Intelligent retrieval system for electronic information of archives

Abstract

The invention relates to the technical field of archive retrieval, in particular to an archive electronic information intelligent retrieval system which comprises a hierarchical extraction module, a frequency comparison module, a character alignment module, a similarity labeling module and an index integration module. According to the method, the archive text and the keyword character string are extracted in a layered mode, the reference ratio is calculated based on the total number of characters and the preset threshold value, the content priority is accurately divided, character distribution and index positions are analyzed in combination with query of the first character, positioning distribution information is generated, and the proportion value is formed by the consistent number of the characters one by one of the matched positions and the length of the query character string. The method comprises the following steps of: screening fragments meeting consistency requirements, generating a similar block labeling list in combination with an index range and a similarity label, integrally realizing layered screening, proportion comparison, character-by-character consistency analysis and similarity labeling, and promoting the linkage of unstructured archive data retrieval from coarse screening to fine matching; and the retrieval accuracy, the matching fragment positioning performance and the result structured expression are obviously improved, and multi-dimensional quick acquisition is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of archive retrieval, and in particular to an intelligent retrieval system for archive electronic information. Background Art

[0002] The field of archive retrieval technology encompasses the archiving, management, and retrieval of electronic information. Its core content includes the digitization, classified storage, indexing, retrieval, data matching, and information extraction of electronic archives. It involves digitizing and storing archival information in paper or electronic form in a database, and using retrieval strategies and query mechanisms to quickly locate and obtain the required archival information, thereby achieving efficient management and utilization of large-scale electronic information resources. This requires not only efficient classification and indexing of archives, but also the establishment of structured or unstructured data organization to support multi-dimensional, multi-conditional retrieval requirements, ensuring that users can easily access target archival content by entering keywords, query conditions, and other methods.

[0003] The intelligent retrieval system for electronic archival information refers to an intelligent retrieval solution for unstructured archival data. It addresses technical issues such as content identification, keyword extraction, query condition matching, and retrieval strategy design for electronic archival information, primarily through multi-condition query methods based on database indexes, text analysis and keyword segmentation processing, document-oriented similarity matching methods, and content classification and tag matching technologies. By establishing an index system for unstructured data and combining keyword weights with condition matching relationships, it supports the rapid location of target information in large-scale electronic archives and returns matching results, encompassing the complete technical process from text data parsing and indexing to multi-condition query execution.

[0004] The existing archival information retrieval process mainly relies on the retrieval process based on conditional query and overall text comparison, lacks a content priority division mechanism and positioning analysis based on character distribution, and there is often high-frequency and inefficient traversal during retrieval, which easily causes key clues to be submerged in a large amount of irrelevant information. In addition, information matching in existing technologies is mostly based on overall similarity, and does not involve character-by-character proportional alignment and consistent block screening of matching fragments, resulting in difficulty in accurately identifying highly relevant areas at the fragment level. The retrieval results lack fragment positioning and similarity labeling. During use, users often need to repeatedly compare and confirm the target content due to vague results and lack of distribution prompts. For example, when responding to large-scale, multi-topic archive queries, existing technologies lack priority screening and fragment distribution analysis, which are prone to problems such as key content submersion, coarse matching granularity, and unclear positioning of related fragments. This directly increases the burden of subsequent manual screening, reduces the efficiency of information utilization, and restricts the rapid response and efficient extraction capabilities of archival information. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent retrieval system for electronic archive information.

[0006] In order to achieve the above-mentioned object, the present invention adopts the following technical solution: an intelligent retrieval system for electronic archive information comprises: The hierarchical extraction module obtains archive text data and keyword strings, counts the total number of archive text characters and sets a preset threshold and calculates a benchmark ratio, records the full-text index position where the keyword first appears and accumulates the total number of first appearances. If the accumulated total number of times is greater than the benchmark ratio, the content corresponding to the keyword is classified into the archive high-priority classification content set; The frequency comparison module obtains the first character of the query string based on the high-priority classified content set of the archive, calculates the number of characters in the layered content that are equal to the first character of the query, determines the first character location distribution, and generates layered first character location distribution information; The character alignment module calculates the number of character-by-character consistency at the start of each matching position based on the hierarchical first character positioning distribution information, forms a ratio value with the query string length, selects position segments whose ratio values ​​meet the consistency threshold requirement, and obtains a hierarchical character-by-character consistent block set; The similarity marking module marks similarity label information based on the hierarchical character-by-character consistent block set and the start and end index positions of the corresponding segments to generate a hierarchical similar block marking list.

[0007] As a further solution of the present invention, the high-priority classified content set of the archive includes priority tags, keyword coverage, and content directionality; the hierarchical first-character positioning distribution information includes character distribution density, position index list, and matching start mark; the hierarchical character-by-character consistent block set includes consistency ratio value, character alignment fragment, and block number; the hierarchical similar block annotation list includes similarity tags, block correspondence, and index range information.

[0008] As a further solution of the present invention, the layered extraction module includes: The text statistics submodule extracts the text content and counts the total number of characters based on the archive text data and keyword strings, sets the character count benchmark value as the text character threshold, calculates the ratio of the total number of text characters to the text character threshold, and obtains the benchmark ratio value; The keyword locating submodule extracts the full-text index position where each keyword first appears based on the text content and the keyword string using a sequential scanning method, and accumulates the number of first appearances of all keywords to obtain the cumulative number of first appearances of the keyword; The priority classification submodule compares the first cumulative number of occurrences of the keyword with the benchmark ratio value to determine whether the former is greater than the latter. If so, the content corresponding to the keyword is classified into a classification set, and a high priority judgment value is calculated at the same time. Combined with the comparison results, a high priority classification content set of the archive is established.

