Named entity consistency correction method and system for long text block alignment

Abstract

This invention discloses a named entity consistency correction method and system for long text block alignment, relating to the field of text processing. The method includes: dividing the text into blocks; constructing pending entity records based on the text blocks and writing the pending entity records into an inter-block state transfer data packet; transmitting the data packet between adjacent text blocks and associating and matching adjacent pending entity records, keeping conflicting pending entity records separate, associating and merging pending entity records that do not conflict and meet the association conditions, and keeping pending entity records that do not conflict but do not meet the association conditions in a pending state; when an explicit rebinding statement appears later, determining a valid text structure range, performing reverse updates to determine the canonical entity identifier for pending entity records that meet the conditions within this range, retaining the association and merging results for pending entity records that are not within the valid text structure range but meet the association conditions, and outputting a correction result set. This invention improves the named entity consistency of long texts.

Description

Technical Field

[0001] This invention relates to the field of long text segmentation processing technology, and specifically to a named entity consistency correction method and system for long text segmentation alignment. Background Technology

[0002] Named entity consistency correction refers to the process of identifying, merging, distinguishing, and standardizing the identification of names of people, organizations, places, roles, numbered objects, and other named entities with specific meanings within the same long text or multiple text segments obtained through block processing. This ensures that different surface writings pointing to the same real object or the same standardized object are assigned the same standardized entity identification, and that entity mentions with the same surface writing but located in different headings, definitions, footnotes, speaking units, or stages maintain different standardized entity identifications.

[0003] With the digitization of legal documents, audit reports, accident investigation materials, full-text technical specifications, and long conference transcripts, performing natural language analysis, chunking, entity recognition, and cross-segment information association on long texts has become an important data processing direction. Existing long text processing technologies typically first parse the electronic text structurally, then maintain the ability to process long contexts through chunking, overlapping windows, block-level representation, or buffered context passing, thereby supporting subsequent named entity recognition, relation extraction, question-answering retrieval, and summary generation. Recent published research also shows that long texts often contain nested chapters, footnote chains, long-distance dependencies, and cross-block entity references; therefore, long text processing has evolved from simple truncated input to refined processing oriented towards structured contexts. In long text processing, text chunking can be based on structural boundaries formed by title transitions, footnote association completion points, speaker transition points, and stage transition points.

[0004] In specific long text scenarios, the same role title can be re-bound to different real entities within different structural text structures. Existing technologies tend to rely on fixed-length blocks, overlapping windows, inter-block alias caching, and global literal unification for consistency correction. This leads to problems when nested chapters, footnote references, definition clauses, speaker switching, and phased redefinitions appear simultaneously in the text. The system is prone to incorrectly merging different entities and incorrectly splitting the same entity. Especially when entities cross block boundaries or when explicit binding statements appear later in the text, existing solutions often make premature decisions in the preceding text, making subsequent unambiguous correction difficult. Recent published literature has exposed this problem from the perspectives of cross-block entities, role drift, footnote and chapter association, and loss of token-level annotations in long documents, but it still lacks a closed-loop correction mechanism based on the precondition that the structural text structure allows for association. Summary of the Invention

[0005] This invention addresses the technical problems in existing technologies for long text processing, such as insufficient reliance on context for named entity consistency correction, failure of cross-block association, and entity merging errors and inconsistencies in identifiers caused by missing local definition constraints. It provides a named entity consistency correction method and system for long text block alignment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A named entity consistency correction method for long text block alignment includes the following steps: S1: Read the input long text, perform text structure range parsing on the input long text, and generate a sentence-by-sentence structure annotation sequence; S2: Determine the block boundaries based on the sentence-by-sentence structure annotation sequence and generate text blocks; S3: Extract candidate named entity mentions within each text block, construct a pending entity record for each candidate named entity mention, and write the pending entity record into the inter-block state transfer data packet; S4: Pass inter-block status data packets between adjacent text blocks, and match the pending entity records in the current text block with the pending entity records passed from the previous text block. Keep pending entity records that conflict separate, perform association merging on pending entity records that do not conflict and meet the association conditions, and keep pending entity records that do not conflict but do not meet the association conditions in a pending state. S5: When an explicit rebinding statement appears in the following text of the current text block, determine the effective text structure range of the explicit rebinding statement, and perform reverse update on the pending entity records that meet the reverse update conditions within the effective text structure range to determine the standard entity identifier. For pending entity records that are not within the effective text structure range but have met the association conditions, retain the association merging result formed in S4, and finally output the correction result set.

[0008] To optimize the above technical solution, the specific measures also include: Furthermore, S1 specifically includes: S1-1: Divide the input long text into sentence unit sequences according to sentence-end punctuation and footnote link positions. ; This represents the Nth sentence unit; S1-2: Perform text structure range parsing on each sentence unit to generate structure-labeled tuples, the first... i The structure annotation tuple of each sentence unit ,in, Sentence unit Title path code, Sentence unit Footnote association code, Sentence unit The definition trigger flag is set to 1 if the current sentence unit contains a local redefinition statement, and 0 if the current sentence unit does not contain a local redefinition statement. Sentence unit The speaking unit code, Sentence unit Stage code; Title path code The footnote association code is obtained by encoding sequentially through a fixed hierarchy from the document root title to the innermost title of the current sentence unit. The unit codes are obtained by sequentially encoding the text from the main text to the footnotes in a fixed citation order. Phase codes are generated based on consecutive speech segments. Generates statements based on time, process, and version switching. S1-3: Write the default footnote link code to sentence units where no footnote link was detected. For sentence units where no speaker switching was detected, a default speaker unit code is written. For sentence units where no phase transition was detected, a default phase code is written. ; S1-4: Traverse each sentence unit to obtain the sentence-by-sentence structure annotation sequence. ; It is the first The structure annotation tuple of each sentence unit.

