Intelligent classification and archiving system for social security electronic records integrating RPA and NLP

Abstract

This invention discloses an intelligent classification and archiving system for social security electronic records that integrates RPA and NLP, belonging to the field of computer digital information processing and natural language parsing technology. It includes extracting the titles of archived files from the archive database to generate a basic rule set; using NLP to generate current text vector data and calculating similarity based on the basic rules to output a first classification label; performing a comparison based on associated metadata to output a second classification label; comparing the differences between the first and second labels to generate label conflict records and extracting the target classification label; and appending conflict records to trigger an update of the rule set. This invention employs a multi-source feature conflict arbitration and rule self-evolution mechanism, which can effectively solve the problem of system flow interruption caused by semantic feature disagreements, improving the fault tolerance rate of complex unstructured data processing and the accuracy of intelligent archiving.

Description

Technical Field

[0001] This invention relates to the fields of computer digital information processing and natural language parsing technology, and in particular to an intelligent classification and archiving system for social security electronic records that integrates RPA and NLP. Background Technology

[0002] Electronic data processing technology plays a crucial role in modern digital transformation, especially in the automated acquisition, feature classification, and archiving of massive amounts of electronic archives. Utilizing computer programs to extract semantic features and perform logical scheduling on unstructured text data is a core application scenario for improving the efficiency of large-scale data processing.

[0003] In related technologies, Chinese invention patent application CN114579719A discloses an electronic document processing method, device, and storage medium combining RPA and AI. The method automatically obtains electronic documents to be delivered from an electronic delivery system; extracts text from the target electronic document in the electronic documents to be delivered to obtain baseline mailing information; extracts the target mailing information from the mailing page in the electronic delivery system based on artificial intelligence technology; compares the target mailing information with the baseline mailing information; if the comparison passes, the method automatically performs a sending operation on the electronic documents to be delivered based on artificial intelligence technology and an RPA robot.

[0004] Regarding the aforementioned technologies, simply performing direct equivalence or consistency comparisons on two sets of extracted single-dimensional text information is insufficient when dealing with electronic archive data with complex semantic information and diverse business flow paths. If the data features extracted from multiple sources show discrepancies or ambiguities, the process is often directly judged as a failure and interrupted. This results in low fault tolerance and accuracy for automated classification and archiving of electronic documents, and a significant limitation on its applicability. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an intelligent classification and archiving system for social security electronic records that integrates RPA and NLP. It employs a multi-source feature conflict arbitration and rule self-evolution mechanism, which can effectively resolve the system flow interruption problem caused by semantic feature disagreements, thereby improving the fault tolerance rate of complex unstructured data processing and the accuracy of intelligent archiving.

[0006] The above objectives can be achieved through the following approach: A smart classification and archiving system for social security electronic records integrating RPA and NLP includes a basic entry generation module, used to read archived electronic files and archiving classification tags from the social security record database, extract title text from the archived electronic files, and generate a set of basic rule entries; a text vector construction module, used to read the current electronic file to be processed and associated metadata through RPA components, and perform stop word removal and phrase segmentation through NLP to generate current text vector data; a rule matching and classification module, used to calculate word vector similarity between the current text vector data and the basic rule entry set based on the basic rule entry set, extract archiving classification tags from the basic rule entry with the highest similarity, and generate a first classification tag; metadata The cross-classification module is used to extract process node identifiers and business type fields based on the associated metadata, perform cross-addressing comparison, and output a second classification label; the label conflict arbitration module is used to perform difference comparison between the first classification label and the second classification label, generate label conflict records, calculate the historical adoption frequency of each classification label in the social security archive database, and extract the classification label with the highest adoption frequency as the target classification label; the archiving and updating module is used to drive the RPA component to write the currently pending electronic file into the electronic archive management system according to the target classification label, and append the target classification label and the label conflict record to the social security archive database, triggering the update of the basic rule entry set.

[0007] Optionally, the basic entry generation module includes: a layout parsing and title extraction unit, used to parse the document layout data based on the archived electronic file, extract the character region sequence at the first line position and perform special symbol cleaning to generate standard title text data; and a business term extraction and rule mapping unit, used to extract core business terms from the standard title text data, perform key-value pair hash mapping association between the core business terms and the archived category tags, and construct a basic rule entry set.

[0008] Optionally, the text vector construction module includes: a node capture and character stream extraction unit, used to drive the RPA component to traverse the document object nodes of the business system to capture associated metadata, and read the current electronic file to be processed to extract the underlying unstructured character stream; and a semantic word segmentation and weight concatenation unit, used to perform stop word removal and phrase segmentation on the underlying unstructured character stream through NLP, extract the word embedding weights of the phrase segmentation results, and concatenate them to generate the current text vector data.

[0009] Optionally, the rule matching and classification module includes: a cosine ranging and matrix generation unit, used to extract a reference text vector based on the basic rule entry set, perform cosine similarity calculation on the current text vector data and the reference text vector, and generate a similarity distribution matrix; and a maximum value mapping and label output unit, used to traverse the similarity distribution matrix to extract the first similarity node, extract the archived classification label of the first similarity node in the basic rule entry set, and generate a first classification label.

[0010] Optionally, the metadata cross-classification module includes: a structure parsing and field separation unit, used to perform key-value pair deserialization parsing on the associated metadata, and separate and extract process node identifier and business type field; and a matrix addressing and label output unit, used to use the process node identifier and the business type field as joint query coordinates, and perform cross-addressing comparison to extract archive classification labels and generate second classification labels.

[0011] Optionally, the system further includes: using the business type field as the main directory name and the target category tag as the subdirectory name, performing string concatenation to generate a physical mounting storage path.

[0012] Optionally, the tag conflict arbitration module includes: a difference analysis and log encapsulation unit, used to perform character sequence difference comparison between the first category tag and the second category tag; if there is a sequence difference, the first category tag, the second category tag, and the current electronic file to be processed are bound together to encapsulate and generate a tag conflict record; and a historical parsing and descending order truncation unit, used to parse the historical archive logs of each category tag extracted from the social security archive database, count the adoption count value of the historical archive logs and perform descending order sorting, and truncate the first category tag of the sequence to output as the target category tag.