[0009] As a further solution of the present invention, the frequency comparison module includes: The first character extraction submodule reads each content segment based on the high-priority classified content set of the archive and extracts the first character of the query string to be retrieved, collects the hierarchical content structure in the text to be analyzed, forms a hierarchical data structure, and obtains a mapping sequence between the first character and the hierarchical structure; The character matching submodule performs character-by-character matching of the query first character extracted from the mapping sequence of the first character and the hierarchical structure with each hierarchical character sequence, counts the frequency of occurrence of the first character, determines the number of characters respectively and counts the total frequency, and calculates and outputs a weighted distribution weight value of the first character; The positioning generation submodule constructs the character index distribution trajectory of the corresponding character in the hierarchical text based on the weight value of the first character weighted distribution and the character index position in the corresponding hierarchical content, combined with all matching points identified in the character matching process, and generates hierarchical first character positioning distribution information.

[0010] As a further solution of the present invention, the character alignment module includes: The position matching submodule extracts a character segment of the same length as the query string from the text based on the hierarchical first character location distribution information and the first character location as the starting point, establishes a pairing relationship between the starting index and the paragraph content, and obtains a query index mapping table; The consistency calculation submodule compares the query fragments recorded in the query index mapping table with the query string character by character, and counts the number of characters in each character fragment that are consistent with the query string. Calculate and output the character consistency ratio value; The block screening submodule compares the character consistency ratio value with the consistency judgment threshold, performs screening operations on all segments, constructs all continuous paragraph blocks that meet the consistency comparison requirements, and integrates to establish a hierarchical character-by-character consistent block set.

[0011] As a further solution of the present invention, the similarity marking module includes: The index mapping submodule extracts the corresponding text segment content based on the hierarchical character-by-character consistent block set, identifies the hierarchical structure label and paragraph number to which the block belongs, and indexes and binds the query character with the block character one by one to obtain an index mapping set; The label generation submodule performs similarity labeling on each set of character mapping segments based on the index mapping set and the number of character matches and the length of the continuous matching segments to obtain a block similarity assessment result; The list output submodule classifies and archives the blocks according to the hierarchical structure based on the block similarity assessment results, combined with the label determination results and the hierarchical label and block number information provided in the index mapping, and establishes a hierarchical similar block annotation list.

[0012] As a further embodiment of the present invention, the system further comprises: The index integration module integrates the similar block annotations and the hierarchical content based on the hierarchical similar block annotation list and the high-priority classified content set of the archive, establishes an overall archive index, and generates an intelligent retrieval index structure for archive electronic information; The intelligent retrieval index structure of archival electronic information includes an integrated index table, hierarchical index nodes, and a global retrieval mapping.

[0013] As a further solution of the present invention, the index integration module includes: The block integration submodule reads the block structure annotation content based on the hierarchical similar block annotation list, checks whether there is overlap between the blocks, and merges them into continuous blocks if overlap is found, and obtains a hierarchical integration block index table; The content fusion submodule performs information level identification processing according to the hierarchical integration block index table, combines the original content of each layer in the archive high-priority classification content set, and performs information level identification processing according to the similarity tags marked on the blocks, adds index anchor points and identification fields, and obtains hierarchical fusion text structure blocks; The index generation submodule performs structural integration of the archives according to the anchor structure and index fields in the hierarchical fusion text structure block, performs field classification processing, sets the query entry of the keyword dimension and the label dimension of the hierarchical level, and generates an intelligent retrieval index structure for the archive electronic information.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, the archival text and keyword character strings are extracted in layers, the benchmark ratio is calculated based on the total number of characters and the preset threshold, the content priority is accurately divided, the character distribution and index position are analyzed in combination with the first character of the query, the positioning distribution information is generated, the number of character-by-character consistency in the matching position and the query string length form a proportional value, the fragments that meet the consistency requirements are screened, and the similar block annotation list is generated in combination with the index range and the similarity label. The overall hierarchical screening, proportional comparison, character-by-character consistency analysis, and similarity annotation are realized, which promotes the connection of unstructured archival data retrieval from coarse screening to fine matching, significantly improves the retrieval accuracy, matching fragment positioning and result structured expression, and realizes multi-dimensional rapid acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the hierarchical extraction module of the present invention; Figure 3 This is a flow chart of the frequency comparison module of the present invention; Figure 4 This is a flow chart of the character alignment module of the present invention; Figure 5 This is a flow chart of the similarity marking module of the present invention; Figure 6 This is a flow chart of the index integration module of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0018] See also Figure 1 , an intelligent retrieval system for electronic archive information includes: The hierarchical extraction module obtains archival text data and keyword strings, counts the total number of archival text characters and a preset threshold (a pre-set character number benchmark value, a standard critical value for classifying text priority), calculates the ratio of the total number of text characters to the preset threshold as a benchmark ratio (a quantitative proportional relationship between the total number of text characters and the preset threshold, used to determine keyword density), records the full-text index position where the keyword first appears based on a sequential scanning operation and accumulates the total number of first appearances, compares the cumulative total number of times with the benchmark ratio, and if the cumulative total number of times is greater than the benchmark ratio, classifies the content corresponding to the keyword into a high-priority classified content set in the archive (a priority text set classified based on the comparison result of the keyword appearance frequency with the benchmark ratio); The frequency comparison module obtains the first character of the query string based on the high-priority classified content set of the archive, calculates the number of characters in the layered content that are equal to the first character of the query, combines the total number of character sequences in the layered content, determines the first character location distribution based on the layered content index position, and generates the layered first character location distribution information; The character alignment module obtains the index of each first character matching position and the query string length based on the hierarchical first character positioning distribution information, calculates the number of character-by-character consistency at the start of each matching position, forms a ratio with the query string length, selects the position segments whose ratios meet the consistency threshold requirements, and obtains a set of hierarchical character-by-character consistent blocks; The similarity annotation module establishes a correspondence between the blocks and the query string based on the hierarchical character-by-character consistent block set, combined with the start and end index positions of the corresponding fragments, annotates the similarity label information, and generates a hierarchical similar block annotation list; The index integration module is based on the hierarchical similar block annotation list and combines the high-priority classified content set of the archives to integrate the similar block annotations with the hierarchical content, establish an overall index of the archives, and generate an intelligent retrieval index structure for the archives' electronic information.

