.

[0111] In this embodiment, each superior policy paragraph in the mixed superior policy paragraph set obtained in step S22 is numbered, so as to construct a corresponding first mapping relationship library. Through the first mapping relationship library, the superior policy paragraph corresponding to the number can be queried based on the number.

[0112] Step S24: converting the sample data in the first sample data set into data suitable for processing by a large language model to obtain a first target sample data set.

[0113] In this embodiment, each sample data in the first sample data set obtained in step S21 is converted into data suitable for processing by a large language model, and all the converted sample data constitute a first target sample data set.

[0114] Specifically: the subordinate policy paragraph of each sample data in the first sample data set is taken as input data, and the number of the superior policy paragraph matching the subordinate policy paragraph is taken as output data, so as to construct corresponding data suitable for processing by the large language model; for sample data in the first sample data set that only has subordinate policy paragraphs but no matching superior policy paragraphs, the subordinate policy paragraph is taken as input data, and the corresponding output data is none.

[0115] Step S25: input the mixed upper-level policy paragraph set, the first target sample data set, and the subordinate policy paragraphs in the subordinate policy paragraph list into a large language model for processing to obtain numbers corresponding to the subordinate policy paragraphs.

[0116] In this embodiment, the mixed superior policy paragraph set, the first target sample data set and a subordinate policy paragraph in the subordinate policy paragraph list obtained by steps S21 to S24 are input into the large language model for processing, and the large language model will output a number that matches the subordinate policy paragraph. It should be understood that some subordinate policy paragraphs in the subordinate policy paragraph list may not have a matching superior policy paragraph in the superior policy paragraph list, and at this time the large language model will output empty, that is, there is no number that matches the subordinate policy paragraph. Among them, for the subordinate policy paragraphs in the subordinate policy paragraph list, only one of the subordinate policy paragraphs is input into the large language model for processing each time, and the mixed superior policy paragraph set and the first target sample data set are also input each time.

[0117] For example, the mixed upper-level policy paragraph set is P, the first target sample data set is Q, and the lower-level policy paragraph list includes lower-level policy paragraphs Y1, Y2, and Y3. The mixed upper-level policy paragraph set P, the first target sample data set Q, and the lower-level policy paragraph Y1 are input into the large language model for processing, and the large language model will output the number that matches the lower-level policy paragraph Y1. Then the mixed upper-level policy paragraph set P, the first target sample data set Q, and the lower-level policy paragraph Y2 are input into the large language model for processing, and the large language model will output the number that matches the lower-level policy paragraph Y2. Then the mixed upper-level policy paragraph set P, the first target sample data set Q, and the lower-level policy paragraph Y3 are input into the large language model for processing, and the large language model will output the number that matches the lower-level policy paragraph Y3.

[0118] Step S26: determining matching results between subordinate policy paragraphs in the subordinate policy paragraph list and superior policy paragraphs in the superior policy paragraph list according to serial numbers corresponding to the subordinate policy paragraphs in the subordinate policy paragraph list and the first mapping relationship library.

[0119] In this embodiment, the matching situation between each subordinate policy paragraph in the subordinate policy paragraph list and the superior policy paragraph in the superior policy paragraph list is determined in the same manner, and a subordinate policy paragraph is used as an example for explanation. After obtaining the number corresponding to the subordinate policy paragraph in the subordinate policy paragraph list through steps S21 to S25, the superior policy paragraph corresponding to the number in the first mapping relationship library constructed is queried, so as to determine which superior policy paragraph in the subordinate policy paragraph list matches each other. It should be understood that for the subordinate policy paragraph without a matching number, there is no superior policy paragraph that matches it accordingly, and it can be directly obtained that the matching situation of the subordinate policy paragraph is the superior policy paragraph that does not match it in the superior policy paragraph list. The application introduces the number corresponding to the superior policy paragraph and the first mapping relationship library here, which can avoid the inefficiency problem caused by the large language model outputting the superior policy paragraph that matches the current subordinate policy paragraph after executing the policy paragraph matching. When the large language model is output at the same time, it may cause the output superior policy paragraph to deviate from the superior policy paragraph that was originally used for policy paragraph matching. By introducing the number corresponding to the superior policy paragraph and the first mapping relationship library, the large language model can output only one number after executing the policy paragraph matching, which will be more efficient. At the same time, based on this number, the most original superior policy paragraph that matches the current subordinate policy paragraph can be obtained by querying the first mapping relationship library, which will not cause the superior policy paragraph output by the large language model that matches the current subordinate policy paragraph to deviate from the most original superior policy paragraph.