[0009] Furthermore, S2 specifically includes: S2-1: For adjacent sentence units and Calculate the inter-sentence boundary strength ,in, This indicates the title path switching indicator, which is the title path code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates the footnote association chain switching indicator, which is the footnote association chain code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates a definition trigger switch indicator; the definition of an adjacent sentence unit triggers the flag. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates the speech unit switching indicator, when the speech unit code of the adjacent sentence unit... and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates a stage switching indicator, which is the stage code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; S2-2: The cumulative length of the current text block is The minimum block length is The maximum block length is When satisfied and At that time, the sentence unit The position at the end of the sentence is determined as the priority block boundary; when the cumulative length from the previous block boundary to the current sentence unit reaches the maximum block length. And if no priority block boundary is formed within this interval, the maximum block length will not be exceeded. The position of the last complete sentence is determined as the mandatory block boundary; if there is no block within the current interval that does not exceed the maximum block length... If the complete sentence end position is determined, then the sentence end position of the current sentence unit will be determined as the forced block boundary; S2-3: Generate a block sequence based on all priority block boundaries and forced block boundaries. ; This represents the Mth text block.

[0010] Furthermore, S3 specifically includes: S3-1: Extract candidate named entity mentions within each text block. and mentions for candidate named entities Constructing undetermined entity records ,in, Indicates candidate named entity mentions A collection of aliases Indicates candidate named entity mentions The set of descriptive words, Indicates candidate named entity mentions Structural range identifier, Indicates candidate named entity mentions Location index in the full text, Indicates candidate named entity mentions The status codes have values ​​of 0 (pending), 1 (defined entity identifier), and 2 (isolated). The descriptor set... References by candidate named entities The alias set is obtained after standardizing the nouns, determiners, and role phrases in the sentence unit, the preceding sentence unit, and the following sentence unit. At least include candidate named entity mentions The original surface form and the standardized surface form after normalization of capitalization, punctuation, and appellation; S3-2: Extract role phrases from the context of candidate named entity mentions according to the fixed priority of appositive phrases in the same sentence, definition trigger phrases, and current title limiting phrases; after extracting the first valid role phrase, encode the first valid role phrase to obtain candidate named entity mentions. Role-restricted identifier code And stop extracting other character phrases; S3-3: Generate structural range identifiers according to the following formula. ,in This represents a fixed-order encoding function. Indicates candidate named entity mentions The title path code of the sentence unit. Indicates candidate named entity mentions The footnote association code of the sentence unit, Indicates candidate named entity mentions Role-restricted identifier code, Indicates candidate named entity mentions The stage code of the sentence unit. Indicates candidate named entity mentions The speech unit code of the sentence unit.

[0011] S3-4: Write the current set of valid structure scope fields, the set of pending entity records, the set of confirmed specification entity records, and the set of explicit rebinding relationships into the inter-block state transfer data packet. .

[0012] Furthermore, S4 specifically includes: S4-1: For the pending entity record passed from the previous text block. With the pending entity records in the current text block Calculate the structural range compatibility coefficient When the structural range compatibility coefficient When the value is 1, it indicates that the two corresponding undetermined entity records meet the prerequisites for entering the subsequent similarity comparison and association score calculation. S4-2: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Alias ​​similarity between ,in, Indicates a candidate named entity reference in the previous text block. A collection of aliases Indicates candidate named entity references for the current text block. n A collection of aliases This indicates the number of elements in the intersection of alias sets. Indicates the number of elements in the union of alias sets; S4-3: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Descriptive similarity between ,in, Indicates a candidate named entity reference in the previous text block. The set of descriptive words, Indicates candidate named entity references for the current text block. The set of descriptive words, This indicates the number of elements in the intersection of the sets of descriptors. Represents a set of descriptive terms The number of elements, Represents a set of descriptive terms The number of elements; S4-4: Calculate the correlation score according to the following formula. ; S4-5: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Conflict count ; S4-6: Let the association threshold be... ,when At that time, determine the pending entity records of the previous text block. Pending entity records in the current text block There is a conflict that blocks the merge; the pending entity will be recorded. and Maintain separation and record the undetermined entities in the current text block. status codes Set to 2; when and When determining the pending entity record Records of entities to be determined The association conditions are met; if the pending entity record status codes If the value is 1, then the entity to be determined will be recorded. Associated with the pending entity record The same canonical entity identifier, and record the entity to be determined. status codes Set to 1; if the entity record is pending. status codes If the value is 0, then the undetermined entity will be recorded. Records of entities to be determined The records are associated with the same pending entity record, and the pending entity record is linked. status codes Maintain at 0; when and When determining the pending entity record The association conditions have not yet been met; therefore, the entity record is retained pending. The pending state and its status code. Maintain at 0.

[0013] Furthermore, S5 specifically includes: S5-1: When an explicit rebinding statement is detected At that time, generate explicit rebinding statements. Corresponding standardized entity identifier And record the starting position of the explicit rebinding statement. ; S5-2: Determine the explicit rebinding statement Valid text structure range , It is the starting position of the explicit rebinding statement. It is the end position of the effective text structure range; S5-3: For any undetermined entity record When the position index relationship is satisfied simultaneously Pending entity records Collection of aliases Includes explicit rebinding statements The re-bound entity surface form Or its normalized surface form, title path compatibility function Footnote association chain contains functions and pending entity records With explicit rebinding statement Conflict count At that time, record the entity to be determined. The canonical entity identifier is determined as an explicit rebinding statement. Corresponding standardized entity identifier And record the entity to be determined. status codes Set to 1; Indicates a record of an entity pending determination. The corresponding candidate named entity references their position index in the full text. Indicates candidate named entity mentions The title path code of the sentence unit. Indicates an explicit rebinding statement The title path code of the sentence unit. Indicates candidate named entity mentions The footnote association code of the sentence unit, Indicates an explicit rebinding statement The footnote association code of the sentence unit; S5-4: For those not falling within the range of valid text structure For pending entity records, retain the association merging result formed in S4. For pending entity records that do not meet the explicit rebinding condition or the association condition in step S4, output them as independent entities and set the status code to 2. S5-5: Output calibration result set ,in, This represents the constructor for the result set. Indicates a record of an entity pending determination. The original surface form mentioned by the corresponding candidate named entities, This indicates the normalized surface form of candidate named entity references after normalization of capitalization, punctuation, and appellation. Indicates a record of an entity pending determination. The final corresponding canonical entity identifier, Indicates a record of an entity pending determination. The corresponding structural range identifier, Indicates a record of an entity pending determination. The block number to which it belongs Indicates a record of an entity pending determination. The corresponding candidate named entity references their position index in the full text. Indicates a record of an entity pending determination. The final status code, Indicates a record of an entity pending determination. The source of the correction is marked.