[0013] Optionally, the archiving and updating module includes: an interface call and standard input unit, used to convert the currently pending electronic file into a standard format file, drive the RPA component to call the archiving interface of the electronic file management system, and write the standard format file and the target classification tag to the storage node of the physical attached storage path; and an incremental insertion and rule overwrite unit, used to insert the tag conflict record into the log data table of the social security file database, establish a new mapping relationship between the currently pending electronic file and the target classification tag, and overwrite the basic rule entry set.

[0014] Based on the same inventive concept, this invention also provides a method for intelligent classification and archiving of social security electronic records that integrates RPA and NLP. The method includes: reading archived electronic files and archiving classification tags from a social security record database; extracting title text from the archived electronic files to generate a set of basic rule entries; reading the current electronic file to be processed and associated metadata using an RPA component, and performing stop word removal and phrase segmentation using NLP to generate current text vector data; calculating word vector similarity between the current text vector data and the set of basic rule entries based on the set of basic rule entries; extracting archiving classification tags from the basic rule entries with the highest similarity; and generating... The system generates a first category label; based on the associated metadata extraction process node identifier and business type field, it performs cross-addressing comparison and outputs a second category label; it performs a difference comparison between the first category label and the second category label, generates label conflict records, calculates the historical adoption frequency of each category label in the social security archive database, and extracts the category label with the highest adoption frequency as the target category label; it drives the RPA component to write the currently pending electronic document into the electronic archive management system according to the target category label, and appends the target category label and the label conflict record to the social security archive database, triggering the update of the basic rule entry set.

[0015] Compared with the prior art, the present invention has the following advantages: 1. By performing low-level cross-comparison between the semantic tags of the text output by NLP and the business flow tags extracted by RPA, and by making dynamic decisions based on the extreme values ​​of historical adoption frequency in the database, the blind spots of a single algorithm are effectively overcome, ensuring high robustness even when facing extremely complex heterogeneous conflicting data. 2. The final arbitration decision is linked to the tag conflict record. By extracting the conflict deviation data, the dictionary key-value reconstruction of the basic rule entry set is triggered in reverse. This closed-loop write-back mechanism can continuously self-correct using abnormal data in daily processing, giving the data processing architecture a strong dynamic growth and adaptive iteration capability; 3. After arbitration, the RPA component is directly driven to generate a physical storage path based on the target classification label, and to perform automated addressing, disk writing, and anti-counterfeiting traceability. This completely eliminates the manual work interruption between "logical data classification" and "physical disk storage," achieving an exponential leap in the efficiency of electronic document archiving.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a framework diagram of the intelligent classification and archiving system for social security electronic records that integrates RPA and NLP, according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of the intelligent classification and archiving system for social security electronic records that integrates RPA and NLP, according to an embodiment of the present invention.

[0020] Figure 3 This is a lollipop diagram showing the word embedding weight distribution of different attribute lexical units in an embodiment of the present invention.

[0021] Figure 4 This is a scatter plot showing the mapping relationship between sequence difference dispersion and rule overwriting confidence in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Reference Figure 1 One embodiment of the present invention proposes an intelligent classification and archiving system for social security electronic records that integrates RPA and NLP. It adopts a multi-source feature conflict arbitration and rule self-evolution mechanism, which can effectively solve the problem of system flow interruption caused by semantic feature divergence, and improve the fault tolerance rate of complex unstructured data processing and the accuracy of intelligent archiving.

[0024] like Figure 2 As shown, the system in this embodiment specifically includes: The basic entry generation module is used to read archived electronic files and archived category tags from the social security archive database, extract title text from the archived electronic files, and generate a set of basic rule entries. Optionally, the basic entry generation module includes: The layout parsing and title extraction unit is used to parse the document layout data based on the archived electronic file, extract the character region sequence at the first line position and perform special symbol cleaning to generate standard title text data; The system accesses the social security archive database to retrieve archived electronic documents generated during historical business processing. These archived electronic documents include electronic format documents and accompanying digital images. Low-level parsing is performed on the archived electronic documents to obtain document layout data. This document layout data records the spatial geometric distribution coordinates of text blocks and image blocks within the document. The layout parsing and title extraction unit calculates the vertical coordinate components of each text line on the page, identifying the text line with the smallest vertical coordinate value as the first line of the document. The character region sequence at the first line position is extracted. This character region sequence contains mixed information such as the original business name, date, and number. Special character cleaning is performed on the character region sequence to remove punctuation marks, special graphic characters, and non-textual information, generating standard title text data. This standard title text data constitutes the basic text source for business attribute identification.

[0025] For example, the layout parsing and title extraction unit reads an archived electronic file from the social security archive database, whose archive category label is "Urban Employee Basic Pension Insurance". The file is parsed to obtain document layout data, identifying a text block within a range of 0.5 cm to 1.2 cm vertically at the top of the page. The character sequence within this range is extracted, containing "

[2023] Urban Employee Basic Pension Insurance Enrollment Registration Form #001". The layout parsing and title extraction unit performs special symbol cleaning, removing square brackets, hash symbols, and numerical codes, generating the standard title text data content as "Urban Employee Basic Pension Insurance Enrollment Registration Form".

[0026] The business term extraction and rule mapping unit is used to extract core business terms from the standard title text data, perform key-value hash mapping to associate the core business terms with the archive category tags, and construct a set of basic rule entries.

