Enterprise policy recommendation method, equipment and medium

Abstract

The invention discloses an enterprise policy recommendation method and device and a medium, and the method comprises the steps: extracting state identifiers through querying features of policy records in a policy database, and generating an unextracted policy ID list; according to the unextracted policy ID list, feature extraction is carried out on each unextracted policy, and a policy portrait corresponding to each unextracted policy is generated; according to a Levenshtein editing distance algorithm and a preset prior rule, performing similarity calculation on the policy portrait and a preset enterprise portrait to obtain the similarity between the enterprise portrait and each policy portrait; according to the similarity, performing descending sorting on each policy portrait to obtain a candidate policy sorting table; and inputting the candidate policy sorting table and a preset enterprise portrait into a preset language large model to obtain a recommendation policy table corresponding to the enterprise portrait. According to the policy recommendation method and device, the long text policy is decomposed through extraction of all dimensions, the policy recommendation efficiency is improved, and the policy recommendation accuracy is improved through combined calculation of the structural features and the semantic similarity.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, device and medium for recommending corporate policies. Background Technology

[0002] Currently, AI-powered policy recommendation systems are widely used in various fields, primarily employing technologies such as collaborative filtering, knowledge graphs, and generative large models.

[0003] Most existing AI-powered policy recommendation methods rely on user behavior data, using instruction fine-tuning and semantic alignment to match enterprises with policies. However, due to the lag in understanding professional terminology and the limitations of context windows in large models, it is difficult to accurately extract features from long policy texts when faced with policy texts containing a large amount of professional terminology or when matching multiple dimensions such as industry classification, geographical scope, and enterprise size. This results in a decrease in policy matching accuracy and fails to meet the necessary needs of enterprises for policy recommendations. Summary of the Invention

[0004] This application provides a method, device, and medium for recommending enterprise policies, which addresses the problem of reduced policy matching accuracy caused by the inability to accurately extract policy features from long texts in existing policy recommendation methods.

[0005] The embodiments of this application adopt the following technical solutions:

[0006] On one hand, this application provides a method for recommending enterprise policies. The method includes: generating a list of unextracted policy IDs by querying the feature extraction status identifiers of policy records in a policy database; extracting features from each unextracted policy based on the list of unextracted policy IDs to generate a policy profile corresponding to each unextracted policy; calculating the similarity between the policy profiles and preset enterprise profiles based on the Levenshtein edit distance algorithm and preset prior rules to obtain the similarity between the enterprise profiles and each policy profile; sorting each policy profile in descending order based on the similarity to obtain a candidate policy ranking table; and inputting the candidate policy ranking table and the preset enterprise profiles into a preset language big data model to obtain a recommended policy table corresponding to each enterprise profile.

[0007] In one example, based on the list of unextracted policy IDs, feature extraction is performed on each unextracted policy to generate a policy profile for each unextracted policy. Specifically, this includes: combining the policy title, tags, and content of each unextracted policy according to a preset language model prompt word format to obtain a policy concatenation combination for each unextracted policy; inputting the policy concatenation combination into a preset language model to obtain a primary policy profile for each unextracted policy; and converting the feature data in the primary policy profile into a structured object using regular expressions, and determining the structured object corresponding to each unextracted policy as the policy profile.

[0008] In one example, based on the Levenshtein edit distance algorithm and preset prior rules, the similarity between the policy profile and the preset enterprise profile is calculated to obtain the similarity between the enterprise profile and each policy profile. Specifically, this includes: using the Levenshtein edit distance algorithm to calculate the text difference between the applicable enterprise description of each policy profile and the corresponding enterprise introduction of the enterprise profile, to obtain the text similarity between the enterprise profile and each policy profile; according to the preset prior rules, matching the enterprise profile and each policy profile for similarity based on region, industry, and scale, to obtain the rule similarity between the enterprise profile and each policy profile; and summing the text similarity and rule similarity using preset similarity score weights to obtain the similarity between the enterprise profile and each policy profile.