[0014] Furthermore, in S4-1, the structural range compatibility coefficient The calculation formula is as follows:

[0015] in, This indicates a function for compatibility with title paths. Indicates a candidate named entity reference in the previous text block. The title path code of the sentence unit. Indicates candidate named entity references for the current text block. The title path code of the sentence unit, when In the position of When it is on the subsequent title chain of the root, The value is 1 if it is set to 1, otherwise the value is 0. This indicates that the footnote association chain contains functions. Indicates a candidate named entity reference in the previous text block. The footnote association code of the sentence unit, Indicates candidate named entity references for the current text block. The footnote association code of the sentence unit, when Include When used as a prefix association chain The value is 1 if it is set to 1, otherwise the value is 0. Functions indicating equality of role-restricted identifiers. Indicates a candidate named entity reference in the previous text block. Role-restricted identifier code, Indicates candidate named entity references for the current text block. Role-restricted identifier code, when and If they are the same, then The value is 1 if it is set to 1, otherwise the value is 0. Indicates the function for stage code equality. Indicates a candidate named entity reference in the previous text block. The stage code of the sentence unit. Indicates candidate named entity references for the current text block. The stage code of the sentence unit, when and If they are the same The value is 1 if it is set to 1, otherwise the value is 0. Functions indicating equality of speech unit codes. Indicates a candidate named entity reference in the previous text block. The speech unit code of the sentence unit it belongs to. Indicates candidate named entity references for the current text block. The speech unit code of the sentence unit, when and If they are the same The value is 1 if it is not 1, otherwise the value is 0.

[0016] Furthermore, in S4-5, the collision counting... The calculation formula is as follows:

[0017] in, This indicates an explicitly defined conflict indicator. It takes the value of 1 when the same alias is explicitly bound to different canonical entities within the comparison interval, and otherwise takes the value of 0. Indicates the number of title path conflict indicators, when and The value is 1 if the elements do not belong to each other's subsequent title chains, and 0 otherwise. Indicates a candidate named entity reference in the previous text block. The title path code of the sentence unit. Indicates candidate named entity references for the current text block. The title path code of the sentence unit; Indicates the phase conflict indicator quantity, when and If they are different, the value is 1; otherwise, the value is 0. Indicates a candidate named entity reference in the previous text block. The stage code of the sentence unit. Indicates candidate named entity references for the current text block. The stage code of the sentence unit it belongs to; Indicates the conflict indication quantity of the speaking unit, when and If they are different, the value is 1; otherwise, the value is 0. Indicates a candidate named entity reference in the previous text block. The speech unit code of the sentence unit it belongs to. Indicates candidate named entity references for the current text block. The speech unit code of the sentence unit.

[0018] Furthermore, in S5-2, the formula for calculating the end position of the effective text structure range interval is as follows:

[0019] in, Indicates the starting position The next heading boundary at the same or higher level. Indicates the starting position The next stage involves switching positions. Indicates the starting position The next speaking unit will switch positions.

[0020] This invention also proposes a named entity consistency correction system for long text block alignment, comprising: The text structure range parsing and structure boundary segmentation module is used to read long input text, perform text structure range parsing on the long input text, generate a sentence-by-sentence structure annotation sequence, determine the segmentation boundary based on the sentence-by-sentence structure annotation sequence, and generate text segments. The module for generating pending entity records and transmitting inter-block status is used to extract candidate named entity mentions in each text block, construct pending entity records for each candidate named entity mention, write the pending entity records into the inter-block status transmission data packet, and transmit the inter-block status transmission data packet between adjacent text blocks. The cross-block association determination and reverse update module is used to match the pending entity records in the current text block with the pending entity records passed from the previous text block. For pending entity records with conflicts, they are kept separate. For pending entity records without conflicts and that meet the association conditions, association merging is performed. For pending entity records without conflicts but that do not meet the association conditions, they are kept in a pending state. When an explicit rebinding statement appears in the following text of the current text block, the effective text structure range of the explicit rebinding statement is determined. Within the effective text structure range, the pending entity records that meet the reverse update conditions are reverse updated to determine the canonical entity identifier. For pending entity records that are not within the effective text structure range but have met the association conditions, their association merging results are retained. Finally, the correction result set is output.

[0021] The beneficial effects of this invention are: This invention transforms the long text named entity consistency correction process from the traditional literal similarity merging process into a process of association, conflict blocking, and canonical entity identification determination under structural scope constraints. First, by performing structural scope compatibility determination before association scoring, the probability of incorrectly merging entities with the same name across different structural scopes is reduced. Second, a pending entity record mechanism preserves intermediate states when there is insufficient evidence for final entity entries, preventing premature irreversible merging results due to insufficient prior evidence. Third, by transmitting the current effective structural scope field set, pending entity record set, determined canonical entity record set, and explicit rebinding relationship set through inter-block state data packets, cross-block alignment is maintained without relying on large-scale original text overlap. Fourth, when explicit rebinding statements appear later, reverse updates can be performed within the effective text structural scope range, ensuring that mentions of the same entity scattered across different blocks receive a unified canonical entity identification, thereby improving the stability of subsequent relation extraction, question-answering retrieval, and knowledge modeling. Attached Figure Description

[0022] Figure 1 This is a flowchart of the named entity consistency correction method for long text block alignment proposed in this invention.