[0120] In combination with the above embodiments, in one implementation, the embodiment of the present application further provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S26 may include: when there is no corresponding number in the subordinate policy paragraph in the subordinate policy paragraph list, determining that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph; when the subordinate policy paragraph in the subordinate policy paragraph list has a corresponding number, determining the target superior policy paragraph corresponding to the number in the first mapping relationship library; when the target superior policy paragraph belongs to the superior policy paragraph in the first target sample data set, determining that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph; when the target superior policy paragraph belongs to the superior policy paragraph in the superior policy paragraph list, determining that the subordinate policy paragraph matches the target superior policy paragraph.

[0121] In this embodiment, for a subordinate policy paragraph with no matching number in the subordinate policy paragraph list, the corresponding matching situation is determined as no superior policy paragraph matching the subordinate policy paragraph in the superior policy paragraph list.

[0122] In this embodiment, for a subordinate policy paragraph with a matching number in the subordinate policy paragraph list, a superior policy paragraph corresponding to the number in the first mapping relationship library is determined, and the superior policy paragraph is referred to as a target superior policy paragraph. Then, it is determined whether the target superior policy paragraph belongs to a superior policy paragraph in the first target sample data set or to a superior policy paragraph in the superior policy paragraph list. In the case where the target superior policy paragraph belongs to a superior policy paragraph in the first target sample data set, the corresponding matching situation is also determined to be that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph. In the case where the target superior policy paragraph belongs to a superior policy paragraph in the superior policy paragraph list, the corresponding matching situation is determined to be that the target superior policy paragraph matches the subordinate policy paragraph.

[0123] In combination with the above embodiments, in one implementation, the embodiment of the present application further provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S22 may include: determining the similarity between the superior policy paragraphs in the first sample data set and the superior policy paragraphs in the superior policy paragraph list; determining the superior policy paragraphs in the first sample data set whose similarity with the superior policy paragraphs in the superior policy paragraph list exceeds a first threshold as duplicate superior policy paragraphs; filtering the duplicate superior policy paragraphs in the first sample data set; mixing the superior policy paragraphs in the filtered first sample data set with the superior policy paragraphs in the superior policy paragraph list to obtain a mixed superior policy paragraph set.

[0124] In this embodiment, in order to prevent the superior policy paragraphs in the first sample data set that are particularly similar to the superior policy paragraphs in the superior policy paragraph list from affecting the subsequent policy paragraph matching of the large language model, before mixing the superior policy paragraphs in the first sample data set with the superior policy paragraphs in the superior policy paragraph list, the present application first filters the superior policy paragraphs in the first sample data set that are particularly similar to the superior policy paragraphs in the superior policy paragraph list and then shuffles and mixes them.

[0125] Specifically: determine the similarity between each superior policy paragraph in the first sample data set and each superior policy paragraph in the superior policy paragraph list, and then determine whether the similarity corresponding to each superior policy paragraph in the first sample data set exceeds the first threshold, and if so, determine the superior policy paragraph in the first sample data set as a duplicate superior policy paragraph. Then filter out all superior policy paragraphs in the first sample data set that are determined to be duplicate superior policy paragraphs, and then shuffle and mix the superior policy paragraphs remaining in the first sample data set after filtering with the superior policy paragraphs in the superior policy paragraph list to obtain a mixed superior policy paragraph set. Among them, the first threshold can be set according to the actual application scenario, and is not specifically limited here. For example, the first sample data set includes superior policy paragraphs Z1, Z2, and Z3, and the superior policy paragraphs in the superior policy paragraph list include X1, X2, and X3. The similarities between superior policy paragraph Z1 and superior policy paragraphs X1, X2, and X3 are determined. If one of the similarities exceeds a first threshold, superior policy paragraph Z1 is filtered out; the similarities between superior policy paragraph Z2 and superior policy paragraphs X1, X2, and X3 are determined. If one of the similarities exceeds the first threshold, superior policy paragraph Z2 is filtered out; the similarities between superior policy paragraph Z3 and superior policy paragraphs X1, X2, and X3 are determined. If one of the similarities exceeds the first threshold, superior policy paragraph Z3 is filtered out.

[0126] In combination with the above embodiments, in one implementation, the embodiment of the present application further provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S4 may include steps S41 to S46:

[0127] Step S41: Construct a second sample data set including a second number of policy point matching sample data and policy point unmatched sample data, wherein the policy point matching sample data are sample data composed of mutually matching upper-level policy points and lower-level policy points, and the policy point unmatched sample data are lower-level policy points without mutually matching upper-level policy points.

