Enterprise service policy label extraction method and system based on natural language analysis
Through the enterprise service policy label extraction method based on natural language analysis, and the natural language analysis model is used to extract policy rule labels, the problem of insufficient flexibility and accuracy of label extraction in the existing technology is solved, and the accuracy and convenience of policy matching are achieved .
Patent Information
- Application Number
- CN202311603048.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-28
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-11-28
AI Technical Summary
The existing enterprise service policy label extraction method is not flexible and cannot meet the precise matching needs under diversified policy information. The number of preset labels is limited, which affects the accuracy of policy matching.
Adopt the enterprise service policy label extraction method based on natural language analysis, through policy text collection, establishing maps and information extraction, and use natural language analysis models to extract policy rule labels, including defining information maps, finding matching policy information maps, positioning words and Logical representation, and rule labels are formed according to preset rules, and the vocabulary is updated dynamically to improve matching accuracy.
It has achieved accurate extraction of policy rule labels, improved the accuracy and convenience of government departments providing policy inquiry services to enterprises, and can accurately match enterprises that have obtained policy subsidies.
Smart Images

Figure CN117609513B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of big data technology, and in particular to a method and system for extracting enterprise service policy labels based on natural language analysis. Background Art
[0002] At present, with the application of big data technology, there are technologies that guide the development of enterprises by collecting enterprise information, drawing enterprise user portraits, and matching them with policies issued by various governments and units based on this.
[0003] The patent with publication number CN116431899A discloses a method for policy matching recommendation for cultural enterprises through big data analysis. It cleans and processes the cultural enterprise data and creates labels, and pre-processes the policy data, performs feature selection, training optimization and creates labels, and finally performs label matching, thereby achieving the effect of matching suitable policies for a large number of cultural enterprises or finding suitable enterprises for different policies.
[0004] According to the above, the implementation of the above method needs to be realized through the combination of tags, etc., and the existing tag extraction methods are mostly:
[0005] 1) Preset tags: pre-define some fixed tags, and users can choose them when posting comments;
[0006] 2) Multi-level tag definition: pre-define the tag categories, then subdivide them level by level, and finally generate specific tags.
[0007] In the first type mentioned above, the number of tags is limited, which requires staff to update regularly and cannot meet all user needs; the second type has poor flexibility and affects the accuracy of matching policies for enterprises. Therefore, this application proposes a new technical solution. Summary of the invention
[0008] In order to conveniently and accurately meet the needs of enterprise policy matching services under the current diversified policy information, the present application provides an enterprise service policy label extraction method and system based on natural language analysis.
[0009] In the first aspect, the present application provides a method for extracting enterprise service policy labels based on natural language analysis, which adopts the following technical solutions:
[0010] A method for extracting enterprise service policy labels based on natural language analysis, including policy text collection, graph building and information extraction;
[0011] The policy text collection includes:
[0012] Obtaining policy texts and performing duplicate checks based on existing records in the policy database; and,
[0013] If the duplicate check result meets the preset storage standards, it will be stored in the policy database;
[0014] The establishing of the map comprises:
[0015] Define the information graph; and,
[0016] Generate corresponding policy information graphs based on the stored policy data and store them in the graph database;
[0017] The information extraction includes:
[0018] Define the problem data corresponding to the information to be extracted;
[0019] Search the graph database based on the problem data to obtain matching policy information graphs and suspected demand information;
[0020] Calling policy texts in the policy database based on the policy information map and locating them in the policy texts based on suspected demand information; and,
[0021] The located characters and words and the data and logical expressions that follow them are extracted, and rules are established according to preset labels to form rule labels; wherein the rule labels include label names, label logical operators and label values, and the rule labels are based on the logical relationships corresponding to the logical expressions to form a rule system.
[0022] Optionally, the establishing of the map includes:
[0023] Identify the policy information graphs based on the same policy text; and,
[0024] The policy information map is classified according to the policy text, the publisher / unit, and the contact person / unit displayed in the policy information map to obtain the policy catalog of each competent caliber and the corresponding map.