[0023] Figure 2This is a block diagram of the named entity consistency correction system for long text block alignment proposed in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] Example 1 This invention proposes a named entity consistency correction method for long text block alignment. The process of this method is as follows: Figure 1 As shown, it includes the following steps: S1: Read the input long text and perform text structure range parsing on the input long text to generate a sentence-by-sentence structure annotation sequence; S1 specifically includes: S1-1: Divide the input long text into sentence unit sequences according to sentence-end punctuation and footnote link positions. ; This represents the Nth sentence unit; S1-2: Perform text structure range parsing on each sentence unit to generate structure-labeled tuples, the first... i The structure annotation tuple of each sentence unit ,in, Sentence unit Title path code, Sentence unit Footnote association code, Sentence unit The definition trigger flag is set to 1 if the current sentence unit contains a local redefinition statement, and 0 if the current sentence unit does not contain a local redefinition statement. Sentence unit The speaking unit code, Sentence unit Stage code; Title path code The footnote association code is obtained by encoding sequentially through a fixed hierarchy from the document root title to the innermost title of the current sentence unit. The unit codes are obtained by sequentially encoding the text from the main text to the footnotes in a fixed citation order. Phase codes are generated based on consecutive speech segments. Generates statements based on time, process, and version switching. S1-3: Write the default footnote link code to sentence units where no footnote link was detected. For sentence units where no speaker switching was detected, a default speaker unit code is written. For sentence units where no phase transition was detected, a default phase code is written. This ensures that all sentence units have a uniformly formatted structure annotation tuple.

[0026] S1-4: Traverse each sentence unit to obtain the sentence-by-sentence structure annotation sequence. ; It is the first The structure annotation tuple of each sentence unit.

[0027] S2: Determine block boundaries based on the sentence-by-sentence structure annotation sequence and generate text blocks; S2 specifically includes: S2-1: To quantify whether a text structure range change occurs between adjacent sentence units sufficient to trigger chunking, and accordingly prioritize chunk boundaries falling at the point where the structure switch is completed, for adjacent sentence units... and Calculate the inter-sentence boundary strength ,in, This indicates the title path switching indicator, which is the title path code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates the footnote association chain switching indicator, which is the footnote association chain code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates a definition trigger switch indicator; the definition of an adjacent sentence unit triggers the flag. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates the speech unit switching indicator, when the speech unit code of the adjacent sentence unit... and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates a stage switching indicator, which is the stage code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; inter-sentence boundary strength It is used to characterize the degree of structural change between adjacent sentence units, and can determine whether a text structure range switch has occurred between sentences, so that the block boundary falls preferentially at the text structure range completion point, rather than falling at an arbitrary length position; The larger the value of , the more suitable the position is as a candidate position for block boundary.

[0028] S2-2: The cumulative length of the current text block is The minimum block length is The maximum block length is When satisfied and At that time, the sentence unit The position at the end of the sentence is determined as the priority block boundary; when the cumulative length from the previous block boundary to the current sentence unit reaches the maximum block length. And if no priority block boundary is formed within this interval, the maximum block length will not be exceeded. The position of the last complete sentence is determined as the mandatory block boundary; if there is no block within the current interval that does not exceed the maximum block length... If the complete sentence end position is determined, then the sentence end position of the current sentence unit will be determined as the forced block boundary; S2-3: Generate a block sequence based on all priority block boundaries and forced block boundaries. ; This represents the Mth text block.

[0029] In long text processing scenarios, entity consistency errors not only originate from the entity recognition process but may also stem from inconsistencies between block boundaries and text structural ranges such as title transitions, footnote additions, definition statements, and stage transitions. Fixed-length blocks or simple overlapping window blocks do not explicitly consider these structural boundaries, easily disrupting the local definition environment. This invention generates block boundaries by prioritizing the determination of text structural range completion points, thereby reducing interference from the block segmentation stage on subsequent entity consistency correction. This invention does not segment first and then compensate; instead, it first determines which positions belong to the text structural range completion points and then prioritizes forming block boundaries at these positions. The advantage of this approach is that each subsequent block retains as much of its local definition environment as possible before entering entity extraction and consistency correction. In other words, this invention first correctly segments the input text before performing subsequent entity judgments, rather than simply relying on more complex post-processing to correct errors caused by the initial segmentation.

[0030] S3: Extract candidate named entity mentions within each text block, construct a pending entity record for each candidate named entity mention, and write the pending entity record into the inter-block state transfer data packet; S3 specifically includes: S3-1: Since the contextual evidence for candidate named entity references within a single block may not be complete, to avoid prematurely determining canonical entity identifiers when the evidence for the final entity is insufficient, candidate named entity references are extracted within each text block. and mentions for candidate named entities Constructing undetermined entity records ,in, Indicates candidate named entity mentions A collection of aliases Indicates candidate named entity mentions The set of descriptive words, Indicates candidate named entity mentions Structural range identifier, Indicates candidate named entity mentions Location index in the full text, Indicates candidate named entity mentions The status codes have values ​​of 0 (pending), 1 (defined entity identifier), and 2 (isolated). The descriptor set... References by candidate named entities The alias set is obtained after standardizing the nouns, determiners, and role phrases in the sentence unit, the preceding sentence unit, and the following sentence unit. At least include candidate named entity mentions The original surface form and the standardized surface form after normalization of capitalization, punctuation, and appellation; S3-2: Extract role phrases from the context of candidate named entity mentions according to the fixed priority of appositive phrases in the same sentence, definition trigger phrases, and current title limiting phrases; after extracting the first valid role phrase, encode the first valid role phrase to obtain candidate named entity mentions. Role-restricted identifier code And stop extracting other character phrases; S3-3: To consolidate the title path, footnote association chain, role qualifying identifier, stage information, and speaker unit information of candidate named entity references into comparable uniform text structure scope identifiers, the structure scope identifier is generated according to the following formula. ,in This represents a fixed-order encoding function. Indicates candidate named entity mentions The title path code of the sentence unit. Indicates candidate named entity mentions The footnote association code of the sentence unit, Indicates candidate named entity mentions Role-restricted identifier code, Indicates candidate named entity mentions The stage code of the sentence unit. Indicates candidate named entity mentions The speech unit code of the sentence unit in which it is located. Structural range identifier. The coefficient is used to calculate the structural range compatibility coefficient in step S4-1 and to determine in step S5-3 whether the pending entity record is within the valid text structure range of the explicit rebinding statement, thereby determining whether different pending entity records have the structural prerequisites for association, merging, or reverse update.