[0027] The business term extraction and rule mapping unit receives standard title text data and performs word segmentation on it to extract core business terms. To filter out highly recognizable business feature words from complex phrases, the business term extraction and rule mapping unit performs filtering by calculating word weight scores. The calculation logic for the word weight scores follows the formula: , Among them, letters Represents the lexical weight score of the target word segment in the current standard title text data; letters This represents the frequency of the target word in the standard title text data; letters The total number of words contained in the standard title text data; letters represents the total number of document samples stored in the social security file database; the letter represents the number of document samples containing the target word segment in the social security file database. The formula calculates the feature intensity through the product operation of the word frequency and the inverse document frequency. Based on the measured distribution of 50,000 historical file data, the word segments with a lexical weight score greater than 0.15 are determined as core business words. The value of 0.15 is obtained through statistical significance analysis and is used to eliminate the generalized words with high frequency and low information. The core business words and the filing classification labels are associated by key-value pair hash mapping. Key-value pair hash mapping means using a hash function to convert the text into a memory index address. The hash calculation logic satisfies the following formula: , [[ID=VII]] [[ID=VIII]]where the letter [[ID=IX]] [[ID=X]]represents the generated hash address value; the letter [[ID=XI]] [[ID=XII]]represents the total number of characters contained in the core business word; the letter [[ID=XIII]] [[ID=XIV]]represents the Unicode integer value corresponding to the [[ID=XV]] [[ID=XVI]]th character in the core business word; the letter [[ID=XVII]] [[ID=XVIII]]represents the prime number of the preset hash bucket capacity. Based on the memory space allocation strategy, the prime number [[ID=XIX]] [[ID=XX]]takes the value of 9973 and is used to map business words of different lengths to a fixed address range. Summarize all the associated data pairs to construct a set of basic rule entries as the logical benchmark for filing comparison. [[ID=XXI]] [[ID=XXII]]

[0028] [[ID=XXIII]]For example, the business word extraction and rule mapping unit obtains the standard title text data as "Work Injury Recognition Application Form". After word segmentation of this text, we get "work injury", "recognition", and "application form". Given that the total number of word segments [[ID=XXIV]] [[ID=XXV]]of this text is 3, and the total number of document samples [[ID=XXVI]] [[ID=XXVII]]in the database is 50,000. Taking the word segment "work injury" as an example, its occurrence frequency [[ID=XXVIII]] [[ID=XXIX]]is 1. The number of document samples containing the word segment "work injury" in the database [[ID=XXX]] [[ID=XXXI]]is 800. Substitute into the lexical weight score calculation formula, and the calculated lexical weight score [[ID=XXXII]] [[ID=XXXIII]]is 0.59. Since 0.59 is greater than the threshold 0.15, "work injury" is extracted as a core business word. The total number of characters [[ID=XXXIV]] [[ID=XXXV]]contained in the core business word "work injury" is 2. The Unicode value of the character "工" [[ID=XXXVI]] [[ID=XXXVII]]is 24037, and the Unicode value of the character "伤" [[ID=XXXVIII]] [[ID=XXXIX]]is 20260. Substitute into the hash calculation formula and perform a modulo operation with the prime number 9973 to generate the hash address value ​The hash address is 1245. The business term extraction and rule mapping unit performs key-value pair mapping between the hash address value 1245 and the corresponding archive category tag "work injury insurance", stores it in memory space, and completes the construction of this entry in the basic rule entry set.

[0029] The text vector construction module is used to read the current electronic file to be processed and associated metadata through the RPA component, and to perform stop word removal and phrase segmentation through NLP to generate the current text vector data. Optionally, the text vector construction module includes: The node capture and character stream extraction unit is used to drive the RPA component to traverse the document object nodes of the business system to capture associated metadata and read the current electronic file to be processed to extract the underlying unstructured character stream. The node capture and character stream extraction unit initiates the robotic process automation (RPA) component. The RPA component enters the software interface of the social security administration system by simulating manual operation logic. The node capture and character stream extraction unit drives the RPA component to traverse the document object nodes of the system. Document object nodes are data structure units in the front-end interface of the system that carry file indexes and business attribute tags. The node capture and character stream extraction unit extracts structured information describing business attributes from the document object nodes, generating associated metadata. This associated metadata includes the business transaction number, processing time, and handler identifier. The currently pending electronic file represents a digital document circulating in the current stage of the system that has not yet entered a fixed archiving state. The node capture and character stream extraction unit further reads the currently pending electronic file, extracts its raw text content, and generates an underlying unstructured character stream. This underlying unstructured character stream is a sequence of raw strings extracted directly from the document's binary stream without semantic filtering.

[0030] For example, the node capture and character stream extraction unit drives the robotic process automation component into the social security business system. On the benefits review interface, it locates the document object node and captures the associated metadata content as "Business Number: SB2024041501". The node capture and character stream extraction unit reads the currently pending electronic file, which is a PDF format pension application form. The node capture and character stream extraction unit extracts the underlying unstructured character stream.

[0031] The semantic word segmentation and weight concatenation unit is used to perform stop word removal and phrase segmentation on the underlying unstructured character stream through NLP, extract the word embedding weights of the phrase segmentation results, and concatenate them to generate the current text vector data.

[0032] The semantic segmentation and weight concatenation unit receives the underlying unstructured character stream and inputs it into the Natural Language Processing (NLP) model. The unit then invokes NLP to perform stop word removal on the unstructured character stream, filtering out auxiliary words and prepositions that have no actual business meaning. After filtering, phrase segmentation is performed, dividing the text into multiple independent lexical units according to semantic logical boundaries, generating phrase segmentation results. To accurately represent the business feature strength of each lexical unit, the semantic segmentation and weight concatenation unit needs to calculate the word embedding weight of each lexical unit in the phrase segmentation results. The formula for calculating the word embedding weight is as follows: , in, The first in the phrase segmentation results Word embedding weights for each lexical unit. This represents the frequency of occurrence of the lexical unit in the underlying unstructured character stream. This represents the total number of words contained in the underlying unstructured character stream. This represents the length of characters contained in the vocabulary unit. This represents the maximum character length limit for a given term. Based on the statistical distribution pattern of the lengths of 30,000 social security business terms, [the following is a partial translation of the original text:] The value is set to 8 to cover frequently used technical terms such as "pension insurance participation". All variables are count values ​​or constants, and the calculation results... Maintaining a dimensionless, pure number state. After calculating the word embedding weights of each lexical unit, the semantic segmentation and weight concatenation unit performs weight concatenation and dimension transformation. To ensure that the vectors generated from texts of different lengths have a fixed spatial dimension, the semantic segmentation and weight concatenation unit uses a sine mapping mechanism for dimension mapping calculation. The formula for dimension transformation is as follows: , in, Represents the first element in the current text vector data. The numerical values ​​of each dimension. This represents the total number of lexical units contained in the phrase segmentation result. This represents the vector dimension index currently being computed. Based on the semantic space complexity test of social security file classification, the total vector dimension is set to 128 dimensions, i.e. The value range is from 1 to 128. By iteratively calculating and sequentially arranging the values ​​of all dimensions, semantic word segmentation and weight concatenation units are concatenated to generate the current text vector data. The current text vector data is a mathematical tensor representing the core semantics of the file to be archived. For example... Figure 3 As shown, the constructed word embedding weights Based on word frequency With character length The joint exponential gain control, a fundamental mathematical mapping mechanism, effectively highlights low-frequency but long-tail core business terms such as "labor capacity assessment," while suppressing high-frequency but extremely low-information-entropy short-character interference words, thus ensuring the semantic fidelity of text vector data.