[0009] In one example, based on preset prior rules, the enterprise profile is matched with each policy profile for similarity in terms of region, industry, and size to obtain the rule similarity between the enterprise profile and each policy profile. Specifically, this includes: matching the enterprise profile with each policy profile for industry similarity based on preset industry hierarchy priority logic to obtain industry similarity; matching the enterprise profile with each policy profile for region similarity based on the geographical scope of application of the policy profile to obtain region similarity; matching the enterprise profile with each policy profile for size similarity based on the mapping rules between enterprise size and different types of enterprises to obtain size similarity; and performing an AND operation on the industry similarity, region similarity, and size similarity to obtain the rule similarity.

[0010] In one example, each policy profile is sorted in descending order based on the similarity to obtain a candidate policy ranking table. Specifically, this includes: sorting each policy profile in descending order based on the similarity to the enterprise profile; if there are policy profiles with the same similarity during the descending sorting process, they are sorted in descending order based on the policy release time, with the later the policy release time, the higher the sorting priority.

[0011] In one example, the candidate policy ranking table and the preset enterprise profile are input into a preset language model to obtain a recommended policy table corresponding to the enterprise profile and the reasons for the recommendation. Specifically, this includes: combining the preset enterprise profile and the policy profile in the candidate policy ranking table according to the preset final prompt word format to obtain the final policy concatenation combination; inputting the final policy concatenation combination into the preset language model to generate recommended policy data in JSON format; the recommended policy data includes matching analysis, recommendation value, and reasons for the value; and filtering the recommended policy data according to a preset policy quantity threshold to obtain the recommended policy table corresponding to the enterprise profile.

[0012] In one example, after filtering the policy data output according to a preset policy quantity threshold to obtain a recommended policy table corresponding to the enterprise profile, the method further includes: generating a recommendation reason for each policy in the recommended policy table based on the matching analysis, recommendation value, and value reasons of each policy, and storing the recommendation reasons on the blockchain.

[0013] In one example, the method also includes: a hierarchical extraction format for the preset language model prompt words; the first-level prompt instructions extract the structured features of the policy title, applicable industry classification, applicable geographical scope, and applicable enterprise size; and the second-level prompt instructions generate a text summary of the target enterprise description based on the full text of the policy.

[0014] On the other hand, embodiments of this application provide an enterprise policy recommendation device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform an enterprise policy recommendation method as described above.

[0015] On the other hand, embodiments of this application provide a non-volatile computer storage medium for recommending enterprise policies, which stores computer-executable instructions capable of executing any of the above-mentioned enterprise policy recommendation methods.

[0016] The at least one technical solution adopted in this application embodiment can achieve the following beneficial effects: By extracting structured features and content summaries of policies through a hierarchical large-scale model suggestion strategy, the problem of processing long text policies is solved. Combining the edit distance algorithm with multi-dimensional rule constraints to calculate the comprehensive matching score significantly improves the policy applicability rate and enhances matching accuracy. Furthermore, it reduces the average time for enterprises to obtain suitable policies from several hours to within seconds, improving policy reach efficiency. By generating explainable recommendation reasons containing industry fit points and support value through a large-scale model, the problem of insufficient explainability in traditional recommendation technologies is solved, enabling enterprises to clearly understand the policy matching logic and increasing their trust in the recommendation results. Attached Figure Description

[0017] To more clearly illustrate the technical solution of this application, some embodiments of this application will be described in detail below with reference to the accompanying drawings, in which:

[0018] Figure 1 A flowchart illustrating a method for recommending corporate policies provided in this application embodiment;

[0019] Figure 2 A schematic diagram of feature extraction code for an enterprise policy recommendation method provided in this application embodiment;

[0020] Figure 3 A schematic diagram of the final prompt word code for a corporate policy recommendation method provided in this application embodiment;

[0021] Figure 4 This is a schematic diagram of the structure of a corporate policy recommendation device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] Some embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart illustrating a corporate policy recommendation method provided in this application embodiment. This method can be applied to different business areas. Certain input parameters or intermediate results in this process can be manually adjusted to help improve accuracy.