[0031] S3-4: Write the current set of valid structure scope fields, the set of pending entity records, the set of confirmed specification entity records, and the set of explicit rebinding relationships into the inter-block state transfer data packet. Therefore, candidate named entity mentions in the current block are converted into pending entity records that can be compared across blocks, providing a unified processing object for subsequent association determination and conflict constraints.

[0032] The current set of valid structural scope fields includes the title path code, footnote association chain code, stage code, and speech unit code that are still in effect at the end of the current text block. This set of valid structural scope fields serves as input for structural scope compatibility determination, association score calculation, and conflict constraint determination for the next text block.

[0033] The set of pending entity records is used to preserve the intermediate state of candidate named entities when there is insufficient evidence for finalizing the entity, so that when new evidence appears later, the unambiguous determination of the canonical entity identifier can be performed.

[0034] Explicit rebinding relationships are structured mapping records generated based on explicit rebinding statements. These records include at least the surface form of the rebinded entity, the corresponding canonical entity identifier, the full-text position index of the explicit rebinding statement, the title path code, footnote association chain code, stage code, speech unit code, and the effective text structure range. Explicit rebinding relationships are used to define the target and scope of the reverse update in step S5. The set of explicit rebinding relationships includes those still within the effective text structure range.

[0035] The set of pending entity records includes pending entity records with a status code of 0; the set of defined entity records includes entity records with a status code of 1 and a defined entity identifier.

[0036] The inter-block status transfer data packet includes the data packet number, the previous text block number, the next text block number, the current valid structure range field set, the pending entity record set, the determined specification entity record set, and the explicit rebinding relationship set. After receiving the inter-block status transfer data packet, the next text block performs the structure range compatibility determination, association score calculation, and conflict constraint determination in step S4 based on the fields in the packet.

[0037] In this step, the present invention does not immediately merge candidate named entity mentions into the final entity; instead, it first saves them as pending entity records. The principle behind this is to first save the currently observed name clues, contextual constraints, and the textual structure range, distinguishing between the known and unconfirmed parts. This way, the system avoids making premature, irreversible judgments due to incomplete evidence, leaving room for subsequent cross-block comparisons and textual corrections. The uniqueness of this step lies not in storing an extra copy of intermediate data, but in breaking down the entity merging process, which should have been completed in one go, into a two-stage process of first retaining and then finalizing.

[0038] This invention does not transmit simple preceding text content, but rather a pre-organized inter-block state transmission data packet. This data packet retains the currently valid title path, footnote association chains, explicit rebinding relationships, and pending entity records. Therefore, the next text block receives not scattered background information, but an intermediate state with a clearly organized structure. The advantage of this approach is that cross-block processing no longer relies on continuously expanding the original context, but rather on the continuous transmission of filtered key state information. Compared to conventional methods, this step elevates cross-block continuity from a text-level extension to a state-level extension.

[0039] S4: Transmit inter-block state transfer data packets between adjacent text blocks, and perform association matching between pending entity records in the current text block and pending entity records transmitted from the previous text block. Pending entity records with conflicts remain separated; pending entity records without conflicts and meeting the association conditions are merged; pending entity records without conflicts but not meeting the association conditions remain in a pending state. S4 specifically includes: S4-1: For the pending entity record passed from the previous text block. With the pending entity records in the current text block Calculate the structural range compatibility coefficient When the structural range compatibility coefficient A value of 1 indicates that the two corresponding undetermined entity records meet the prerequisites for subsequent similarity comparison and association scoring calculation; structural range compatibility coefficient The calculation formula is as follows:

[0040] in, This indicates a function for compatibility with title paths. Indicates a candidate named entity reference in the previous text block. The title path code of the sentence unit. Indicates the candidate named entity mentions of the current text block. The title path code of the sentence unit, when In the position of When it is on the subsequent title chain of the root, The value is 1 if it is set to 1, otherwise the value is 0. This indicates that the footnote association chain contains functions. Indicates a candidate named entity reference in the previous text block. The footnote association code of the sentence unit, Indicates the candidate named entity mentions of the current text block. The footnote association code of the sentence unit, when Include When used as a prefix association chain The value is 1 if it is set to 1, otherwise the value is 0. Functions indicating equality of role-restricted identifiers. Indicates a candidate named entity reference in the previous text block. Role-restricted identifier code, Indicates the candidate named entity mentions of the current text block. Role-restricted identifier code, when and If they are the same, then The value is 1 if it is set to 1, otherwise the value is 0. Indicates the function for stage code equality. Indicates a candidate named entity reference in the previous text block. The stage code of the sentence unit. Indicates the candidate named entity mentions of the current text block. The stage code of the sentence unit, when and If they are the same The value is 1 if it is set to 1, otherwise the value is 0. Functions indicating equality of speech unit codes. Indicates a candidate named entity reference in the previous text block. The speech unit code of the sentence unit it belongs to. Indicates the candidate named entity mentions of the current text block. The speech unit code of the sentence unit, when and If they are the same The value is 1 if it is not 1, otherwise the value is 0.

[0041] S4-2: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Alias ​​similarity between ,in, Indicates a candidate named entity reference in the previous text block. A collection of aliases Indicates candidate named entity references for the current text block. n A collection of aliases This indicates the number of elements in the intersection of alias sets. The number of elements in the union of aliases; alias similarity. Used to characterize the degree of overlap between two pending entity records at the name level.