[0033] For example, the semantic word segmentation and weighted concatenation unit obtains the underlying unstructured character stream content as "The applicant meets the retirement conditions and applies for pension filing". Stop word filtering is performed to remove "meets" and "conditions". Phrase segmentation is then performed to obtain the lexical units "applicant", "retirement", "application", and "pension". The total number of lexical units contained in the phrase segmentation result is... The total vocabulary size of the underlying unstructured character stream is 4. The frequency is 7. Taking the vocabulary unit "retirement" as an example, its occurrence frequency is... The length is 1, and the character length is 1. Set to 2. Set the maximum word length limit. The value is 8. Substituting into the word embedding weight calculation formula, the word embedding weight of "retirement" is calculated. The value is 0.18. The semantic segmentation and weight concatenation unit calculates the values ​​for all dimensions, indexed by the first dimension, i.e., the vector dimension. Taking 1 as an example, the embedding weights of each word are multiplied by their corresponding sine mapping values ​​and summed to calculate the value of the first dimension. The value is 0.72. All calculations are performed by traversing dimensions 1 to 128, and the results are concatenated to generate the current text vector data.

[0034] The rule matching and classification module is used to calculate the word vector similarity between the current text vector data and the set of basic rule entries based on the set of basic rule entries, extract the archive classification label from the basic rule entry with the highest similarity, and generate the first classification label. Optionally, the rule matching and classification module includes: The cosine ranging and matrix generation unit is used to extract a reference text vector based on the set of basic rule entries, perform cosine similarity calculation on the current text vector data and the reference text vector, and generate a similarity distribution matrix; The cosine ranging and matrix generation unit reads the set of basic rule entries and extracts a baseline text vector from it. The set of basic rule entries is a reference dictionary containing historical core business features and classification mapping relationships. The baseline text vector is a baseline mathematical tensor formed by multi-dimensional numerical transformation of core business terms in the set of basic rule entries. After obtaining the baseline text vector, the cosine ranging and matrix generation unit performs cosine similarity calculation on the current text vector data and the baseline text vector. The current text vector data is a multi-dimensional numerical tensor representing the semantic features of the current electronic document to be processed. Cosine similarity calculation refers to a low-level mathematical approximation process that quantitatively evaluates the directional consistency and semantic similarity between two tensors by measuring the cosine value of the angle between them in the multi-dimensional feature space. The cosine similarity calculation formula is as follows: , in, This represents the current text vector data and its index number. The cosine similarity value between the baseline text vectors. Represents the first element in the current text vector data. Quantified values ​​for the location dimension. The index number represents The first in the baseline text vector Quantified values ​​for the location dimension. The total dimension of the vector representing the tensor. Based on the sine dimension mapping mechanism, the total dimension of the vector is... The value is strictly limited to 128 to ensure absolute alignment of dimensions in high-dimensional space operations. The physical dimensions of the numerator and denominator cancel each other out during division, ensuring the calculated cosine similarity value remains a dimensionless pure number with a stable range between -1 and +1. The cosine ranging and matrix generation unit traverses the set of basic rule entries to calculate the cosine similarity of all tensor pairs, and then combines all the results sequentially according to the index order of the basic rules to generate a similarity distribution matrix. The similarity distribution matrix is ​​a structured numerical array that records the matching strength between the current electronic document to be processed and each historical business rule.

[0035] For example, the cosine ranging and matrix generation unit reads the set of basic rule entries and extracts the baseline text vectors with indices 1 to 3. The current text vector data contains values ​​with dimension indices 1 to 128. The cosine ranging and matrix generation unit performs a cosine similarity calculation on the current text vector data and the baseline text vector with index number 1. The sum of the products of the corresponding values ​​in the 128 dimensions of the current text vector data and the baseline text vector with index number 1, i.e., the numerator of the formula, is 85.5. The square root of the sum of the squares of the values ​​in each dimension of the current text vector data is 10.0. The square root of the sum of the squares of the values ​​in each dimension of the baseline text vector with index number 1 is 9.5. Substituting these values ​​into the cosine similarity calculation formula, the cosine similarity value between the current text vector data and the baseline text vector with index number 1 is mathematically calculated. The value is 0.90. Using the same ranging calculation logic, the cosine similarity value with the baseline text vector at index number 2 is calculated. The cosine similarity score is 0.45, which is the same as the score of the baseline text vector with index number 3. The value is 0.21. The cosine ranging and matrix generation unit concatenates these three dimensionless values ​​in index order to generate a similarity distribution matrix. The matrix content is recorded as a one-dimensional numerical distribution: 0.90, 0.45, and 0.21.

[0036] The maximum value mapping and label output unit is used to traverse the similarity distribution matrix to extract the first similarity node, extract the archived classification label of the first similarity node in the basic rule entry set, and generate the first classification label.