[0025] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a controller as an example.

[0026] Based on this Figure 1 The process may include the following steps:

[0027] S101: Generate a list of unextracted policy IDs by querying the feature extraction status identifiers of policy records in the policy database; the policy database data is obtained by collecting policies from government websites within a preset collection period; the extracted policies and unextracted policies in the policy database have different feature extraction status identifiers.

[0028] In some embodiments of this application, to carry out the enterprise policy recommendation process, it is first necessary to establish an enterprise policy database. The policy database includes policies that the enterprise is already using, as well as policies that have been recently collected and have not yet been matched.

[0029] It should be noted that this application will set a policy collection cycle (e.g., once a week) to regularly collect newly released policies from government websites or other official websites and store them in the policy database.

[0030] Furthermore, identifiers are added to the policies in the policy database. Taking binary identifiers as an example, "0" indicates that no features have been extracted, and "1" indicates that features have been extracted. By querying the records in the policy database table that are identified as "0", a list of unextracted policy IDs is generated.

[0031] S102: Based on the list of unextracted policy IDs, feature extraction is performed on each unextracted policy to generate a policy profile corresponding to each unextracted policy; the policy profile includes the applicable industry, geographical scope, applicable enterprise size, and applicable enterprise description.

[0032] In some embodiments of this application, after generating a list of policy IDs for which no policies have been extracted, feature extraction is performed on each policy in the list. Specifically, the policy title, policy tag, and full text of the policy are concatenated into a preset large model prompt (including role definitions, JSON output format, and field explanations) to obtain a policy concatenation combination corresponding to each policy for which no policy has been extracted. Then, the policy concatenation combination is input into a preset language large model to generate a primary policy profile in JSON format, which includes national standard industry (major / medium / minor category), enterprise information tracing industry (category / major / medium / minor category), applicable province, applicable city, applicable enterprise size, and applicable enterprise description. The JSON string of the primary policy profile is then extracted using the regular expression "\s*json(.*?)" and converted into a structured policy feature object. Finally, the structured object corresponding to each policy for which no policy has been extracted is determined as the policy profile.

[0033] It should be noted that the preset language model prompt word format adopts a hierarchical extraction format; the first-level prompt instructions extract the structured features of policy title, applicable industry classification, applicable geographical scope, and applicable enterprise size; the second-level prompt instructions generate a text summary of the target enterprise description based on the full text of the policy.

[0034] This application solves the problem of processing long policy texts by extracting structured features and content summaries of policies through a hierarchical large model suggestion strategy, thereby improving the efficiency and accuracy of enterprise policy recommendations.

[0035] S103: Based on the Levenshtein edit distance algorithm and preset prior rules, the similarity between the policy profile and the preset enterprise profile is calculated to obtain the similarity between the enterprise profile and each policy profile.

[0036] In some embodiments of this application, after obtaining the policy profile during the feature extraction stage, the policy profile is matched with a pre-built enterprise profile (including enterprise industry classification, registered region, size, and brief introduction) for similarity. Specifically:

[0037] First, the Levenshtein edit distance algorithm is used to calculate the textual difference between the applicable enterprise description and the corresponding enterprise profile for each policy profile. The core idea is to transform one string into another with the fewest possible editing operations (insertion, deletion, and character replacement). More operations indicate a greater distance between the two strings, resulting in lower similarity, while fewer operations indicate higher similarity. The calculation formula is as follows:

[0038] Edit distance = min(insertion count, deletion count, replacement count)

[0039] Then, the similarity score, T, is calculated using normalization. 政策 Introduction of target companies representing the policy profile, T 公司 The company profile represents the enterprise's identity, with a similarity range of [0,1] (0 = no match, 1 = perfect match). The final formula is as follows:

[0040]

[0041] This formula yields the textual similarity between the enterprise profile and each policy profile.