[0128] In this embodiment, the present application first constructs a second sample data set including a small amount of sample data, the second sample data set includes some policy point matching sample data and some policy point non-matching sample data, the number of sample data included in the second sample data set is the second number, the second number can be set according to the actual application scenario, and is not specifically limited here. Among them, the policy point matching sample data refers to the sample data composed of the upper policy point and the lower policy point that match each other; the policy point non-matching sample data refers to the lower policy point that does not have a matching upper policy point, that is, the policy point non-matching sample data only has the lower policy point.

[0129] Step S42: Mix the superior policy points in the mutually matching superior policy paragraphs with the superior policy points in the second sample data set to obtain a mixed superior policy point set.

[0130] In this embodiment, the superior policy points in the second sample data set obtained in step S41 are mixed with all superior policy points in the mutually matching superior policy sections to obtain a mixed superior policy point set.

[0131] Step S43: numbering the superior policy points in the mixed superior policy point set to obtain a corresponding second mapping relationship library.

[0132] In this embodiment, each superior policy point in the mixed superior policy point set obtained in step S42 is numbered, so as to construct a corresponding second mapping relationship library. Through the second mapping relationship library, the superior policy point corresponding to the number can be queried based on the number.

[0133] Step S44: converting the sample data in the second sample data set into data suitable for processing by a large language model to obtain a second target sample data set.

[0134] In this embodiment, each sample data in the second sample data set obtained in step S41 is converted into data suitable for processing by a large language model, and all the converted sample data constitute a second target sample data set.

[0135] Specifically: the subordinate policy point of each sample data in the second sample data set is taken as input data, and the number of the superior policy point matching the subordinate policy point is taken as output data, so as to construct corresponding data suitable for processing by the large language model; for sample data in the first sample data set that only has subordinate policy points but no matching superior policy points, the subordinate policy point is taken as input data, and the corresponding output data is none.

[0136] Step S45: inputting the mixed upper-level policy point set, the second target sample data set and the subordinate policy points in the mutually matching subordinate policy paragraphs into a large language model for processing to obtain numbers corresponding to the subordinate policy points.

[0137] In this embodiment, the mixed upper-level policy point set, the second target sample data set and a lower-level policy point of the matching lower-level policy paragraphs obtained through steps S41 to S44 are input into the large language model for processing, and the large language model will output a number that matches the lower-level policy point. It should be understood that some lower-level policy points in the matching lower-level policy paragraphs may not have a matching upper-level policy point in the matching upper-level policy paragraphs. In this case, the large language model will output empty, that is, there is no number that matches the lower-level policy point. Among them, for the lower-level policy points in the lower-level policy paragraphs, only one of the lower-level policy points is input into the large language model for processing each time, and the mixed upper-level policy point set and the second target sample data set are also input each time.

[0138] For example, the mixed upper-level policy point set is M, the second target sample data set is N, and the matching lower-level policy paragraphs include lower-level policy points y11, y12, and y13. The mixed upper-level policy point set M, the second target sample data set N, and the lower-level policy point y11 are input into the large language model for processing, and the large language model will output the number that matches the lower-level policy point y11. Then the mixed upper-level policy point set M, the second target sample data set N, and the lower-level policy point y12 are input into the large language model for processing, and the large language model will output the number that matches the lower-level policy point y12. Then the mixed upper-level policy point set M, the second target sample data set N, and the lower-level policy point y13 are input into the large language model for processing, and the large language model will output the number that matches the lower-level policy point y13.

[0139] Step S46: Determine the matching result between the subordinate policy points in the mutually matching subordinate policy paragraphs and the superior policy points in the mutually matching superior policy paragraphs according to the corresponding numbers of the subordinate policy points in the mutually matching subordinate policy paragraphs and the second mapping relationship library.

[0140] In this embodiment, the method for determining the matching situation between each subordinate policy point in the mutually matching subordinate policy paragraphs and the superior policy paragraph in the mutually matching superior policy paragraphs is the same, and a subordinate policy point is used as an example for explanation. After obtaining the corresponding number of the subordinate policy point in the mutually matching subordinate policy paragraphs through steps S41 to S45, the superior policy point corresponding to the number in the constructed second mapping relationship library is queried to determine which superior policy point in the superior policy paragraph matches the subordinate policy point. It should be understood that for subordinate policy points that do not have matching numbers, there are correspondingly no superior policy points that match them, and it can be directly obtained that the matching situation of the subordinate policy point is that there is no superior policy point that matches it in the mutually matching superior policy paragraphs.