[0025] Optionally, the policy texts in the policy database are classified and stored according to the policy catalogs of each competent authority;
[0026] The extracted and located characters, words and the data and logical expressions following them include:
[0027] Match the corresponding vocabulary with each competent policy catalog;
[0028] When the duplicate checking results of the new policy text meet the preset vocabulary update standards, the new policy text will be used as a sample;
[0029] Compare the word library, extract the new words in the sample, and search the dictionary to obtain the definition;
[0030] Add the newly appeared words to the corresponding phrases in the vocabulary according to the definitions.
[0031] Optionally, the extracted located characters, words and the data and logical expressions following them also include:
[0032] Locate the new words in the sample and determine whether the words that represent the meaning appear in the first n and / or last m characters. If so, then:
[0033] Compare the content following the interpretation with the interpretation of the phrase in the corresponding vocabulary. If they are the same, add the phrase; if they are different, generate a new phrase based on the content following the interpretation.
[0034] Optionally, the updated information and / or newly added information in the phrases in the vocabulary is sent to a designated user terminal, and user feedback is received, and whether to implement the vocabulary update / addition action is determined based on the user feedback; or, if no feedback is received within a preset feedback deadline T, the vocabulary update / addition action is implemented.
[0035] Optionally, the duplicate checking process includes processing the policy text based on a locally sensitive hashing algorithm, comparing the results, and calculating the repetition rate.
[0036] Optionally, the duplicate checking process includes: before checking the content of the policy text for duplicates, first checking the name / number and publication time of the policy text for duplicates. If the result is 100% identical, stop checking the content for duplicates; if the result is not 100% identical, check the content for duplicates.
[0037] In the second aspect, the present application provides an enterprise service policy label extraction system based on natural language analysis, which adopts the following technical solutions:
[0038] A system for extracting enterprise service policy labels based on natural language analysis comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any of the above-mentioned methods for extracting enterprise service policy labels based on natural language analysis.
[0039] To summarize, the present application includes at least one of the following beneficial technical effects: the natural language analysis model can be used to realize the extraction of policy rule labels, so that when government departments provide policy query services to enterprises, they can accurately match enterprises that can obtain policy subsidies, providing convenience for enterprises or governments. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 is a schematic flow chart of the method of the present application;
[0041] Figure 2 It is a structural diagram of the policy information map of this application. DETAILED DESCRIPTION
[0042] The following is combined with Figure 1-2 This application is described in further detail.
[0043] The embodiment of the present application discloses a method for extracting enterprise service policy labels based on natural language analysis.
[0044] Reference Figure 1 ,The enterprise service policy label extraction method based on natural language analysis includes ,policy text collection, graph building and information extraction.
[0045] Policy texts refer to the policy documents issued by various government departments in the application scenarios targeted by this method. Policy text collection includes:
[0046] Obtaining policy texts and performing duplicate checks based on existing records in the policy database; and,
[0047] If the duplicate check results meet the preset storage standards, they will be stored in the policy database.
[0048] It is known that policy documents are sometimes released by multiple departments, and important policy documents will be forwarded through multiple channels and platforms after release. Therefore, if crawler technology is used to capture policy documents on the Internet, there is a high probability of duplicate text. If it is not checked for duplication, a large amount of unnecessary data analysis will be added.
[0049] In this embodiment, the duplicate checking process can be implemented through various duplicate checking algorithms, such as: processing the policy text based on the local sensitive hash algorithm, and comparing the results to calculate the repetition rate. The advantage of the local sensitive hash algorithm is that the text can be converted into two-dimensional data such as: 1001010001, which simplifies the comparison. However, considering that some policy documents are long and such documents will be actively distinguished in the name, in order not to waste computing power, before using the algorithm to check the content of the text for duplicate checking, the name / number and release time of the text are first checked for duplicate processing. If the result is 100% the same, stop; otherwise, check the content of the text for duplicate checking.
[0050] Regarding the establishment of the map, it includes:
[0051] Define an information graph; for example, define entities, attributes, and relationships, see Figure 2 , entities such as: Science and Technology Progress Award, Patent Award; attributes such as: award category, application time-December 20, 2023; relationships such as: affiliated association-Electronic Information Association, affiliated province-Province X.
[0052] After the staff has made the above definition, they can use the natural language analysis model as a tool to generate the corresponding policy information map based on the stored policy data and store it in the map database.