[0042] At the same time, based on the preset prior rules, the similarity between the enterprise profile and each policy profile is matched by region, industry and scale to obtain the rule similarity between the enterprise profile and each policy profile.

[0043] The industry matching follows a hierarchical priority logic. If the policy specifies a sub-category, the company's sub-category must match. The matching logic is to determine whether the company's industry belongs to the policy industry set. If it belongs to the policy industry set, the intermediate and major categories are checked for matching. Qichacha's industry matching logic is the same as the national standard, thus assigning industry similarity.

[0044] Based on the geographical scope of application of the policy profile, the regional matching is performed to match the enterprise profile with each policy profile based on geographical similarity. The company's province must be in the list of provinces where the policy applies, and the company's city or district must be in the list of cities where the policy applies (multi-level matching is supported, such as if the policy says "Yuhang District of Hangzhou City", it can match the company's district "Yuhang District"), thus obtaining the regional similarity.

[0045] Based on the mapping rules between enterprise size and various types of enterprises, the scale matching is performed to match the enterprise profile with each policy profile based on scale similarity. The policy requires that "large enterprises" correspond to "large" in terms of matching enterprise size; "small and micro enterprises" correspond to "medium, small, and micro"; "micro and small enterprises" correspond to "small and micro"; "micro enterprises" correspond to "micro"; and "unlimited" corresponds to all sizes, thus obtaining the scale similarity.

[0046] Then, the industry similarity, regional similarity, and scale similarity are ANDed to obtain the rule similarity.

[0047] It should be noted that industry similarity, regional similarity, and scale similarity only have two values: 0 and 1. 0 represents no match, and 1 represents a match. The AND operation 0 ∩ 1 = 0.

[0048] Finally, by using preset similarity score weights, the text similarity and rule similarity are summed to obtain the similarity between the enterprise profile and each policy profile. The formula is:

[0049] S 综合 =α·S 文本 +β·S 规则

[0050] Where α = 0.4 and β = 0.6.

[0051] This application significantly improves the policy applicability rate and enhances the matching accuracy by combining the edit distance algorithm with multi-dimensional rule constraints to calculate the comprehensive matching score. It also reduces the average time for enterprises to obtain applicable policies from several hours to within seconds, thereby improving the efficiency of policy reach.

[0052] S104: Based on the similarity, sort each policy profile in descending order to obtain a candidate policy ranking table.

[0053] In some embodiments of this application, after obtaining the similarity between the enterprise profile and each policy profile, the similarity is calculated according to S. 综合 Sort in descending order to ensure the most relevant policies are listed first.

[0054] If multiple policies have the same score, they are further sorted by the most recent policy release date, with later release dates receiving higher priority. Furthermore, a large K value (e.g., K=50) is set to ensure sufficient coverage of candidate policies and avoid missing potential matches.

[0055] The final output is a candidate policy ranking table with the structure (Company ID, [Policy ID1, Policy ID2, ..., Policy IDK]), which is used for subsequent fine ranking.

[0056] S105: Input the candidate policy ranking table and the preset enterprise profile into the preset language big model to obtain the recommended policy table corresponding to the enterprise profile.

[0057] In some embodiments of this application, the preset enterprise profile and the policy profile in the candidate policy ranking table are combined according to the preset final prompt word format to obtain the final policy splicing combination;

[0058] Furthermore, the final policy data is concatenated and input into a pre-defined language model to generate policy data output in JSON format; the policy data output includes matching analysis, recommendation value, and reasons for the value.

[0059] Based on a preset policy quantity threshold, the policy data output is filtered to obtain a recommended policy table corresponding to the enterprise profile.