[0037] The maximum value mapping and label output unit receives the similarity distribution matrix. It performs a global traversal operation on the similarity distribution matrix, comparing the numerical values ​​of each element in the numerical array to extract the first similarity node. The first similarity node represents the element node in the similarity distribution matrix whose value exhibits the largest global feature; its physical meaning is the target rule location point in the multidimensional semantic space that has the highest matching degree with the current electronic document to be processed. After locking the first similarity node, the maximum value mapping and label output unit reads the address index information of this node and performs reverse memory addressing based on this address index information to extract the archive classification label corresponding to the first similarity node in the basic rule entry set. The archive classification label is a standard string that clearly indicates the business archive category to which the historical digital voucher belongs. The maximum value mapping and label output unit uses the extracted archive classification label as the recognition output of the text stage to generate the first classification label. The first classification label represents the preliminary business classification tendency deduced purely based on natural language content semantic matching.

[0038] For example, the maximum value mapping and label output unit receives a similarity distribution matrix with contents of 0.90, 0.45, and 0.21. It performs a global traversal and size comparison calculation, identifying the value 0.90 as the global maximum in the matrix. The maximum value mapping and label output unit extracts the value 0.90 as the first similarity node and locks its permutation index in the matrix as 1. Based on index 1, it performs hash key-value addressing on the set of basic rule entries, accurately extracting the archived category label bound to index 1, which is "Work Injury Determination". The maximum value mapping and label output unit encapsulates and outputs "Work Injury Determination", generating the first category label for cross-modal cross-comparison and label conflict arbitration scheduling.

[0039] The metadata cross-classification module is used to extract process node identifiers and business type fields based on the associated metadata, perform cross-addressing comparison, and output a second classification label. Optionally, the metadata cross-classification module includes: The structure parsing and field separation unit is used to perform key-value pair deserialization parsing on the associated metadata, and separate and extract the process node identifier and business type fields; The structure parsing and field separation unit acquires the retrieved associated metadata. This associated metadata is an attribute dataset encapsulated in standardized data exchange formats such as JSON or XML. The structure parsing and field separation unit performs key-value pair deserialization on the associated metadata. Key-value pair deserialization is the process of breaking down and mapping a continuous sequence of strings formatted according to a predetermined protocol into independent data structures of "keys" and "values" that can be directly addressed and read from computer memory. After completing the key-value pair deserialization, the structure parsing and field separation unit locates specific "keys" to separate and extract the corresponding data "values." Specifically, it extracts values ​​representing the progress of business processes to generate process node identifiers; and extracts values ​​representing macro-level business segment categories to generate business type fields. Process node identifiers are unique numerical codes indicating the specific stage of the currently pending electronic document within the social security approval engine. Business type fields are standard English strings identifying the top-level business pool to which the currently pending electronic document belongs.

[0040] For example, the structure parsing and field separation unit receives associated metadata, the underlying format of which is "{"Node_ID":"04","Biz_Type":"Pension"}". The structure parsing and field separation unit performs key-value pair deserialization parsing, mapping the string to an in-memory object. It locates the key name "Node_ID", separates and extracts its corresponding value, generating the process node identifier content as "04". It locates the key name "Biz_Type", separates and extracts its corresponding value, generating the business type field content as "Pension".

[0041] The matrix addressing and label output unit is used to use the process node identifier and the business type field as joint query coordinates, perform cross-addressing comparison to extract archive classification labels and generate second classification labels.

[0042] The matrix addressing and label output unit receives the separately generated process node identifier and business type field. The process node identifier and business type field are used as the joint query coordinates. The joint query coordinates are two-dimensional index vectors composed of two independent business feature dimensions, used for data location in the standard business mapping matrix. The matrix addressing and label output unit performs cross-addressing comparison based on these joint query coordinates. Cross-addressing comparison refers to the calculation process of mapping the two dimensions of the joint query coordinates to row and column indices in a pre-defined standard business mapping matrix, and then locating and reading the value in the target storage cell through the intersection of the rows and columns. To ensure that text-type identifiers can be converted into matrix-recognizable numeric indices, the decimal index value of the joint query coordinates needs to be calculated. The formula for calculating the decimal index value is as follows: , in, This represents the decimal index value on which the cross-addressing comparison is based. The total length of characters contained in the process node identifier. The first node in the process node identifier from right to left The Unicode integer value corresponding to each position character. This represents the preset column offset base. Based on the maximum capacity test of the social security business matrix, [the following will be implemented]. Set to 100000000 to ensure that the row index segment and the column index segment never overlap in numerical space. The total length of characters contained in the business type field. The business type field, from right to left, is the first... The formula assigns different weights to the Unicode characters at each position and combines this with an offset base, resulting in a decimal index value that is dimensionless and absolutely unique. Based on the calculated decimal index value, the matrix addressing and label output unit extracts the archive category label from the corresponding storage cell in the standard business mapping matrix. The matrix addressing and label output unit outputs the extracted archive category label as the result of the business logic dimension, generating a second category label.

[0043] For example, the matrix addressing and label output unit receives the process node identifier "04" and the business type field "Pension". These are used as coordinates for a joint query to prepare for calculating the decimal index value. The total character length of the process node identifier... The value is 2. The Unicode integer value of the character "4" from right to left. 52; the Unicode integer value of the character "0". The column offset base is 48. The value is 100000000. This represents the total character length of the business type field. The value is 7. The Unicode integer value of the character "n" from right to left. 110; the Unicode integer value of the character "o". 111; the Unicode integer value of the character "i". 105; the Unicode integer value of the character "s". 115; the Unicode integer value of the character "n". 110; the Unicode integer value of the character "e". 101; the Unicode integer value of the character "P" The value is 80. Substitute this value into the formula to calculate the decimal index: the first half... The second half Adding the two together yields the decimal index value. The value is 53281326720. The matrix addressing and label output unit locates itself in the standard business mapping matrix based on the value 53281326720, extracts the archive category label content of the intersection cell as "Enterprise Retirement Approval Category", and outputs this content to generate the second category label.

[0044] Optionally, the system further includes: The business type field is used as the main directory name and the target category tag is used as the subdirectory name. String concatenation is performed to generate the physical mounting storage path.