[0053] Regarding information extraction, it includes:
[0054] S11. Define the question data corresponding to the information to be extracted; for example: Where can I apply for a science and technology award? What are the categories of science and technology awards?
[0055] S12. Search the graph database based on the problem data to obtain matching policy information graphs and suspected demand information.
[0056] It is understandable that by querying the policy information map based on questions, a dependency tree such as "where" to "application" and then to "science and technology awards" can be formed, and a question map can be further generated. Therefore, not only can the corresponding policy information map be located, but also the answer can be further obtained based on the question, and the answers after integration are suspected demand information.
[0057] S13. Call the policy text in the policy database based on the policy information map, and locate it in the policy text based on the suspected demand information.
[0058] The above positioning is to search in the corresponding policy text according to each answer in the suspected demand information, and determine its position according to the search result.
[0059] S14. Extract the located words and words and the data that follows them (here refers to the numerical value, such as: R&D expenses account for 20% of 20%), logical expressions, and establish rules based on preset labels to form rule labels.
[0060] For example: Policy text information: Nomination for the Youth Science and Technology Innovation Award of Province X, unqualified content in the formal review: 1. The candidate is not a Chinese citizen; 2. The candidate's work unit is not an institution affiliated with Province X; 3. No representative papers or monographs published in China have been submitted (technical research and development categories do not need to submit papers and monographs, but if submitted, at least one paper must be published in China). One of the tags extracted from this information can be: {Province X} and {Youth Science and Technology Innovation Award} and {Formal Review} and {Unqualified} and {Not Submitted} and {{Paper} or {Monograph}}.
[0061] Among them, the rule tags include tag names (such as the above-mentioned "Youth Science and Technology Innovation Award"), tag logical operators (such as the above-mentioned "and", "or") and tag values (such as the above-mentioned display tags), and the rule tags form a rule system based on the logical relationship corresponding to the logical expression. For example, all the rule tags corresponding to the "Youth Science and Technology Innovation Award Nomination" under the policy text form the nomination rule system of the "Youth Science and Technology Innovation Award" according to the logical expression.
[0062] According to the above settings, this method can use the natural language analysis model to realize the extraction of policy rule labels, so that when government departments provide policy query services to enterprises, they can accurately match enterprises that can obtain policy subsidies, providing convenience for enterprises or governments.
[0063] In one embodiment of the method, the above-mentioned establishment of the map further includes:
[0064] Identify the policy information graphs based on the same policy text; and,
[0065] The policy information map is classified according to the policy text, the publisher / unit, and the contact person / unit displayed in the policy information map to obtain the policy catalog of each competent caliber and the corresponding map.
[0066] It is understandable that for policy texts, matters corresponding to different competent calibers are identifiable. For example, the policy texts of the Bureau of Economics and Information Technology and the Bureau of Culture and Tourism are obviously different. Classification according to the publisher / unit and the contact person / unit can facilitate the management of the corresponding map by the staff, which facilitates the implementation of subsequent content.
[0067] In this method, the policy texts in the policy database are classified and stored according to the policy catalogs of each competent authority. The aforementioned extracted and located words and phrases and the data and logical expressions that follow them also include:
[0068] Each policy catalogue with different calibers is matched with a corresponding vocabulary; it can be understood that the vocabulary includes phrases and words belonging to the phrases; and according to the above, different calibers have different management contents and the policy texts are quite different, so the vocabulary is matched separately to improve the efficiency and accuracy of information extraction.
[0069] At the same time, in this method, the vocabulary is not fixed, but:
[0070] When the duplicate checking results of the new policy text meet the preset vocabulary update standards, the new policy text will be used as a sample;
[0071] Compare the word library, extract the new words in the sample, and search the dictionary to obtain the definition;
[0072] Add the newly appeared words to the corresponding phrases in the vocabulary according to the definitions.
[0073] According to the above, the information extraction of the label established by this method not only matches the vocabulary according to the caliber corresponding to the policy document, but the vocabulary can also be updated with the newly released policy text.
[0074] Furthermore, the above-mentioned extracted and located characters, words and the data and logical expressions following them also include:
[0075] Locate the new words in the sample and determine whether the words that represent the meaning appear in the first n and / or last m characters. If so, then:
[0076] Compare the content following the interpretation with the interpretation of the phrase in the corresponding vocabulary. If they are the same, add the phrase; if they are different, generate a new phrase based on the content following the interpretation.