[0060] Finally, based on the matching analysis, recommendation value, and reasons for the value of each policy in the recommended policy table, the recommendation reasons for each policy in the recommended policy table are generated and stored on the blockchain.

[0061] This application generates explainable recommendation reasons that include industry fit and support value through a large model, which solves the problem of insufficient explainability of traditional recommendation technology, enabling enterprises to clearly understand the policy matching logic and increasing their trust in the recommendation results.

[0062] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S101 to S105 will be described sequentially, but this does not mean that steps S101 and S105 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S101 to S105 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S101 to S105 can be appropriately adjusted according to actual needs.

[0063] pass Figure 1 This method extracts structured features and content summaries of policies through a hierarchical large-scale model suggestion strategy, solving the challenge of processing long-text policies. By combining edit distance algorithms with multi-dimensional rule constraints to calculate a comprehensive matching score, it significantly improves policy applicability and enhances matching accuracy. Furthermore, it reduces the average time for enterprises to obtain suitable policies from several hours to within seconds, improving policy outreach efficiency. The large-scale model generates explainable recommendation reasons including industry fit and support value, addressing the lack of interpretability in traditional recommendation technologies. This allows enterprises to clearly understand the policy matching logic, increasing their trust in the recommendation results.

[0064] Figure 2 This is a schematic diagram of feature extraction prompt words for a corporate policy recommendation method provided in an embodiment of this application.

[0065] exist Figure 2 The example demonstrates how to extract a policy profile for each piece of policy data using large-scale model prompts. The policy data to be concatenated includes the policy title, policy tags, and policy content. The returned result is a JSON string, which is then converted into a Python object using a string regular expression matching algorithm.

[0066] Figure 3 This is a schematic diagram of the final prompt word code for a corporate policy recommendation method provided in an embodiment of this application.

[0067] exist Figure 3 The document demonstrates the code for semantic-level fine-tuning of prompt words by combining the full text of the policy with enterprise profiles, and the output format is JSON.

[0068] Figure 4 A schematic diagram of a corporate policy recommendation device provided in this application embodiment includes:

[0069] At least one processor; and,

[0070] A memory that is communicatively connected to at least one processor; wherein,

[0071] The memory stores instructions that can be executed by at least one processor, which, when executed by at least one processor, enable the at least one processor to perform any of the above-mentioned enterprise policy recommendation methods.

[0072] Some embodiments of this application provide a non-volatile computer storage medium for recommending corporate policies, which stores computer-executable instructions capable of executing any of the above-described corporate policy recommendation methods.

[0073] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0074] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0075] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0077] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0079] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0080] Memory may include non-persistent storage in computer-readable media, random access memory (RAM), and non-volatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0081] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0082] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0083] The above are merely embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the technical principles of this application should fall within the protection scope of this application.

Claims

1. A method for recommending corporate policies, characterized in that, The method includes: By querying the feature extraction status identifiers of policy records in the policy database, a list of unextracted policy IDs is generated; the extracted policies and unextracted policies in the policy database have different feature extraction status identifiers. Based on the list of unextracted policy IDs, feature extraction is performed on each unextracted policy to generate a policy profile for each unextracted policy. The feature data of the policy profile includes the applicable industry, geographical scope of application, applicable enterprise size, and description of the applicable enterprise. Based on the Levenshtein edit distance algorithm and preset prior rules, the similarity between each policy profile and the enterprise profile of the enterprise to be recommended is calculated to obtain the similarity between the enterprise profile and each policy profile. Based on the similarity, each policy profile is sorted in descending order to obtain a candidate policy ranking table; Input the candidate policy ranking table and the enterprise profile into a preset language big model to obtain the recommended policy table corresponding to the enterprise profile.