[0141] In combination with the above embodiments, in one implementation, the embodiment of the present application also provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S46 may include: when the subordinate policy point in the mutually matching subordinate policy paragraph has no corresponding number, determining that there is no superior policy point in the mutually matching superior policy paragraph that matches the subordinate policy point; when the subordinate policy point in the mutually matching subordinate policy paragraph has a corresponding number, determining the target superior policy point corresponding to the number in the second mapping relationship library; when the target superior policy point belongs to the superior policy point in the second target sample data set, determining that there is no superior policy point in the mutually matching superior policy paragraph that matches the subordinate policy point; when the target superior policy point belongs to the superior policy point in the mutually matching superior policy paragraph, determining that the subordinate policy point matches the target superior policy point.

[0142] In this embodiment, for a subordinate policy paragraph with no matching number in the mutually matching subordinate policy paragraphs, the corresponding matching situation is determined as a superior policy point with no matching subordinate policy point in the mutually matching superior policy paragraphs.

[0143] In this embodiment, for a subordinate policy paragraph with a matching number in the mutually matching subordinate policy paragraphs, determine the superior policy point corresponding to the number in the second mapping relationship library, and refer to the superior policy point as the target superior policy point. Then determine whether the target superior policy point belongs to the superior policy point in the second target sample data set, or to the superior policy point in the mutually matching superior policy paragraph. In the case where the target superior policy point belongs to the superior policy point in the second target sample data set, it is also determined that the corresponding matching situation is that there is no superior policy point in the mutually matching superior policy paragraph that matches the subordinate policy point. And in the case where the target superior policy point belongs to the superior policy point in the mutually matching superior policy paragraph, it is determined that the corresponding matching situation is that the target superior policy point matches the subordinate policy point.

[0144] In combination with the above embodiments, in one implementation, the embodiment of the present application further provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, the method further includes: determining whether there are multiple subordinate policy points that match the same superior policy point among the mutually matching subordinate policy points and the target superior policy point; merging the multiple subordinate policy points that match the same superior policy point to obtain a merged single subordinate policy point that matches the same superior policy point.

[0145] In this embodiment, after obtaining the policy point matching result, the upper policy points and lower policy points that form a matching relationship with each other in the upper policy paragraph and the lower policy paragraph that match each other are extracted to determine whether there are multiple lower policy points that match with one upper policy point at the same time. If there are, the multiple lower policy points are merged to obtain a merged lower policy point, and then it is determined that the merged lower policy point matches with the upper policy point. For example, in the upper policy paragraph X4 and the lower policy paragraph Y4 that match each other, the upper policy paragraph X4 includes upper policy points x41 and x42, and the lower policy paragraph Y4 includes lower policy points y41, y42, and y43. Through policy point matching, it is determined that the lower policy points y41 and y42 match with the upper policy point x41 at the same time, and the lower policy point y43 matches with the upper policy point x42. Therefore, the lower policy points y41 and y42 are merged to form a new lower policy point that matches with the upper policy point x41.

[0146] In combination with the above embodiments, in one implementation, the embodiment of the present application also provides a policy comparison method based on a large language model. In the policy comparison method based on a large language model, step S42 may include: determining the similarity between the superior policy points in the second sample data set and the superior policy points in the mutually matching superior policy paragraphs; determining the superior policy points in the second sample data set whose similarity with the superior policy points in the mutually matching superior policy paragraphs exceeds a second threshold as duplicate superior policy points; filtering the duplicate superior policy points in the second sample data set; mixing the superior policy points in the filtered second sample data set with the superior policy points in the mutually matching superior policy paragraphs to obtain a mixed superior policy point set.

[0147] In this embodiment, in order to prevent the superior policy points in the second sample data set that are particularly similar to the superior policy points in the mutually matching superior policy paragraphs from affecting the subsequent policy point matching of the large language model, the present application first filters the superior policy points in the second sample data set that are particularly similar to the superior policy points in the mutually matching superior policy paragraphs before mixing them.

[0148] Specifically: determine the similarity between each superior policy point in the second sample data set and each superior policy point in the mutually matching superior policy paragraphs, and then determine whether the similarity corresponding to each superior policy point in the second sample data set exceeds the second threshold, and if so, determine the superior policy point in the second sample data set as a duplicate superior policy point. Then filter out all superior policy points in the second sample data set that are determined to be duplicate superior policy points, and then shuffle and mix the remaining superior policy points in the filtered second sample data set with the superior policy points in the mutually matching superior policy paragraphs to obtain a mixed superior policy point set. Among them, the second threshold can be set according to the actual application scenario and is not specifically limited here. For example, the second sample data set includes superior policy points z11, z12, and z13, and the superior policy points in the superior policy paragraph include x11, x12, and x13. The similarities between the superior policy point z11 and the superior policy points x11, x12, and x13 are determined. If one of the similarities exceeds the second threshold, the superior policy point z11 is filtered out; the similarities between the superior policy point z12 and the superior policy points x11, x12, and x13 are determined. If one of the similarities exceeds the second threshold, the superior policy point z12 is filtered out; the similarities between the superior policy point z13 and the superior policy points x11, x12, and x13 are determined. If one of the similarities exceeds the second threshold, the superior policy point z13 is filtered out.