[0077] According to the above settings, this method can also actively add or update phrases in the policy document to explain certain newly defined "words".
[0078] In order to reduce a series of errors caused by AI analysis errors, this method:
[0079] The updated information and / or newly added information in the phrases in the vocabulary is sent to the designated user terminal via the Internet or SMS, and user feedback is received. Based on the user feedback, it is determined whether to implement the vocabulary update / addition action; or, if no feedback is received within the preset feedback deadline T, the vocabulary update / addition action is implemented.
[0080] That is, the user can manually stop or continue the word library update or addition.
[0081] In summary, this method can realize the extraction of policy rule labels, so that when government departments provide policy query services to enterprises, they can accurately match enterprises that can obtain policy subsidies, providing convenience for enterprises or governments.
[0082] The embodiment of the present application also discloses an enterprise service policy label extraction system based on natural language analysis.
[0083] The enterprise service policy label extraction system based on natural language analysis includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any of the above-mentioned enterprise service policy label extraction methods based on natural language analysis.
[0084] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.
Claims
1. A method for extracting enterprise service policy labels based on natural language analysis, characterized in that: Including policy text collection, map building and information extraction; The policy text collection includes: Obtaining policy texts and performing duplicate checks based on existing records in the policy database; and, If the duplicate check result meets the preset storage standards, it will be stored in the policy database; The establishing of the map comprises: Define the information graph; and, Generate corresponding policy information graphs based on the stored policy data and store them in the graph database; The information extraction includes: Define the problem data corresponding to the information to be extracted; Search the graph database based on the problem data to obtain matching policy information graphs and suspected demand information; Calling policy texts in the policy database based on the policy information graph and locating them in the policy texts based on suspected demand information; and, Extract the located words and phrases and the data and logical expressions that follow them, and establish rules according to preset labels to form rule labels; wherein the rule labels include label names, label logical operators and label values, and the rule labels are based on the logical relationships corresponding to the logical expressions to form a rule system; The establishing of the map comprises: Identify the policy information graphs based on the same policy text; and, Classify the policy information map according to the policy text, the publisher / unit, and the contact person / unit displayed in the policy information map to obtain the policy catalogs of each competent caliber and the corresponding maps; classify and store the policy texts in the policy database according to the policy catalogs of each competent caliber; The extracted and located characters, words and the data and logical expressions following them include: Match the corresponding vocabulary with each competent policy catalog; When the duplicate checking results of the new policy text meet the preset vocabulary update standards, the new policy text will be used as a sample; Compare the word library, extract the new words in the sample, and search the dictionary to obtain the definition; Add the newly appeared words to the corresponding phrases in the vocabulary according to the meaning; The extracted and located characters, words and the data and logical expressions following them also include: Locate the new words in the sample and determine whether the words that represent the meaning appear in the first n and / or last m characters. If so, then: Compare the content following the interpretation with the interpretation of the phrase in the corresponding vocabulary, if they are the same, add the phrase; if they are not the same, generate a new phrase according to the content following the interpretation; Send the updated information and / or newly added information in the phrases in the vocabulary to the designated user terminal, receive user feedback, and determine whether to implement the vocabulary update / addition action based on the user feedback; or, if no feedback is received within the preset feedback deadline T, implement the vocabulary update / addition action.
2. The method for extracting enterprise service policy labels based on natural language analysis according to claim 1 is characterized by: The duplicate checking process includes processing the policy text based on a locally sensitive hashing algorithm, comparing the results, and calculating the repetition rate.
3. The method for extracting enterprise service policy labels based on natural language analysis according to claim 2 is characterized in that: The duplicate checking process includes: before checking the content of the policy text for duplicates, first check the name / number and release time of the policy text for duplicates. If the result is 100% identical, stop checking the content for duplicates; if the result is not 100% identical, check the content for duplicates.
4. An enterprise service policy label extraction system based on natural language analysis, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes the enterprise service policy label extraction method based on natural language analysis as claimed in any one of claims 1 to 3.
Citation Information
Cited By
A dynamic image-based enterprise management method and management platform
CN121352630B