[0019] The collection of high-priority classified content of archives includes priority labels, keyword coverage, and content directionality; the hierarchical first-character positioning distribution information includes character distribution density, position index list, and matching start mark; the hierarchical character-by-character consistent block collection includes consistency ratio value, character alignment fragment, and block number; the hierarchical similar block annotation list includes similarity labels, block correspondence, and index range information; the intelligent retrieval index structure of archive electronic information includes integrated index table, hierarchical index node, and global retrieval mapping.

[0020] See also Figure 2 , the hierarchical extraction module includes: The text statistics submodule extracts text content and counts the total number of characters based on the archive text data and keyword strings, sets the character count benchmark value as the text character threshold, calculates the ratio of the total number of text characters to the text character threshold, and obtains the benchmark ratio value; Based on the archival text data and keyword strings, the target document text data needs to be extracted from the archive. For example, the total number of characters in a company's annual report text is 15,400 characters, and the keyword strings are ["revenue", "net profit", "research and development"]. The text character threshold is set to 5,000 characters. The threshold is set based on the average number of characters in the past 10 company annual reports. The calculated benchmark ratio value is the ratio of the total number of characters to the threshold, specifically: 15,400 ÷ 5,000 = 3.08, indicating that the scale of this text has a higher information density than the benchmark. The total number of text characters is calculated by counting the number of characters in each paragraph of the document paragraph by paragraph and summing them up. For example, the number of characters in a paragraph is 1,500, 1,800, 2,500, 3,000, 4,600, and 1,000 characters respectively, and the sum of each item is 15,400 characters. The text character threshold is used to calculate the total number of characters. The source is the average value obtained by counting the characters of 10 recent archival materials and eliminating the maximum and minimum values. For example, the number of characters in the 10 texts is 5200, 5100, 5050, 4900, 4800, 5000, 5300, 4700, 4950, and 5000. After removing 5300 and 4700, there are 8 samples left, and their average is (5200+5100+5050+4900+4800+5000+4950+5000) ÷ 8 = 5000 characters. The rationality of setting the character threshold refers to the general requirement of no less than 4800 characters in the industry annual report writing specifications. Combined with the average value, 5000 characters is set as the benchmark threshold, which will be used as a reference for judging the distribution density of keywords in subsequent analysis. The ratio is 3.08 in this sample text, and the final benchmark ratio value is 3.08.

[0021] The keyword positioning submodule uses a sequential scanning method to extract the full-text index position where each keyword first appears based on the text content and keyword string, and accumulates the number of times all keywords first appear to obtain the cumulative number of first appearances of the keyword; Combining the text content with the keyword string, the text needs to be segmented at the character level, and sequential matching is performed on each keyword to extract the full-text character index position where it first appears. For example, the keyword "revenue" first appears at the 120th character, "net profit" first appears at the 245th character, and "research and development" first appears at the 360th character. "Gross profit margin" as a noise word first appears at the 490th character. The number of times all keywords appear is counted one by one. "Revenue" appears 3 times in the text, with index positions 120, 900, and 1310. "Net profit" appears 2 times, at 245 and 1300. "research and development" appears 4 times, with corresponding positions 360, 710, 1150, and 1490. "Gross profit margin" appears once, with an index of 490. The above keywords and their positions and frequencies are counted as follows to form index and frequency data, as shown in the following table: Table 1 Keyword statistics ; As shown in Table 1, the first character index and corresponding frequency of all keywords are counted. By adding the total number of occurrences, that is, 3+2+4+1=10 times, the cumulative number of first occurrences of the keyword in the text is obtained as 10.

[0022] The priority classification submodule compares the first cumulative number of occurrences of the keyword with the benchmark ratio value to determine whether the former is greater than the latter. If so, the content corresponding to the keyword is classified into the classification set, and the formula is used: ; Calculate high priority judgment value , combined with the comparison results to establish a high-priority classification content set of archives, among which, Indicates the cumulative number of times a keyword appears for the first time. The average of the character index values ​​where the keyword first appears. Indicates the number of different keywords in the keyword string, represents the base ratio value, Represents the average of the total length of all keywords, Indicates the character position where the i-th keyword first appears, Indicates the character spacing between the i-th keyword. Indicates the sum operation of all keyword index data. Indicates absolute value operation; The judgment is made by comparing the cumulative number of first appearances of the keyword with the benchmark ratio value. If the former is greater than the latter, the content corresponding to the keyword is classified into the archive high-priority classification content set. It is known that the cumulative number of appearances is 10 and the benchmark ratio is 3.08, which meets the condition range (the keyword frequency exceeds the benchmark value multiple). The following formula is introduced to quantify the judgment process.