[0149] Based on the same inventive concept, the present application provides a policy comparison system based on a large language model, such as Figure 2 As shown, the system 200 includes:

[0150] The policy paragraph splitting module 201 is used to split the policy paragraphs of the upper and lower policy documents according to the title structure of the upper and lower policy documents, and obtain a list of upper policy paragraphs corresponding to the upper policy document and a list of lower policy paragraphs corresponding to the lower policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents;

[0151] A policy paragraph matching module 202 is used to perform policy paragraph matching on the lower-level policy paragraphs in the lower-level policy paragraph list and the upper-level policy paragraphs in the upper-level policy paragraph list to obtain corresponding paragraph matching results;

[0152] A policy point splitting module 203 is used to split the policy points of the matched upper policy paragraphs and lower policy paragraphs respectively based on the paragraph matching results, and obtain each upper policy point in the matched upper policy paragraphs and each lower policy point in the lower policy paragraphs;

[0153] A policy point matching module 204 is used to perform policy point matching on each upper policy point in the upper policy paragraphs and each lower policy point in the lower policy paragraphs that match each other, and obtain corresponding policy point matching results;

[0154] The matching analysis module 205 is used to analyze the corresponding matching situations of the paragraph matching results and the policy point matching results using the corresponding large language model prompt words to obtain the corresponding matching analysis results.

[0155] Optionally, the policy paragraph splitting module 201 includes:

[0156] A title determination module is used to determine the titles in the upper and lower policy files by matching the title serial numbers in the upper and lower policy files through setting rules;

[0157] A policy content determination module is used to determine the content between titles and between a title and the end of a policy document as the policy content of the previous title according to the titles in the determined superior and subordinate policy documents;

[0158] The policy paragraph list determination module is used to create a superior policy paragraph list corresponding to the superior policy file and a subordinate policy paragraph list corresponding to the subordinate policy file according to the determined titles in the superior and subordinate policy files and the policy contents corresponding to the titles.

[0159] Optionally, the policy paragraph matching module 202 includes:

[0160] A first sample data set construction module is used to construct a first sample data set including a first number of policy paragraph matching sample data and policy paragraph non-matching sample data, wherein the policy paragraph matching sample data is sample data composed of mutually matching upper-level policy paragraphs and lower-level policy paragraphs, and the policy paragraph non-matching sample data is lower-level policy paragraphs without mutually matching upper-level policy paragraphs;

[0161] A first mixing module, configured to mix the superior policy paragraphs in the superior policy paragraph list with the superior policy paragraphs in the first sample data set to obtain a mixed superior policy paragraph set;

[0162] A first numbering module, used for numbering the superior policy paragraphs in the mixed superior policy paragraph set to obtain a corresponding first mapping relationship library;

[0163] A first data conversion module, used to convert the sample data in the first sample data set into data suitable for processing by a large language model to obtain a first target sample data set;

[0164] A first matching analysis module is used to input the mixed upper-level policy paragraph set, the first target sample data set, and the lower-level policy paragraphs in the lower-level policy paragraph list into a large language model for processing, and obtain the numbers corresponding to the lower-level policy paragraphs;

[0165] The first matching relationship determination module is used to determine the matching results between the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list according to the numbers corresponding to the subordinate policy paragraphs in the subordinate policy paragraph list and the first mapping relationship library.

[0166] Optionally, the first matching relationship determination module includes:

[0167] A first determination submodule is used to determine that there is no upper-level policy paragraph in the upper-level policy paragraph list that matches the lower-level policy paragraph if there is no corresponding number in the lower-level policy paragraph list;

[0168] A second determination submodule is used to determine a target upper-level policy paragraph corresponding to the number in the first mapping relationship library when the lower-level policy paragraph in the lower-level policy paragraph list has a corresponding number;

[0169] A third determination submodule is used to determine that there is no upper-level policy paragraph in the upper-level policy paragraph list that matches the lower-level policy paragraph if the target upper-level policy paragraph belongs to the upper-level policy paragraph in the first target sample data set;

[0170] The fourth determination submodule is configured to determine that the subordinate policy paragraph matches the target superior policy paragraph when the target superior policy paragraph belongs to a superior policy paragraph in the superior policy paragraph list.