2. The method according to claim 1, characterized in that, The step of extracting features from each unextracted policy based on the list of unextracted policy IDs to generate a policy profile for each unextracted policy specifically includes: Based on the preset language model prompt word format, the policy title, policy content, and policy tags of each unextracted policy are combined to obtain the policy splicing combination corresponding to each unextracted policy; The policies are spliced ​​and combined and input into a preset language model to obtain a primary policy profile for each policy that has not been extracted. The feature data in the primary policy profile is converted into structured objects using regular expressions, and each structured object corresponding to a policy that has not been extracted is identified as a policy profile.

3. The method according to claim 1, characterized in that, The similarity calculation between each policy profile and the enterprise profile of the enterprise to be recommended is performed based on the Levenshtein edit distance algorithm and preset prior rules to obtain the similarity between the enterprise profile and each policy profile. Specifically, this includes: The Levenshtein edit distance algorithm is used to calculate the textual difference between the applicable enterprise description of each policy profile and the corresponding enterprise profile, and the textual similarity between the enterprise profile and each policy profile is obtained. Based on the preset prior rules, the similarity between the enterprise profile and each policy profile is matched by region, industry and scale to obtain the rule similarity between the enterprise profile and each policy profile. The text similarity and the rule similarity are weighted and summed to obtain the similarity between the enterprise profile and each policy profile.

4. The method according to claim 3, characterized in that, The process involves matching the enterprise profile with each policy profile based on preset prior rules, considering factors such as region, industry, and scale, to obtain the rule similarity between the enterprise profile and each policy profile. Specifically, this includes: Based on the preset industry hierarchy priority logic, the enterprise profile is matched with each policy profile to obtain the industry similarity. Based on the geographical scope of application of the policy profile, the enterprise profile is matched with each policy profile to obtain the geographical similarity score. Based on the mapping rules between the applicable enterprise size of the policy profile and various types of enterprises, the enterprise profile is matched with each policy profile for size similarity to obtain the size similarity; the various types of enterprises specifically include large, medium and small enterprises and micro enterprises. The rule similarity is obtained by performing a bitwise AND operation on industry similarity, regional similarity, and scale similarity.

5. The method according to claim 1, characterized in that, The step of sorting each policy profile in descending order based on the similarity to obtain a candidate policy ranking table specifically includes: The similarity between each policy profile and the enterprise profile is sorted in descending order. If policy profiles with the same similarity exist during the descending sorting process, they are sorted in descending order according to the policy release time, with the later the policy release time, the higher the sorting priority.

6. The method according to claim 1, characterized in that, The step of inputting the candidate policy ranking table and the enterprise profile into a preset language big data model to obtain the recommended policy table corresponding to the enterprise profile specifically includes: Based on the preset final prompt word format, the preset enterprise profile and the policy profile in the candidate policy ranking table are combined to obtain the final policy splicing combination; The final policies are concatenated and combined and input into a preset language model to generate recommended policy data in JSON format; the recommended policy data includes matching analysis, recommendation value, and reasons for the value. Based on a preset policy quantity threshold, the recommended policy data is filtered to obtain a recommended policy table corresponding to the enterprise profile.

7. The method according to claim 6, characterized in that, After filtering the recommended policy data according to a preset policy quantity threshold to obtain a recommended policy table corresponding to the enterprise profile, the method further includes: Based on the matching analysis, recommendation value, and reasons for the value of each policy in the recommended policy table, the recommendation reasons for each policy in the recommended policy table are generated, and the recommendation reasons are stored on the blockchain.

8. The method according to claim 1, characterized in that, The method further includes: The preset language model prompt word format adopts a hierarchical extraction format; the first-level prompt instruction extracts the structured features of policy title, applicable industry classification, applicable geographical scope, and applicable enterprise size; the second-level prompt instruction generates a text summary of the target enterprise description based on the full text of the policy.

9. A device for recommending enterprise policies, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a corporate policy recommendation method as described in any one of claims 1-8.

10. A corporate policy recommendation storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of executing the enterprise policy recommendation method described in any one of claims 1-8.

Images