[0045] The system first extracts the separated business type field. Simultaneously, it receives the target category label, representing the final authoritative ruling, output from the label conflict arbitration module. The business type field is then designated as the main directory name. The main directory name is the name of the first-level data storage folder, constructed within the server operating system's underlying file system, defining the top-level business hierarchy. The target category label is then designated as the subdirectory name. The subdirectory name is the name of the second-level data storage folder, nested within the main directory and established based on subdivided file categories. Following the path separation rules of the underlying file system, the main directory name and subdirectory name are concatenated to generate the physical mounting storage path. The physical mounting storage path is the absolute path string instructing the computer operating system to precisely write the currently processed electronic file to the underlying storage disk sector.

[0046] For example, the extracted business type field is "Pension". The received output target category tag is "Enterprise Retirement Approval Class". "Pension" is used as the first-level main directory name, and "Enterprise Retirement Approval Class" is used as the nested subdirectory name. Following the slash-separated rules of Linux systems, a string concatenation operation is performed to generate a physical mounting storage path of " / Pension / Enterprise Retirement Approval Class / ", which can be directly invoked for archiving and disk persistence.

[0047] The tag conflict arbitration module is used to perform a difference comparison between the first category tag and the second category tag, generate a tag conflict record, calculate the historical adoption frequency of each category tag in the social security archive database, and extract the category tag with the highest adoption frequency as the target category tag. Optionally, the tag conflict arbitration module includes: The difference analysis and log encapsulation unit is used to perform character sequence difference comparison between the first category label and the second category label. If there is a sequence difference, the first category label, the second category label and the current electronic file to be processed are bound to the data and encapsulated to generate a label conflict record. The difference analysis and log encapsulation unit receives the generated first category label and the generated second category label. The first category label represents the preliminary result based on natural language semantic matching. The second category label represents the logical result based on the underlying metadata structure attribute mapping. The difference analysis and log encapsulation unit performs character sequence difference comparison on the first and second category labels. Character sequence difference comparison refers to the process of quantitatively assessing whether two strings are absolutely identical at the underlying memory level by comparing their character encoding sequences bit by bit. To digitize the comparison result, the sequence difference dispersion needs to be calculated. The formula for calculating the sequence difference dispersion is as follows: , in, This represents the calculated sequence difference dispersion. This represents the total length of characters contained in the first category label. This represents the total length of characters contained in the second category label. This represents the step size of the character position index during the traversal and comparison process. Represents the first category label in the first position The decimal value of the Unicode corresponding to the position character. Represents the second category label in the first position The decimal value of the Unicode corresponding to the position character. The formula generates a dimensionless pure number for the sequence difference dispersion by aggregating the length deviation and the absolute value of the coding difference at each corresponding position. If the sequence difference dispersion is 0, it indicates that the two tags are exactly the same; if the sequence difference dispersion is greater than 0, the trigger condition "sequence difference exists" is determined. When there is a sequence difference, the difference analysis and log encapsulation unit extracts the first classification tag and the second classification tag, and synchronously extracts the unique identification code of the currently to-be-processed electronic file as the identifier of the currently to-be-processed electronic file. The three data elements are subjected to data binding. Data binding means allocating continuous pointer references for heterogeneous data in the computer memory to form a unified structured object body. After the binding is completed, it is encapsulated and packed to generate a tag conflict record. The tag conflict record serves as the underlying structured log describing the divergence of business characteristics for evolutionary learning.

[0048] Exemplarily, the difference analysis and log encapsulation unit receives the first classification tag as "work injury recognition class" and the second classification tag as "work injury investigation class". Perform character sequence difference comparison. The total length of the characters in the first classification tag is 5, and the total length of the characters in the second classification tag is 5, and the absolute value of the length difference is 0. Start comparing the absolute value of the difference in the decimal value of Unicode one by one according to the position step For the character at the 3rd position, the first classification tag is "recognition", and the Unicode value [[ID=!!]] is 35748; the second classification tag is "investigation", and the Unicode value is 35843, and the absolute difference is 95. For the character at the 4th position, the first classification tag is "determination", and the Unicode value is 23450; the second classification tag is "investigation", and the Unicode value is 26597, and the absolute difference is 3147. The absolute differences at the remaining positions are 0. Substitute into the formula for calculation and summation to obtain the sequence difference dispersion is 3242. Since 3242 is greater than 0, it is determined that there is a sequence difference. The difference analysis and log encapsulation unit extracts the identifier "DOC_202405" of the currently to-be-processed electronic file, and binds the "work injury recognition class", "work injury investigation class" and "DOC_202405" to the memory pointer, and encapsulates and generates a tag conflict record in the format of a JSON object.

[0049] The historical parsing and descending truncation unit is used to parse the historical archived logs of each classification tag extracted from the social security file database, count the adopted count values of the historical archived logs and perform a descending sort, and truncate the classification tag at the first position of the sequence and output it as the target classification tag.

[0050] The historical parsing and descending order truncation unit is activated after being triggered by the difference analysis and log encapsulation unit. It accesses the social security archive database and, based on the string names of the first and second category labels that caused the conflict, parses and retrieves the underlying data tables to extract the associated historical archive logs. The historical archive logs are a collection of data rows recording the authoritative classification results of similar archives in the past. The historical parsing and descending order truncation unit calculates the adoption count value for each category label that caused the conflict based on the extracted historical archive logs. The adoption count value refers to the cumulative frequency of a specific category label as the final archive category in historical authoritative results. After calculating the adoption count value for each category label, a sorting algorithm is called to sort the adoption count values ​​in descending order. Descending order sorting refers to rearranging the row and column indices of data groups containing the "category label-adoption count value" mapping relationship from high to low based on the numerical value. The historical parsing and descending order truncation unit extracts the category label at the top of the rearranged sequence and outputs it as the arbitration decision, establishing it as the target category label. The target classification label represents the sole standard used to drive subsequent physical disk placement, based on the highest confidence level obtained through historical data voting and deduction.