[0171] Optionally, the first mixing module comprises:

[0172] A first similarity determination module, configured to determine the similarity between the superior policy paragraphs in the first sample data set and the superior policy paragraphs in the superior policy paragraph list;

[0173] a duplicate superior policy paragraph determination module, configured to determine superior policy paragraphs in the first sample data set whose similarity with superior policy paragraphs in the superior policy paragraph list exceeds a first threshold as duplicate superior policy paragraphs;

[0174] A first filtering module, configured to filter repeated superior policy paragraphs in the first sample data set;

[0175] The first mixing submodule is used to mix the upper-level policy paragraphs in the filtered first sample data set with the upper-level policy paragraphs in the upper-level policy paragraph list to obtain a mixed upper-level policy paragraph set.

[0176] Optionally, the policy point matching module 204 includes:

[0177] A second sample data set construction module is used to construct a second sample data set including a second number of policy point matching sample data and policy point non-matching sample data, wherein the policy point matching sample data is sample data composed of mutually matching upper-level policy points and lower-level policy points, and the policy point non-matching sample data is lower-level policy points without mutually matching upper-level policy points;

[0178] A second mixing module, configured to mix the superior policy points in the mutually matching superior policy paragraphs with the superior policy points in the second sample data set to obtain a mixed superior policy point set;

[0179] A second numbering module, used for numbering the superior policy points in the mixed superior policy point set to obtain a corresponding second mapping relationship library;

[0180] A second data conversion module, used to convert the sample data in the second sample data set into data suitable for processing by a large language model to obtain a second target sample data set;

[0181] A second matching analysis module is used to input the mixed upper-level policy point set, the second target sample data set and the lower-level policy points in the mutually matched lower-level policy paragraphs into a large language model for processing to obtain the serial numbers corresponding to the lower-level policy points;

[0182] The second matching relationship determination module is used to determine the matching results between the subordinate policy points in the mutually matching subordinate policy paragraphs and the superior policy points in the mutually matching superior policy paragraphs based on the corresponding numbers of the subordinate policy points in the mutually matching subordinate policy paragraphs and the second mapping relationship library.

[0183] Optionally, the second matching relationship determination module includes:

[0184] A fifth determination submodule, configured to determine that there is no upper-level policy point in the mutually matching upper-level policy paragraph that matches the lower-level policy point if there is no corresponding number in the lower-level policy paragraph that matches the lower-level policy point;

[0185] A sixth determination submodule, configured to determine, when the subordinate policy points in the mutually matching subordinate policy paragraphs have corresponding numbers, a target superior policy point corresponding to the number in the second mapping relationship library;

[0186] A seventh determination submodule, configured to determine, when the target superior policy point belongs to a superior policy point in the second target sample data set, that there is no superior policy point in the mutually matched superior policy paragraphs that matches the subordinate policy point;

[0187] An eighth determination submodule is configured to determine that the subordinate policy point matches the target superior policy point when the target superior policy point belongs to a superior policy point in the mutually matching superior policy paragraph.

[0188] Optionally, the system 200 further includes:

[0189] A policy point matching relationship determination module is used to determine whether there are multiple subordinate policy points that match each other and the target superior policy point that match the same superior policy point;

[0190] The policy point merging module is used to merge the multiple subordinate policy points that match the same superior policy point to obtain a merged single subordinate policy point that matches the same superior policy point.

[0191] Optionally, a second hybrid module includes:

[0192] A second similarity determination module, used to determine the similarity between the superior policy point in the second sample data set and the superior policy point in the mutually matching superior policy paragraph;

[0193] a duplicate superior policy point determination module, configured to determine superior policy points in the second sample data set whose similarity with superior policy points in the mutually matching superior policy paragraph exceeds a second threshold as duplicate superior policy points;

[0194] A second filtering module, used to filter repeated superior policy points in the second sample data set;

[0195] The second mixing module is used to mix the upper-level policy points in the filtered second sample data set with the upper-level policy points in the mutually matching upper-level policy paragraphs to obtain a mixed upper-level policy point set.