[0042] S4-3: To measure the proximity of two pending entity records at the local semantic description level, calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Descriptive similarity between ,in, Indicates a candidate named entity reference in the previous text block. The set of descriptive words, Indicates candidate named entity references for the current text block. The set of descriptive words, This indicates the number of elements in the intersection of the sets of descriptors. Represents a set of descriptive terms The number of elements, Represents a set of descriptive terms The number of elements; S4-4: Calculate the correlation score according to the following formula. Related ratings Used to quantify the reliability of the association between two undetermined entity records, within the limits allowed by the text structure; S4-5: To avoid erroneously merging pending entity records that fall within different valid text structure ranges even when the correlation score is high, calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Conflict count Conflict count The calculation formula is as follows:

[0043] in, This indicates an explicitly defined conflict indicator. It takes the value of 1 when the same alias is explicitly bound to different canonical entities within the comparison interval, and otherwise takes the value of 0. Indicates the number of title path conflict indicators, when and The value is 1 if the elements do not belong to each other's subsequent title chains, and 0 otherwise. Indicates a candidate named entity reference in the previous text block. The title path code of the sentence unit. Indicates candidate named entity references for the current text block. The title path code of the sentence unit; Indicates the phase conflict indicator quantity, when and If they are different, the value is 1; otherwise, the value is 0. Indicates a candidate named entity reference in the previous text block. The stage code of the sentence unit. Indicates candidate named entity references for the current text block. The stage code of the sentence unit it belongs to; Indicates the conflict indication quantity of the speaking unit, when and If they are different, the value is 1; otherwise, the value is 0. Indicates a candidate named entity reference in the previous text block. The speech unit code of the sentence unit it belongs to. Indicates candidate named entity references for the current text block. The speech unit code of the sentence unit in which it is located. Conflict count. Used as a priority over correlation scoring The pre-merge condition is used to prevent entities with the same name in different text structures from being mistakenly merged.

[0044] S4-6: Let the association threshold be... ,when At that time, determine the pending entity records of the previous text block. Pending entity records in the current text block There is a conflict that blocks the merge; the pending entity will be recorded. and Maintain separation and record the undetermined entities in the current text block. status codes Set to 2; when and When determining the pending entity record Records of entities to be determined The association conditions are met; if the pending entity record status codes If the value is 1, then the entity to be determined will be recorded. Associated with the pending entity record The same canonical entity identifier, and record the entity to be determined. status codes Set to 1; if the entity record is pending. status codes If the value is 0, then the undetermined entity will be recorded. Records of entities to be determined The records are associated with the same pending entity record, and the pending entity record is linked. status codes Maintain at 0; when and When determining the pending entity record The association conditions have not yet been met; therefore, the entity record is retained pending. The pending state and its status code. Maintain at 0.

[0045] This invention changes the judgment order in this step. First, it checks whether the two pending entity records are within the allowed text structure range for association; then it checks if the names and partial descriptions are similar; and finally, it specifically checks for conflicts that should prevent merging. The working principle of this approach is: first, it filters out objects that should not be compared in principle; then it compares comparable objects; and finally, it sets up a special veto step for those cases that are superficially similar but should not be merged in substance. The role of this step in this solution is not only to improve accuracy, but more importantly, to elevate the prohibition of erroneous merging to a separate, explicit judgment step, rather than incidentally relying on the natural differentiation of similarity.

[0046] S5: When an explicit rebinding statement appears in the following text of the current text block, determine the effective text structure range of the explicit rebinding statement, and within the effective text structure range, perform reverse update on the pending entity records that meet the reverse update conditions to determine the canonical entity identifier. For pending entity records that are not within the effective text structure range but have met the association conditions, retain the association merging results formed in S4, and finally output the correction result set. S5 specifically includes: S5-1: When an explicit rebinding statement is detected At that time, generate explicit rebinding statements. Corresponding standardized entity identifier And record the starting position of the explicit rebinding statement. ; S5-2: Determine the explicit rebinding statement Valid text structure range , It is the starting position of the explicit rebinding statement. This is the ending position of the valid text structure range; the formula for calculating the ending position of the valid text structure range is as follows:

[0047] in, Indicates the starting position The next heading boundary at the same or higher level. Indicates the starting position The next stage involves switching positions. Indicates the starting position The next speaking unit will switch positions.

[0048] S5-3: For any undetermined entity record When the position index relationship is satisfied simultaneously Pending entity records Collection of aliases Includes explicit rebinding statements The re-bound entity surface form Or its normalized surface form, title path compatibility function Footnote association chain contains functions and pending entity records With explicit rebinding statement Conflict count At that time, record the entity to be determined. The canonical entity identifier is determined as an explicit rebinding statement. Corresponding standardized entity identifier And record the entity to be determined. status codes Set to 1; Indicates a record of an entity pending determination. The corresponding candidate named entity references their position index in the full text. Indicates candidate named entity mentions The title path code of the sentence unit. Indicates an explicit rebinding statement The title path code of the sentence unit. Indicates candidate named entity mentions The footnote association code of the sentence unit, Indicates an explicit rebinding statement The footnote association code of the sentence unit; S5-4: For those not falling within the range of valid text structure For pending entity records, retain the association merging result formed in S4. For pending entity records that do not meet the explicit rebinding condition or the association condition in step S4, output them as independent entities and set the status code to 2. S5-5: To uniformly output the corrected canonical entity identifiers along with their corresponding text structure range information, block location information, status information, and correction source information, output the correction result set. ,in, This represents the constructor for the result set. Indicates a record of an entity pending determination. The original surface form mentioned by the corresponding candidate named entities, This indicates the normalized surface form of candidate named entity references after normalization of capitalization, punctuation, and appellation. Indicates a record of an entity pending determination. The final corresponding canonical entity identifier, Indicates a record of an entity pending determination. The corresponding structural range identifier, Indicates a record of an entity pending determination. The block number to which it belongs Indicates a record of an entity pending determination. The corresponding candidate named entity references their position index in the full text. Indicates a record of an entity pending determination. The final status code, Indicates a record of an entity pending determination. The source of the correction is marked.

[0049] Final status code in the calibration result set The value can be 0, 1, or 2. Status code 0 indicates that the entity correction record is still a pending record in the current processing round. Status code 1 indicates that the entity correction record has been identified as a standard entity. Status code 2 indicates that the entity correction record is output as an independent entity due to structural conflicts or failure to meet association conditions. When the entity record to be determined is associated and merged in step S4 to determine the canonical entity identifier, its correction source mark is... Write to determine the association; when the pending entity record undergoes explicit rebinding and reverse update in step S5 to determine the canonical entity identifier, its correction source mark is determined. Write as a rebinding update; when the pending entity record does not meet the association conditions and is not overwritten by explicit rebinding, its correction source flag is... Write as an independent output.