[0023] in: , the cumulative number of first appearances of the keyword; : The average value of the first character index position of the keyword; : The number of different keywords in the keyword string; : The ratio of the total number of characters in the text to the character threshold; : The average value of the total frequency of keywords; ; After substituting the actual values, the calculation is as follows: ; The results show that when the text character density reaches the benchmark of 3.08, the structural aggregation calculation value of the keyword density has reached 400,000, which is greater than the preset interval critical value (such as setting the upper limit of the interval standard to 200,000), and is therefore classified as a high-priority classification content set of archives.

[0024] The high priority judgment value is a composite quantitative indicator used to measure the relationship between the distribution density and position concentration of keywords in a text and the overall information carrying capacity of the text. The larger the value, the more frequent the keyword appears and the more concentrated it is in the core area of ​​the text. There is also a significant imbalance between its appearance structure and the overall character density of the text, reflecting that the information weight carried by the keyword in the text content is significantly high, and therefore it is necessary to be extracted and marked preferentially. This value integrates the coupling relationship between the first appearance position of the keyword, the total number of appearances and the text length ratio, and connects the structural characteristics of the keyword with the semantic center of gravity of the text, thereby realizing the identification and priority classification of key content fragments. It is a direct quantitative basis for supporting the construction of high-priority content collections in archives.

[0025] The operational logic of this formula is a comprehensive measure of keyword distribution density, text structure complexity and keyword index dynamics, where the numerator is the cumulative number of keyword appearances. The square root of the square of the keyword index mean Adding, it means introducing quantitative consideration of position distribution concentration based on keyword frequency. The square operation of keyword position reflects its increasing influence as it moves away from the starting position. The total number of keywords is added to stabilize the fluctuation range of the position mean. Further combined with the denominator , that is, the text character ratio The average frequency of keywords The difference reflects the tension between the overall density of the text and the sparseness of keywords. When the difference is close to zero, the formula amplification coefficient tends to increase, increasing the sensitivity to the mismatch between keyword and text density. Subsequently, it is multiplied by the sum of the ratios of the first occurrence position of all keywords to their frequency. This item represents the bias degree and frequency regulation effect of keyword index in distribution. Ultimately, the absolute value operation is used to ensure that all structural outputs are positive numbers to accommodate the possibility of positive and negative disturbances under different text structures. The overall logic is reflected in the following: taking keyword frequency as the main trunk, combining the nonlinear interaction results of structural distribution strength and its tension with text density, the keyword priority is quantitatively judged.

[0026] See also Figure 3 , the frequency comparison module includes: The first character extraction submodule reads each content fragment and extracts the first character of the query string to be retrieved based on the archive's high-priority classified content set. It collects the hierarchical content structure of the text to be analyzed, establishes an index system for each content level and records the character sequence range, establishes the total number of character sequences for each level, forms a hierarchical data structure, and obtains a mapping sequence between the first character and the hierarchical structure. Obtain a collection of high-priority classified content from the archive, extract the first keyword in each text in turn, and extract the first character from the keyword as the query target character. For example, the first character of the query keyword "net profit growth rate" is "net", and this character is used as the subsequent matching object. Then, the hierarchical content of the target archive is extracted in a structured manner. In this example, the text is divided into three logical modules: "corporate strategy", "financial analysis", and "R&D planning". The number of characters in each module is independently counted, and the layers are numbered L1, L2, and L3, with a total number of characters of 820, 1130, and 940, respectively. When counting, the text parsing interface needs to be called to traverse each layer character by character, and special symbols such as separators and line breaks are removed before being counted into the net character count. Based on the character hierarchy, the position and frequency of the first appearance of the first character "net" in each layer are extracted, and its index position in the character list is identified. At the same time, it is recorded whether it appears on an odd-numbered index and whether it is in a character paragraph position adjacent to a numeric character. Finally, a complete data structure is formed and summarized as shown in the following table: Table 2 First character frequency statistics ;

[0027] As shown in Table 2, character sequence statistics and first character index data extraction are completed for all three layers of content. The mapping sequence between the first character and the hierarchical structure obtained will serve as the basic index set for the next comparison process.

[0028] The character matching submodule matches the first character of the query extracted from the mapping sequence of the first character and the hierarchical structure with each hierarchical character sequence character by character, counts the frequency of the first character, determines the number of characters respectively and calculates the total frequency, using the formula: ; Calculate the weighted distribution weight of the output first character , combined with each layer index number mapping table for interpolation archiving, where, Indicates the The number of occurrences of the first character in the query layer, Indicates the index position where the first character of the corresponding layer appears for the first time. is the total number of characters in the corresponding layer, is the number of times the first character of the corresponding layer appears in odd positions, is the number of times the first character of the corresponding layer appears when the adjacent characters are numbers, Indicates the total number of layers, Indicates the sum of all layer data. Indicates absolute value operation; Based on the "net" character occurrence feature information recorded in the mapping sequence between the first character and the hierarchical structure, characters are matched bit by bit in each hierarchical text to identify the "net" character occurrence position and count its occurrence frequency. The character distribution strength value is calculated based on the first occurrence position, the total number of characters, whether the occurrence position is an odd number, and whether it is adjacent to a number. The distribution value is used to judge the density and structural activity of the first character. After collecting the above data, it is inserted into the distribution calculation formula for actual calculation, which is executed according to the following structure: ; Substitute the corresponding values ​​in Table 2 for each layer and perform the expansion calculation: The L1 layer distribution value is: ; The L2 layer distribution value is: ; The L3 layer distribution values ​​are: ; Combine and sum to get the absolute value: ; This value is the comprehensive weighted distribution strength of the first character, which can reflect the distribution density and fluctuation of the "net" characters in the hierarchical text structure. The final weight value of the first character weighted distribution is 27.16.