[0051] For example, the historical parsing and descending order truncation unit accesses the social security archive database and extracts historical archived logs containing "Work Injury Determination" and "Work Injury Investigation" categories. The unit performs statistical calculations. Traversing the extracted data tables, it finds that a total of 8500 log entries are ultimately confirmed as "Work Injury Determination," thus calculating an adoption count of 8500 for "Work Injury Determination." Similarly, the adoption count for "Work Injury Investigation" is calculated to be 1500. The unit then sorts this data set ("Work Injury Determination": 8500, "Work Injury Investigation": 1500) in descending order, placing the largest value item at the top. The unit then extracts the top-ranked category label, i.e., "Work Injury Determination" with an adoption count of 8500, and outputs it as the final target category label.

[0052] The archiving and updating module is used to drive the RPA component to write the currently pending electronic file into the electronic file management system according to the target classification label, and to append the target classification label and the label conflict record to the social security file database, thereby triggering the update of the basic rule entry set.

[0053] Optionally, the archiving and updating module includes: The interface call and standard data entry unit is used to convert the currently pending electronic file into a standard format file, drive the RPA component to call the archiving interface of the electronic document management system, and write the standard format file and the target classification tag location to the storage node of the physical attached storage path; The interface call and standard data entry unit retrieve the currently pending electronic file. This file is typically a easily tamper-proof streaming document such as Word or Excel. The interface call and standard data entry unit then invoke the format conversion engine to convert the file into a standard format file. A standard format file is a PDF or OFD digital document that conforms to the standards of the State Archives Administration, has a fixed layout, and is tamper-proof. To verify that no underlying data corruption occurred during the format conversion process, the interface call and standard data entry unit calculate the bitstream integrity check code of the standard format file. The formula for calculating the bitstream integrity check code is as follows: , in, This represents the calculated bitstream integrity check code. This represents the total number of bytes contained in a standard format file. Represents the physical position index in the file byte stream, with values ​​ranging from 1 to... . The standard format file is located at the first position. The decimal integer value corresponding to the byte of the physical location index. This represents a preset check prime number. Based on a 100MB-level single-file hash collision prevention test, it will... The value is set to 999983 to ensure a balance between computational power consumption and collision prevention capabilities during the verification process. The formula deeply binds byte values ​​to physical locations and performs modulo operations with large prime numbers, ensuring that any shift or flip of a single byte will drastically change the calculation result. The generated bitstream integrity check code is a dimensionless pure number. After the interface call verifies the bitstream integrity check code against the standard data entry unit, it drives the Robotic Process Automation (RPA) component. The RPA component calls the application programming interface (API) exposed by the electronic records management system, namely the archiving interface. Through this archiving interface, the RPA component performs low-level disk addressing based on the generated physical storage path, locating and writing the data stream of the standard format file and the metadata of the target classification label to the storage node mapped by that physical storage path. The storage node is the smallest data-carrying unit at the bottom layer of the electronic records management system.

[0054] For example, the interface call and standard input unit convert a Word document into a standard PDF document. The bitstream integrity checksum of the PDF file is calculated. The total number of bytes in the PDF file is also calculated. The value is 200,000. First physical location index. A decimal value of 1 byte 37; Second physical location index A decimal value of 2 bytes The value is 80. Perform cumulative calculation, i.e. Traversing all bytes yields a cumulative product of 1250000000. (Check for prime numbers.) The value is 999983. Substituting this into the calculation formula, we calculate the remainder when 1250000000 is divided by 999983, thus obtaining the bitstream integrity check code. The value is 133648. After verifying that the check code is correct, the interface call and the standard data entry unit drive the Robotic Process Automation (RPA) component to call the archiving interface of the electronic records management system. Through the archiving interface, the RPA component locates and writes the PDF file and the target category label "Work Injury Determination" to the disk storage node corresponding to the physical storage path " / Injury / Work Injury Determination / ".

[0055] The incremental insertion and rule overwriting unit is used to insert the tag conflict record into the log data table of the social security file database, establish a new mapping relationship between the current electronic file to be processed and the target classification tag, and overwrite the set of basic rule entries.

[0056] The incremental insertion and rule overwriting unit starts after detecting a successful write response message from the archiving interface. It extracts the generated tag conflict record and inserts it into a specific structured log data table in the social security archive database. This insertion operation forcibly establishes a new mapping relationship between the current electronic file to be processed and the target classification label in the underlying data architecture. This new mapping relationship represents the latest and most authoritative text classification benchmark learned after historical data voting arbitration. Based on this new mapping relationship, the incremental insertion and rule overwriting unit reverses the set of basic rule entries. Overwriting refers to finding the old erroneous rule in the set of basic rule entries with the highest similarity to the underlying semantic features of the current electronic file to be processed, and replacing its original erroneous archive classification label with the target classification label. To prevent occasional dirty data from damaging the set of basic rule entries, the rule overwriting confidence score must be calculated before overwriting. The formula for calculating the rule overwriting confidence score is as follows: , in, This represents the calculated rule overwrite confidence level. This represents the adoption count value of the target category label in the arbitration module. This represents the total number of historical archived log entries in the social security archive database that participated in statistical analysis. This represents the calculated sequence difference dispersion. This represents the maximum tolerance constant. Based on 5000 real-world network stress test data, The value is set to 10000. All variables are count values ​​or discrete characteristic values. The calculated rule overwrite confidence score remains a dimensionless pure number, and its value is strictly limited to between 0 and 1. The incremental insertion and rule overwrite unit determines whether the rule overwrite confidence score exceeds a preset learning threshold. Based on the low fault tolerance requirements of core social security business, the learning threshold is rigidly set to 0.65. Only when the rule overwrite confidence score exceeds 0.65 will the incremental insertion and rule overwrite unit perform the overwrite action on the basic rule entry set. For example... Figure 4 As shown, as the dispersion of sequence differences gradually increases, the confidence of rule overwriting exhibits a strict exponential decay trend. When the confidence is lower than the learning threshold, the system will automatically block the overwriting action, thereby effectively protecting the set of basic rule entries from contamination by extreme abnormal data.