[0050] This invention introduces a reverse update following explicit rebinding at this step. Instead of simply recording subsequent information, it uses explicit correspondences appearing later as new criteria for judgment, returning to the scope of the text structure where they are legally valid. It then re-unifies and standardizes the entity identifiers of records that are still pending or can be corrected. The purpose of this step is to bring late-arriving information back into the preceding context, freeing the system from the restriction of the order in which information appears. Compared to conventional techniques, this invention does not wait for errors to be naturally exposed, but actively corrects them when highly certain clues appear later in the text, thus enabling the definition information scattered throughout a long text to truly coordinate.

[0051] Example 2 This invention proposes a named entity consistency correction system for long text block alignment, corresponding to the method in Embodiment 1. The system's structural block diagram is as follows: Figure 2 ,include: The text structure range parsing and structure boundary segmentation module is used to read long input text, perform text structure range parsing on the long input text, generate a sentence-by-sentence structure annotation sequence, determine the segmentation boundary based on the sentence-by-sentence structure annotation sequence, and generate text segments. The module for generating pending entity records and transmitting inter-block status is used to extract candidate named entity mentions in each text block, construct pending entity records for each candidate named entity mention, write the pending entity records into the inter-block status transmission data packet, and transmit the inter-block status transmission data packet between adjacent text blocks. The cross-block association determination and reverse update module is used to match the pending entity records in the current text block with the pending entity records passed from the previous text block. For pending entity records with conflicts, they are kept separate. For pending entity records without conflicts and that meet the association conditions, association merging is performed. For pending entity records without conflicts but that do not meet the association conditions, they are kept in a pending state. When an explicit rebinding statement appears in the following text of the current text block, the effective text structure range of the explicit rebinding statement is determined. Within the effective text structure range, the pending entity records that meet the reverse update conditions are reverse updated to determine the canonical entity identifier. For pending entity records that are not within the effective text structure range but have met the association conditions, their association merging results are retained. Finally, the correction result set is output.

[0052] The implementation methods of each module and its function in the system are completely consistent with the steps of the method in Implementation Example 1, so they will not be repeated here.

[0053] The most direct improvement of this invention is addressing the problem of mistakenly merging entities with the same name in long texts. For texts interwoven with hierarchical headings, footnotes, local definitions, and stage transitions, conventional methods often prematurely treat different objects as the same entity simply because their names appear identical. This invention first limits the scope of the text structure and then decides whether to allow correlation comparisons, thus preventing many cases where the names are the same but the meanings have changed. This not only reduces the number of errors but also reduces the possibility of subsequent processing steps that depend on entity results being misled. Another important function of this invention is preserving the ability to absorb subsequent supplementary explanations. In long texts, the statement that truly determines the meaning of a term may not appear first; it may appear later in definition clauses, footnotes, or stage transitions. Conventional methods, once a fixed merging is made in the preceding text, often only provide additional explanations when more accurate information appears later, making overall correction difficult. This invention, through pending entity recording and a reverse update mechanism, keeps the preceding judgment in a correctable state when evidence is insufficient, allowing for unified correction once a clear correspondence appears in the following text. This invention also eliminates the need for long text processing to rely solely on continuously increasing block lengths or blindly expanding overlapping areas to maintain contextual coherence. Conventional approaches, in order to minimize cross-block breaks, often involve making blocks longer or retaining more repetitive content between adjacent blocks. However, this leads to increased processing redundancy and dilutes key structural information. This invention, through structural boundary blocking and state transfer between blocks, preserves the key states that truly affect the continuity of entity meaning, ensuring that cross-block connections are based on effective state continuation rather than on the mechanical repetition of the original text.

[0054] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0055] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A named entity consistency correction method for long text block alignment, characterized in that, Includes the following steps: S1: Read the input long text, perform text structure range parsing on the input long text, and generate a sentence-by-sentence structure annotation sequence; S2: Determine the block boundaries based on the sentence-by-sentence structure annotation sequence and generate text blocks; S3: Extract candidate named entity mentions within each text block, construct a pending entity record for each candidate named entity mention, and write the pending entity record into the inter-block state transfer data packet; S4: Pass inter-block status data packets between adjacent text blocks, and match the pending entity records in the current text block with the pending entity records passed from the previous text block. Keep pending entity records that conflict separate, perform association merging on pending entity records that do not conflict and meet the association conditions, and keep pending entity records that do not conflict but do not meet the association conditions in a pending state. S5: When an explicit rebinding statement appears in the following text of the current text block, determine the effective text structure range of the explicit rebinding statement, and perform reverse update on the pending entity records that meet the reverse update conditions within the effective text structure range to determine the standard entity identifier. For pending entity records that are not within the effective text structure range but have met the association conditions, retain the association merging result formed in S4, and finally output the correction result set.

2. The named entity consistency correction method for long text block alignment as described in claim 1, characterized in that, S1 specifically includes: S1-1: Divide the input long text into sentence unit sequences according to sentence-end punctuation and footnote link positions. ; This represents the Nth sentence unit; S1-2: Perform text structure range parsing on each sentence unit to generate structure-labeled tuples, the first... i The structure annotation tuple of each sentence unit ,in, Sentence unit Title path code, Sentence unit Footnote association code, Sentence unit The definition trigger flag is set to 1 if the current sentence unit contains a local redefinition statement, and 0 if the current sentence unit does not contain a local redefinition statement. Sentence unit The speaking unit code, Sentence unit Stage code; Title path code The footnote association code is obtained by encoding sequentially through a fixed hierarchy from the document root title to the innermost title of the current sentence unit. The unit codes are obtained by sequentially encoding the text from the main text to the footnotes in a fixed citation order. Phase codes are generated based on consecutive speech segments. Generates statements based on time, process, and version switching. S1-3: Write the default footnote link code to sentence units where no footnote link was detected. For sentence units where no speaker switching was detected, a default speaker unit code is written. For sentence units where no phase transition was detected, a default phase code is written. ; S1-4: Traverse each sentence unit to obtain the sentence-by-sentence structure annotation sequence. ; It is the first The structure annotation tuple of each sentence unit.