[0029] The weighted distribution weight value of the first character is a numerical indicator used to measure the distribution density and structural concentration of the first character of the query in multiple hierarchical text contents. This value comprehensively considers the frequency of the first character in each layer, the position of its first appearance, the regularity of its distribution at odd index positions, and the semantic complexity adjacent to numeric characters, and is normalized based on the total number of characters in the layer. Through the quantitative calculation of these structural factors, it reflects whether the first character is clustered, regular, or sparsely distributed in each layer. The larger the value, the more concentrated the character distribution in the structure and the higher the weight. It can be used as a matching strength measure between the spatial position of the character and the hierarchical structure of the text, thereby providing a data basis for subsequent positioning generation and text screening.

[0030] The calculation logic of the formula is based on a comprehensive evaluation of the frequency of occurrence of the first character in different layers and its distribution characteristics, and the weighted quantification of the character distribution weight value is achieved through a structural combination of multiple parameters, where Indicates that the first character is in The number of occurrences in a layer is used to reflect the concentration of the character in the layer, multiplied by This is to introduce the influence of the first appearance position of the first character on the overall structure. The earlier the first appearance, the more likely it is to constitute the central topic in that layer. Therefore, the position index is positively included in the product term, and 1 is added to prevent multiplication by zero. The denominator is used to compress the total number of characters. The weight of , so that the level with a large total number of characters will not be distorted due to the large number of characters, +1 prevents illegal operations caused by the square root of 0, and It represents the difference between the frequency of the first character appearing at an odd-index position (usually representing the beginning of a sentence or a key node in semantics) and its adjacent digital characters. The larger the difference between R and T, the more regular the position of the first character and the lower the noise. The overall formula uses this "frequency × position" as the numerator and the "square root of the character size + structural difference" as the compressed denominator to form a multi-dimensional evaluation. Finally, all hierarchical items are summed up and the absolute value is taken in order to unify the positive and negative directionality and avoid the hierarchical symbols affecting the global weighted judgment results.

[0031] The positioning generation submodule constructs the character index distribution trajectory of the corresponding character in the hierarchical text based on the weighted distribution weight value of the first character and the character index position in the corresponding hierarchical content, combined with all the matching points identified in the character matching process, and converts the trajectory into visual coordinates and segments in hierarchical order to generate hierarchical first character positioning distribution information; According to the above calculation results, the weighted distribution weight value of the first character is 27.16, and according to the specific position index of the "net" character in each layer of content, a character positioning array is constructed. For example, in the L1 layer, "net" appears at indexes 24, 177, 326, etc., in the L2 layer at positions 19, 204, 385, 612, 1022, etc., and in the L3 layer at positions 32, 187, 463, 623, 721, 850, etc. These indexes are represented in an array, and the character distribution trajectory segments are generated in combination with the total character range of each layer. For example, in L1, the "net" character is mainly distributed in the index segment 24–420, in the L2 layer distribution segment is 19–1022, and in the L3 layer is concentrated in 32–850. A pairing mapping is established between each hierarchical positioning segment and the level number, and a hierarchical character space coordinate representation is constructed. Finally, the positioning distribution matrix is ​​output to establish the hierarchical first character positioning distribution information.

[0032] See also Figure 4 , the character alignment module includes: The position matching submodule records the index position of the first character of each layer based on the distribution information of the first character position in each layer, obtains the length of the corresponding query string, and extracts character segments of the same length as the query string from the text, using the position of the first character as the starting point, establishes a pairing relationship between the starting index and the paragraph content, and obtains a query index mapping table; To obtain the hierarchical first character positioning distribution information, it is necessary to perform a comparison structure preparation operation for each identified first character index position. First, the starting position of each matching point is extracted according to the hierarchical index information, and the length of the query keyword string is used as the extraction range of the subsequent comparison segment. For example, in the scenario of searching for the keyword "net profit growth rate", the query string length is 6 characters. Therefore, for any starting index position, 6 consecutive characters need to be extracted after the position to form a comparison segment. If the matching position is 120, the characters corresponding to index positions 120 to 125 are extracted as the first segment. Other starting points such as 245, 463, etc. are processed in sequence to obtain several character segment structures with the same length as the query string. They are identified and classified by numbering, and finally the mapping data structure shown in the following table is generated: Table 3 Character consistency comparison table ; As shown in Table 3, the number, length, starting position and related span of each matching segment have been structured. The constructed index segment mapping can be used for character-level consistency judgment and obtain the query index mapping table.