[0196] As for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0197] It should be noted that, for the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0198] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0199] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the embodiments of the present application may adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the embodiments of the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0200] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0201] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0202] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable terminal device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0203] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0204] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal 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 terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0205] The above is a detailed introduction to a policy comparison method and system based on a large language model provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A policy comparison method based on a large language model, characterized in that: The method comprises: According to the title structure of the superior and subordinate policy documents, the superior and subordinate policy documents are split into policy paragraphs to obtain a superior policy paragraph list corresponding to the superior policy document and a subordinate policy paragraph list corresponding to the subordinate policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents; Perform policy paragraph matching on the lower-level policy paragraphs in the lower-level policy paragraph list and the upper-level policy paragraphs in the upper-level policy paragraph list to obtain corresponding paragraph matching results; Based on the paragraph matching results, policy point splitting is performed on the mutually matched upper-level policy paragraphs and lower-level policy paragraphs respectively, so as to obtain each upper-level policy point in the mutually matched upper-level policy paragraphs and each lower-level policy point in the lower-level policy paragraphs; Perform policy point matching on each superior policy point in the mutually matching superior policy paragraphs and each subordinate policy point in the subordinate policy paragraphs to obtain corresponding policy point matching results; For various types of matching situations in the paragraph matching results and the policy point matching results, corresponding large language model prompt words are used to analyze the corresponding matching situations to obtain corresponding matching analysis results.

2. According to the large language model-based policy comparison method of claim 1, it is characterized in that: According to the title structure of the superior and subordinate policy files, the superior and subordinate policy files are split into policy paragraphs to obtain a superior policy paragraph list corresponding to the superior policy file and a subordinate policy paragraph list corresponding to the subordinate policy file, including: Determine the titles in the upper and lower policy files by setting rules to match the title serial numbers in the upper and lower policy files; Based on the titles of the determined superior and subordinate policy documents, the content between titles and between a title and the end of a policy document is determined as the policy content of the previous title; Based on the determined titles in the superior and subordinate policy documents and the policy contents corresponding to the titles, a superior policy paragraph list corresponding to the superior policy document and a subordinate policy paragraph list corresponding to the subordinate policy document are created.

3. According to the large language model-based policy comparison method of claim 1, it is characterized in that: The lower policy paragraphs in the lower policy paragraph list are matched with the upper policy paragraphs in the upper policy paragraph list to obtain corresponding paragraph matching results, including: Constructing a first sample data set including a first number of policy paragraph matching sample data and policy paragraph non-matching sample data, wherein the policy paragraph matching sample data is sample data consisting of mutually matching upper-level policy paragraphs and lower-level policy paragraphs, and the policy paragraph non-matching sample data is lower-level policy paragraphs without mutually matching upper-level policy paragraphs; Mixing the superior policy paragraphs in the superior policy paragraph list with the superior policy paragraphs in the first sample data set to obtain a mixed superior policy paragraph set; Numbering the superior policy paragraphs in the mixed superior policy paragraph set to obtain a corresponding first mapping relationship library; Converting the sample data in the first sample data set into data suitable for processing by a large language model to obtain a first target sample data set; Inputting the mixed upper-level policy paragraph set, the first target sample data set, and the lower-level policy paragraphs in the lower-level policy paragraph list into a large language model for processing, and obtaining numbers corresponding to the lower-level policy paragraphs; According to the serial numbers corresponding to the subordinate policy paragraphs in the subordinate policy paragraph list and the first mapping relationship library, a matching result between the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list is determined.

4. According to claim 3, a policy comparison method based on a large language model is characterized in that: Determining, according to the serial numbers corresponding to the subordinate policy paragraphs in the subordinate policy paragraph list and the first mapping relationship library, matching results between the subordinate policy paragraphs in the subordinate policy paragraph list and the superior policy paragraphs in the superior policy paragraph list, including: In the case where the subordinate policy paragraph in the subordinate policy paragraph list has no corresponding number, determining that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph; In the case where the subordinate policy paragraph in the subordinate policy paragraph list has a corresponding number, determining a target superior policy paragraph corresponding to the number in the first mapping relationship library; In the case where the target superior policy paragraph belongs to the superior policy paragraph in the first target sample data set, determining that there is no superior policy paragraph in the superior policy paragraph list that matches the subordinate policy paragraph; In the case that the target superior policy paragraph belongs to the superior policy paragraphs in the superior policy paragraph list, it is determined that the subordinate policy paragraph matches the target superior policy paragraph.

5. The policy comparison method based on a large language model according to claim 3, characterized in that: Mixing the superior policy paragraphs in the superior policy paragraph list with the superior policy paragraphs in the first sample data set to obtain a mixed superior policy paragraph set, including: Determining the similarity between the superior policy paragraphs in the first sample data set and the superior policy paragraphs in the superior policy paragraph list; Determine the superior policy paragraphs in the first sample data set whose similarity with the superior policy paragraphs in the superior policy paragraph list exceeds a first threshold as duplicate superior policy paragraphs; Filtering repeated superior policy paragraphs in the first sample data set; The upper-level policy paragraphs in the filtered first sample data set are mixed with the upper-level policy paragraphs in the upper-level policy paragraph list to obtain a mixed upper-level policy paragraph set.