[0057] For example, the incremental insertion and rule overriding unit detects that the PDF file has been successfully stored in the node. Tag conflict records in JSON object format are extracted and inserted into the log data table of the social security archive database. A mapping relationship is established between the text features of the PDF file and the newly added target classification tag "Work Injury Determination". Before overriding the set of basic rule entries, the rule overriding confidence is calculated. The adoption count value of the target classification tag is known. The total number of historical archived log entries included in the statistics is 8500. The value is 10000. Sequence difference dispersion. The maximum tolerance difference constant is 3242. The value is 10000. Substitute this into the calculation formula: Natural constant of The power of, i.e. It is approximately 0.7231. Calculate. The rule overwrite confidence level is obtained. The confidence level is approximately 0.6146. Since the calculated rule overwrite confidence level of 0.6146 is less than the system's preset hard learning threshold of 0.65, the incremental insertion and rule overwrite unit refuses to perform the overwrite action, retains the current label conflict record as a log backup, and abandons updating the basic rule entry set, thereby protecting the basic rule entry set from contamination by low-confidence data.

[0058] Based on the same inventive concept, this invention also provides a method for intelligent classification and archiving of social security electronic records that integrates RPA and NLP, the method comprising: By reading archived electronic files and archived category tags from the social security archive database, the title text of the archived electronic files is extracted to generate a set of basic rule entries; The RPA component reads the current electronic file to be processed and its associated metadata, and the NLP component performs stop word removal and phrase segmentation to generate the current text vector data. Based on the set of basic rule entries, word vector similarity is calculated between the current text vector data and the set of basic rule entries. Archive classification labels are extracted from the basic rule entries with the highest similarity to generate the first classification label. Based on the node identifier and business type field of the associated metadata extraction process, cross-addressing comparison is performed to output the second category label; A difference comparison is performed between the first category label and the second category label to generate a label conflict record. The historical adoption frequency of each category label is calculated in the social security archive database, and the category label with the highest adoption frequency is extracted as the target category label. The RPA component is driven to write the currently pending electronic document into the electronic file management system according to the target classification label, and to append the target classification label and the label conflict record to the social security file database, triggering the update of the basic rule entry set.

[0059] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. This is a conventional technical method. The electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0060] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A smart classification and archiving system for social security electronic records integrating RPA and NLP, characterized in that, The system includes: The basic entry generation module is used to read archived electronic files and archived category tags from the social security archive database, extract title text from the archived electronic files, and generate a set of basic rule entries. The text vector construction module is used to read the current electronic file to be processed and associated metadata through the RPA component, and to perform stop word removal and phrase segmentation through NLP to generate the current text vector data. The rule matching and classification module is used to calculate the word vector similarity between the current text vector data and the set of basic rule entries based on the set of basic rule entries, extract the archive classification label from the basic rule entry with the highest similarity, and generate the first classification label. The metadata cross-classification module is used to extract process node identifiers and business type fields based on the associated metadata, perform cross-addressing comparison, and output a second classification label. The tag conflict arbitration module is used to perform a difference comparison between the first category tag and the second category tag, generate a tag conflict record, calculate the historical adoption frequency of each category tag in the social security archive database, and extract the category tag with the highest adoption frequency as the target category tag. The archiving and updating module is used to drive the RPA component to write the currently pending electronic file into the electronic file management system according to the target classification label, and to append the target classification label and the label conflict record to the social security file database, thereby triggering the update of the basic rule entry set.

2. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 1, characterized in that, The basic entry generation module includes: The layout parsing and title extraction unit is used to parse the document layout data based on the archived electronic file, extract the character region sequence at the first line position and perform special symbol cleaning to generate standard title text data; The business term extraction and rule mapping unit is used to extract core business terms from the standard title text data, perform key-value hash mapping to associate the core business terms with the archive category tags, and construct a set of basic rule entries.

3. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 1, characterized in that, The text vector construction module includes: The node capture and character stream extraction unit is used to drive the RPA component to traverse the document object nodes of the business system to capture associated metadata and read the current electronic file to be processed to extract the underlying unstructured character stream. The semantic word segmentation and weight concatenation unit is used to perform stop word removal and phrase segmentation on the underlying unstructured character stream through NLP, extract the word embedding weights of the phrase segmentation results, and concatenate them to generate the current text vector data.

4. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 1, characterized in that, The rule matching and classification module includes: The cosine ranging and matrix generation unit is used to extract a reference text vector based on the set of basic rule entries, perform cosine similarity calculation on the current text vector data and the reference text vector, and generate a similarity distribution matrix; The maximum value mapping and label output unit is used to traverse the similarity distribution matrix to extract the first similarity node, extract the archived classification label of the first similarity node in the basic rule entry set, and generate the first classification label.

5. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 1, characterized in that, The metadata cross-classification module includes: The structure parsing and field separation unit is used to perform key-value pair deserialization parsing on the associated metadata, and separate and extract the process node identifier and business type fields; The matrix addressing and label output unit is used to use the process node identifier and the business type field as joint query coordinates, perform cross-addressing comparison to extract archive classification labels and generate second classification labels.

6. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 1, characterized in that, The system also includes: The business type field is used as the main directory name and the target category tag is used as the subdirectory name. String concatenation is performed to generate the physical mounting storage path.

7. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 1, characterized in that, The label conflict arbitration module includes: The difference analysis and log encapsulation unit is used to perform character sequence difference comparison between the first category label and the second category label. If there is a sequence difference, the first category label, the second category label and the current electronic file to be processed are bound to the data and encapsulated to generate a label conflict record. The historical parsing and descending order truncation unit is used to parse the historical archive logs of each category label from the social security archive database, count the adoption count value of the historical archive logs and sort them in descending order, and truncate the category label at the beginning of the sequence as the target category label.

8. The intelligent classification and archiving system for social security electronic records integrating RPA and NLP as described in claim 6, characterized in that, The archiving and updating module includes: The interface call and standard data entry unit is used to convert the currently pending electronic file into a standard format file, drive the RPA component to call the archiving interface of the electronic document management system, and write the standard format file and the target classification tag location to the storage node of the physical attached storage path; The incremental insertion and rule overwriting unit is used to insert the tag conflict record into the log data table of the social security file database, establish a new mapping relationship between the current electronic file to be processed and the target classification tag, and overwrite the set of basic rule entries.

Images