[0033] The consistency calculation submodule compares the query fragments recorded in the query index mapping table with the query string character by character, and counts the number of characters in each character fragment that are consistent with the query string in turn, using the formula: ; Calculate and output character consistency ratio value ,in, Indicates the number of matching fragments, Indicates the number of characters in the th fragment that match the query string. Indicates the The length of the fragment in characters, Indicates the The global index value corresponding to the starting character position of the fragment, Indicates the The span of the last character of each fragment relative to the beginning of the line, Indicates the sum operation of all matching fragments. Indicates absolute value processing; According to the query index mapping table, a character-by-character consistency analysis is performed on the comparison segments extracted from each record and the query string. Each character position is compared according to the character sequence order, and the number of characters in each comparison segment that are completely consistent with the query string is counted. For example, 4 characters in segment P1 are consistent, 5 characters in segment P2, and 6 characters in segment P3 are completely consistent. The consistency ratio is then calculated based on the comparison length, starting index, and span information from the end of the segment to the beginning of each segment. This calculation process uses the following formula for structural normalization adjustment.

[0034] Substituting the values ​​in Table 3 into the formula, the results are as follows: P1 segment ratio: ; P2 segment ratio: ; P3 segment ratio: ; Taking the average of the above results: ; The resulting character consistency ratio is 0.0421, which reflects the degree of segment consistency in the global character structure.

[0035] The character consistency ratio value indicates the overall similarity between the multiple character fragments extracted and the query string at the character-by-character level in a hierarchical text structure, based on the first character matching point as the starting position. This ratio value comprehensively measures the matching strength of each fragment in character content, the stability of the fragment length structure, the global distribution of the starting index, and the clustering characteristics of local fragments in their layer segments. The larger the value, the higher the degree of consistency of multiple fragments in content, position and structure. It can reflect the comprehensive fit between the matching fragments and the query items at the character level during the entire comparison process, and is used to guide subsequent block screening and information extraction operations.

[0036] The formula uses the following structure for the calculation logic design: The formula as a whole normalizes the performance of each comparison segment in terms of character consistency. First, the parameter Indicates the number of characters in the current segment that are consistent with the query string. It is a core variable that directly reflects the quality of segment matching. The larger its value, the closer the segment is to the target string. Represents the square root of the length of the fragment itself. The purpose is to introduce a nonlinear balance factor between fragments of different lengths to suppress the imbalance in the proportion of overly long fragments in the denominator-dominated structure. and The addition can reflect the weighted combination of matching strength and fragment structure stability, forming an enhanced expression of the local matching ability of the fragment. The denominator is composed of Composition, of which Indicates the index position of the starting character of the fragment in the global text, reflecting its distribution position in the overall text. It represents the span between the last character of the segment and the beginning of the paragraph where it is located. The difference represents the relative stability position of the segment within the layer. The smaller the value, the more concentrated the matching block is and the more compact the structure is. Therefore, it can be used as the denominator to form a structural stability inhibition factor. The overall score structure expresses the balanced ratio between the strength of character consistency and the stability of structural position. Finally, the overall score is used to represent the stability of the whole segment. The sum of the fragments is performed and the average is taken, that is, Expressing the global consistency trend, the outer wrapping absolute value symbol is to avoid imbalance in the average result due to negative index difference. This combination logic constructs fragment-level comparison evaluation indicators through three dimensions: character consistency, structural diffusion, and text index position, to achieve consistent expression of the entire comparison area.

[0037] The block screening submodule compares the character consistency ratio with the consistency judgment threshold, performs screening operations on all segments, constructs all continuous paragraph blocks that meet the consistency comparison requirements, and integrates them to establish a hierarchical set of character-by-character consistent blocks; According to the character consistency ratio value of 0.0421, the system performs screening and judgment based on the set consistency judgment threshold. If the threshold is 0.04, the system will screen out all comparison segments with ratio values ​​greater than or equal to 0.04. According to the ratio data in Table 3, only the P1 segment meets the screening conditions. The system retains the P1 segment and packages and integrates the character segment content corresponding to the segment and its index information in the original hierarchical structure with the hierarchical label, marking it as a content segment that meets the consistency requirements. Finally, a structured paragraph alignment result set is generated, and a hierarchical character-by-character consistent block set is established.

[0038] See also Figure 5 ,Similarity annotation modules include: The index mapping submodule extracts the corresponding text segment content based on the hierarchical character-by-character consistent block set, establishes character-level correspondences based on the original query string, identifies the hierarchical structure label and paragraph number to which the block belongs, extracts the character sequence range of the block in the hierarchical structure, and indexes and binds the query characters to the block characters one by one to obtain the index mapping set; Based on the hierarchical character-by-character consistent block set, each block needs to be accurately located and structurally identified. First, the starting index and ending index corresponding to the block in the original text are extracted, and the hierarchical label and number to which the block belongs are determined. Then, the character sequence is obtained from the original query string. For example, the query string is "net profit growth rate" and its length is 6 characters. These 6 characters are mapped and bound bit by bit with the character content of the same length in the current block in sequence. The actual position of each query character in the text is recorded, and a corresponding chain of mapping structure "query character-text character-index" is formed. An index pairing set with block number, start and end index and hierarchical label as the primary key is established to build a basic data framework for subsequent comparison and judgment, and obtain the index mapping set.