6. The policy comparison method based on a large language model according to claim 1, characterized in that: The policy points of each superior policy point in the mutually matching superior policy paragraphs and each subordinate policy point in the subordinate policy paragraphs are matched to obtain corresponding policy point matching results, including: Constructing a second sample data set including a second number of policy point matching sample data and policy point non-matching sample data, wherein the policy point matching sample data is sample data composed of mutually matching upper-level policy points and lower-level policy points, and the policy point non-matching sample data is lower-level policy points without mutually matching upper-level policy points; Mixing the superior policy points in the mutually matching superior policy paragraphs with the superior policy points in the second sample data set to obtain a mixed superior policy point set; Numbering the superior policy points in the mixed superior policy point set to obtain a corresponding second mapping relationship library; Converting the sample data in the second sample data set into data suitable for processing by a large language model to obtain a second target sample data set; Inputting the mixed upper-level policy point set, the second target sample data set, and the lower-level policy points in the mutually matched lower-level policy paragraphs into a large language model for processing to obtain numbers corresponding to the lower-level policy points; According to the corresponding numbers of the subordinate policy points in the mutually matching subordinate policy paragraphs and the second mapping relationship library, the matching results between the subordinate policy points in the mutually matching subordinate policy paragraphs and the superior policy points in the mutually matching superior policy paragraphs are determined.

7. A policy comparison method based on a large language model according to claim 6, characterized in that: Determining the matching result between the lower-level policy point in the lower-level policy paragraphs and the upper-level policy point in the upper-level policy paragraphs according to the serial numbers corresponding to the lower-level policy points in the lower-level policy paragraphs and the second mapping relationship library, including: In the case where the subordinate policy point in the mutually matching subordinate policy paragraph has no corresponding number, determining that there is no superior policy point in the mutually matching superior policy paragraph that matches the subordinate policy point; In the case where the subordinate policy points in the mutually matching subordinate policy paragraphs have corresponding numbers, determining a target superior policy point corresponding to the number in the second mapping relationship library; In the case where the target superior policy point belongs to the superior policy point in the second target sample data set, determining that there is no superior policy point in the mutually matched superior policy paragraphs that matches the subordinate policy point; In the case where the target superior policy point belongs to the superior policy point in the mutually matching superior policy paragraphs, it is determined that the subordinate policy point and the target superior policy point match each other.

8. The policy comparison method based on a large language model according to claim 7, characterized in that: The method further comprises: Determine whether there are multiple subordinate policy points that match each other and the target superior policy point that match the same superior policy point; The multiple subordinate policy points that match the same superior policy point are merged to obtain a merged single subordinate policy point that matches the same superior policy point.

9. The policy comparison method based on a large language model according to claim 6, characterized in that: Mixing the superior policy points in the mutually matching superior policy paragraphs with the superior policy points in the second sample data set to obtain a mixed superior policy point set, including: Determining the similarity between the superior policy point in the second sample data set and the superior policy point in the mutually matching superior policy paragraph; Determine the superior policy points in the second sample data set whose similarity with the superior policy points in the mutually matching superior policy paragraph exceeds a second threshold as duplicate superior policy points; filtering repeated superior policy points in the second sample data set; The upper-level policy points in the filtered second sample data set are mixed with the upper-level policy points in the mutually matching upper-level policy paragraphs to obtain a mixed upper-level policy point set.

10. A policy comparison system based on a large language model, characterized in that: The system comprises: A policy paragraph splitting module is used to split the policy paragraphs of the upper and lower policy documents according to their title structures, and obtain a list of upper-level policy paragraphs corresponding to the upper-level policy document and a list of lower-level policy paragraphs corresponding to the lower-level policy document, wherein the elements in the policy paragraph list are policy paragraphs including policy titles and policy contents; A policy paragraph matching module, used to perform policy paragraph matching on the lower-level policy paragraphs in the lower-level policy paragraph list and the upper-level policy paragraphs in the upper-level policy paragraph list to obtain corresponding paragraph matching results; A policy point splitting module is used to split the policy points of the mutually matched upper policy paragraphs and lower policy paragraphs respectively based on the paragraph matching results, so as to obtain each upper policy point in the mutually matched upper policy paragraphs and each lower policy point in the lower policy paragraphs; A policy point matching module is used to match each upper policy point in the upper policy paragraphs and each lower policy point in the lower policy paragraphs that match each other, and obtain corresponding policy point matching results; The matching analysis module is used to analyze the corresponding matching situations of the paragraph matching results and the policy point matching results using the corresponding large language model prompt words to obtain the corresponding matching analysis results.

Images