[0039] The label generation submodule is based on the index mapping set and judges according to the number of character matches and the length of continuous matching segments. It performs similarity annotation processing on each set of character mapping segments, records the start and end positions of each block and the relative position of the query character in the paragraph, and obtains the block similarity assessment result; Based on the index mapping set, a character-level comparison is performed to compare each query string character with the target block character segment by segment to see if they are consistent. The number of completely consistent characters in each segment is counted and their continuous length is detected. When the number of consistent characters is greater than half the length of the original query string, and the length of the continuous consistent character segment is greater than 3 characters, the system determines that the block is "highly similar". If the number of consistent characters is lower than this ratio but there are still partial matches, it is classified as "lowly similar". If there is no valid match, it is marked as "dissimilar". At the same time, the similarity label output results are recorded by block number. For example, block P1 matches 5 characters, 4 consecutive characters, meeting the high similarity condition. P2 matches 3 characters, scattered distribution, and is classified as lowly similar. P3 has no match and is dissimilar. The marking results are shown in the following table: Table 4 Similar block annotation list ; As shown in Table 4, each block data has completed character-level consistency marking to obtain the block similarity evaluation results.

[0040] The list output submodule, based on the block similarity assessment results, combines the label determination results with the hierarchical label and block number information provided in the index mapping, classifies and archives them according to the hierarchical structure, meets the requirements of block-by-block marking, consistency recording, and block attribution, and establishes a hierarchical similar block annotation list; Based on the block similarity assessment results and the structural annotation content of each block, all block information is summarized and output as a structural list. The list must include the block number, start index, end index, hierarchical label and similarity label fields. The data must be classified according to the hierarchical label and output in numerical order. The summarized list not only indicates the specific position of each block in the original text, but also aligns the similarity label with the actual index information, clearly identifying the structural status and comparison level of each matching paragraph. The final output structural list is the hierarchical similar block annotation list.

[0041] See also Figure 6 , the index integration module includes: The block integration submodule reads the block structure annotation content based on the hierarchical similar block annotation list, sorts the blocks in the same layer in index order, and checks whether there is overlap between the blocks. If adjacent block indexes overlap, they are merged into continuous blocks, and the new index range and original block number of the merged block are recorded to obtain the hierarchical integration block index table; Based on the hierarchical similar block annotation list, it is necessary to extract the block number, start index, end index and corresponding hierarchical label in each block record, establish a basic mapping structure, and sort the starting positions of all block numbers under the same hierarchical label by reading the block range information marked in the structure field. According to the sorting result, it is determined whether the index intervals between adjacent blocks overlap or are continuous. If the index spacing between two blocks does not exceed 5 characters, they are merged. At the same time, the merged index segments are renumbered, and the original block number mapping relationship is recorded in the structure, thereby constructing a one-to-one fusion mapping table. This table realizes the organization, standardization and repositioning of block segments at the structural level, and finally obtains a hierarchical integrated block index table.

[0042] The content fusion submodule integrates the block index table according to the layered integration, combines the original content of each layer in the archive's high-priority classification content set, and intercepts the character content of the corresponding position in the original text according to the start and end index of each block in the original text. It also performs information hierarchical identification processing according to the similarity tags marked on the blocks, embeds annotation information into the blocks in the same layer, and retains the original hierarchical content structure. The fused text paragraphs must retain the original logical division without changing the order and hierarchy. Index anchor points and identification fields are added to establish a traceable fusion content body and obtain the hierarchical fusion text structure blocks. Read the structure fields identified in the hierarchical integration block index table, combine the text paragraph information of the original archive high-priority classification content set, extract the text content according to the block start and end index, and add block annotation labels to the extracted content. For example, in block P1, the character range extracted is from index 120 to 125, the corresponding text is "net profit growth", and its similarity label is "high similarity", then add "high similarity" as the structural anchor mark before and after the text segment, and complete the replacement embedding in the corresponding layer L1. When multiple blocks are embedded in the same layer, they are written in sequence, and the original paragraph structure of the text is kept unchanged. No additional line breaks or indentations are added, forming the following structure table: Table 5 Archive integration index structure table ; As shown in Table 5, all block contents have been embedded with the original layered content to establish a layered fusion text structure block.

[0043] The index generation submodule integrates archives into a structured manner based on the anchor structure and index fields in the hierarchical fusion text structure block, performs field classification processing, sets the query entry of the keyword dimension and the label dimension of the hierarchical level, and constructs a structured multi-dimensional retrieval structure with a unified entry to generate an intelligent retrieval index structure for archive electronic information; According to the anchor information and block index position written in the hierarchical fusion text structure block, the retrieval index is classified and integrated according to the structural hierarchy, and the "layer number", "fusion block number", "start index", "end index", and "similar label" fields recorded in the table are extracted. The layer number is set as the first-level entry and the fusion block number is set as the second-level jump label. A text retrieval structure that supports jumps, positioning, and classification is constructed. It supports users to jump to the corresponding layer position by keyword during the search, or to expand and view similar fragments by tag clustering, and records the character offset of each index point in the text. This completes the structured, searchable, and jumpable index design and generates an intelligent retrieval index structure for archival electronic information.

[0044] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent retrieval system for electronic archive information, characterized in that: The system comprises: The hierarchical extraction module obtains archive text data and keyword strings, counts the total number of archive text characters and sets a preset threshold and calculates a benchmark ratio, records the full-text index position where the keyword first appears and accumulates the total number of first appearances. If the accumulated total number of times is greater than the benchmark ratio, the content corresponding to the keyword is classified into the archive high-priority classification content set; The frequency comparison module obtains the first character of the query string based on the high-priority classified content set of the archive, calculates the number of characters in the layered content that are equal to the first character of the query, determines the first character location distribution, and generates layered first character location distribution information; The character alignment module calculates the number of character-by-character consistency at the start of each matching position based on the hierarchical first character positioning distribution information, forms a ratio value with the query string length, selects position segments whose ratio values ​​meet the consistency threshold requirement, and obtains a hierarchical character-by-character consistent block set; The similarity marking module marks similarity label information based on the hierarchical character-by-character consistent block set and the start and end index positions of the corresponding segments to generate a hierarchical similar block marking list.

2. The intelligent retrieval system for electronic archive information according to claim 1, characterized in that: The high-priority classified content set of the archive includes priority tags, keyword coverage, and content directionality; the hierarchical first-character positioning distribution information includes character distribution density, position index list, and matching start mark; the hierarchical character-by-character consistent block set includes consistency ratio value, character alignment segment, and block number; the hierarchical similar block annotation list includes similarity tags, block correspondence, and index range information.

3. The intelligent retrieval system for electronic archive information according to claim 1, characterized in that: The layered extraction module includes: The text statistics submodule extracts text content and counts the total number of characters based on the archive text data and keyword strings, sets the character count benchmark value as the text character threshold, calculates the ratio of the total number of text characters to the text character threshold, and obtains the benchmark ratio value; The keyword locating submodule extracts the full-text index position where each keyword first appears based on the text content and the keyword string using a sequential scanning method, and accumulates the number of first appearances of all keywords to obtain the cumulative number of first appearances of the keyword; The priority classification submodule compares the first cumulative number of occurrences of the keyword with the benchmark ratio value to determine whether the former is greater than the latter. If so, the content corresponding to the keyword is classified into a classification set, and a high priority judgment value is calculated at the same time. Combined with the comparison results, a high priority classification content set of the archive is established.

4. The intelligent retrieval system for electronic archive information according to claim 1, characterized in that: The frequency comparison module includes: The first character extraction submodule reads each content fragment and extracts the first character of the query string to be retrieved based on the high-priority classified content set of the archive, collects the hierarchical content structure in the text to be analyzed, forms a hierarchical data structure, and obtains a mapping sequence between the first character and the hierarchical structure; The character matching submodule performs character-by-character matching of the query first character extracted from the mapping sequence between the first character and the hierarchical structure with each hierarchical character sequence, counts the frequency of occurrence of the first character, determines the number of characters respectively and counts the total frequency, and calculates and outputs a weighted distribution weight value of the first character; The positioning generation submodule constructs the character index distribution trajectory of the corresponding character in the hierarchical text based on the weight value of the first character weighted distribution and the character index position in the corresponding hierarchical content, combined with all matching points identified in the character matching process, and generates hierarchical first character positioning distribution information.

5. The intelligent retrieval system for electronic archive information according to claim 1, characterized in that: The character alignment module includes: The position matching submodule extracts a character segment of the same length as the query string from the text based on the hierarchical first character location distribution information and the first character location as the starting point, establishes a pairing relationship between the starting index and the paragraph content, and obtains a query index mapping table; The consistency calculation submodule compares the query fragments recorded in the query index mapping table with the query string character by character, and counts the number of characters in each character fragment that are consistent with the query string. Calculate and output the character consistency ratio value; The block screening submodule compares the character consistency ratio value with the consistency judgment threshold, performs screening operations on all segments, constructs all continuous paragraph blocks that meet the consistency comparison requirements, and integrates to establish a hierarchical character-by-character consistent block set.

6. The intelligent retrieval system for electronic archive information according to claim 1, characterized in that: The similarity annotation module includes: The index mapping submodule extracts the corresponding text segment content based on the hierarchical character-by-character consistent block set, identifies the hierarchical structure label and paragraph number to which the block belongs, and indexes and binds the query character with the block character one by one to obtain an index mapping set; The label generation submodule performs similarity labeling on each set of character mapping segments based on the index mapping set and the number of character matches and the length of the continuous matching segments to obtain a block similarity assessment result; The list output submodule classifies and archives the blocks according to the hierarchical structure based on the block similarity assessment results, combined with the label determination results and the hierarchical label and block number information provided in the index mapping, and establishes a hierarchical similar block annotation list.

7. The intelligent retrieval system for electronic archive information according to claim 1, characterized in that: The system further comprises: The index integration module integrates the similar block annotations and the hierarchical content based on the hierarchical similar block annotation list and the high-priority classified content set of the archive, establishes an overall archive index, and generates an intelligent retrieval index structure for archive electronic information; The intelligent retrieval index structure of archival electronic information includes an integrated index table, hierarchical index nodes, and a global retrieval mapping.

8. The intelligent retrieval system for electronic archive information according to claim 7, characterized in that: The index integration module includes: The block integration submodule reads the block structure annotation content based on the hierarchical similar block annotation list, checks whether there is overlap between the blocks, and merges them into continuous blocks if overlap is found, and obtains a hierarchical integration block index table; The content fusion submodule performs information level identification processing according to the hierarchical integration block index table, combines the original content of each layer in the archive high-priority classification content set, and performs information level identification processing according to the similarity tags marked on the blocks, adds index anchor points and identification fields, and obtains hierarchical fusion text structure blocks; The index generation submodule performs structural integration of archives based on the anchor structure and index fields in the hierarchical fusion text structure block, performs field classification processing, sets the query entry of the keyword dimension and the label dimension of the hierarchical level, and generates an intelligent retrieval index structure for archive electronic information.

Images