3. The named entity consistency correction method for long text block alignment as described in claim 2, characterized in that, S2 specifically includes: S2-1: For adjacent sentence units and Calculate the inter-sentence boundary strength ,in, This indicates the title path switching indicator, which is the title path code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates the footnote association chain switching indicator, which is the footnote association chain code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates a definition trigger switch indicator; the definition of an adjacent sentence unit triggers the flag. and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates the speech unit switching indicator, when the speech unit code of the adjacent sentence unit... and If they are not the same, the value is 1. and If they are the same, the value is 0; This indicates a stage switching indicator, which is the stage code of the adjacent sentence unit. and If they are not the same, the value is 1. and If they are the same, the value is 0; S2-2: The cumulative length of the current text block is The minimum block length is The maximum block length is When satisfied and At that time, the sentence unit The position at the end of the sentence is determined as the priority block boundary; when the cumulative length from the previous block boundary to the current sentence unit reaches the maximum block length. And if no priority block boundary is formed within this interval, the maximum block length will not be exceeded. The position of the last complete sentence is determined as the mandatory block boundary; if there is no block within the current interval that does not exceed the maximum block length... If the complete sentence end position is determined, then the sentence end position of the current sentence unit will be determined as the forced block boundary; S2-3: Generate a block sequence based on all priority block boundaries and forced block boundaries. ; This represents the Mth text block.

4. The named entity consistency correction method for long text block alignment as described in claim 1, characterized in that, S3 specifically includes: S3-1: Extract candidate named entity mentions within each text block. and mentions for candidate named entities Constructing undetermined entity records ,in, Indicates candidate named entity mentions A collection of aliases Indicates candidate named entity mentions The set of descriptive words, Indicates candidate named entity mentions Structural range identifier, Indicates candidate named entity mentions Location index in the full text, Indicates candidate named entity mentions The status codes have values ​​of 0 (pending), 1 (defined entity identifier), and 2 (isolated). The descriptor set... References by candidate named entities The alias set is obtained after standardizing the nouns, determiners, and role phrases in the sentence unit, the preceding sentence unit, and the following sentence unit. At least include candidate named entity mentions The original surface form and the standardized surface form after normalization of capitalization, punctuation, and appellation; S3-2: Extract role phrases from the context of candidate named entity mentions according to the fixed priority of appositive phrases in the same sentence, definition trigger phrases, and current title limiting phrases; after extracting the first valid role phrase, encode the first valid role phrase to obtain candidate named entity mentions. Role-restricted identifier code And stop extracting other character phrases; S3-3: Generate structural range identifiers according to the following formula. ,in This represents a fixed-order encoding function. Indicates candidate named entity mentions The title path code of the sentence unit. Indicates candidate named entity mentions The footnote association code of the sentence unit, Indicates candidate named entity mentions Role-restricted identifier code, Indicates candidate named entity mentions The stage code of the sentence unit. Indicates candidate named entity mentions The speech unit code of the sentence unit it belongs to; S3-4: Write the current set of valid structure scope fields, the set of pending entity records, the set of confirmed specification entity records, and the set of explicit rebinding relationships into the inter-block state transfer data packet. .

5. The named entity consistency correction method for long text block alignment as described in claim 1, characterized in that, S4 specifically includes: S4-1: For the pending entity record passed from the previous text block. With the pending entity records in the current text block Calculate the structural range compatibility coefficient When the structural range compatibility coefficient When the value is 1, it indicates that the two corresponding undetermined entity records meet the prerequisites for entering the subsequent similarity comparison and association score calculation. S4-2: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Alias ​​similarity between ,in, Indicates a candidate named entity reference in the previous text block. A collection of aliases Indicates candidate named entity references for the current text block. n A collection of aliases This indicates the number of elements in the intersection of alias sets. Indicates the number of elements in the union of alias sets; S4-3: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Descriptive similarity between ,in, Indicates a candidate named entity reference in the previous text block. The set of descriptive words, Indicates candidate named entity references for the current text block. The set of descriptive words, This indicates the number of elements in the intersection of the sets of descriptors. Represents a set of descriptive terms The number of elements, Represents a set of descriptive terms The number of elements; S4-4: Calculate the correlation score according to the following formula. ; S4-5: Calculate the pending entity records passed from the previous text block. With the pending entity records in the current text block Conflict count ; S4-6: Let the association threshold be... ,when At that time, determine the pending entity records of the previous text block. Pending entity records in the current text block There is a conflict that blocks the merge; the pending entity will be recorded. and Maintain separation and record the undetermined entities in the current text block. status codes Set to 2; when and When determining the pending entity record Records of entities to be determined The association conditions are met; if the pending entity record status codes If the value is 1, then the entity to be determined will be recorded. Associated with the pending entity record The same canonical entity identifier, and record the entity to be determined. status codes Set to 1; if the entity record is pending. status codes If the value is 0, then the undetermined entity will be recorded. Records of entities to be determined The records are associated with the same pending entity record, and the pending entity record is linked. status codes Maintain at 0; when and When determining the pending entity record The association conditions have not yet been met; therefore, the entity record is retained pending. The pending state and its status code. Maintain at 0.

6. The named entity consistency correction method for long text block alignment as described in claim 1, characterized in that, S5 specifically includes: S5-1: When an explicit rebinding statement is detected At that time, generate explicit rebinding statements. Corresponding standardized entity identifier And record the starting position of the explicit rebinding statement. ; S5-2: Determine the explicit rebinding statement Valid text structure range , It is the starting position of the explicit rebinding statement. It is the end position of the effective text structure range; S5-3: For any undetermined entity record When the position index relationship is satisfied simultaneously Pending entity records Collection of aliases Includes explicit rebinding statements The re-bound entity surface form Or its normalized surface form, title path compatibility function Footnote association chain contains functions and pending entity records With explicit rebinding statement Conflict count At that time, record the entity to be determined. The canonical entity identifier is determined as an explicit rebinding statement. Corresponding standardized entity identifier And record the entity to be determined. status codes Set to 1; Indicates a record of an entity pending determination. The corresponding candidate named entity references their position index in the full text. Indicates candidate named entity mentions The title path code of the sentence unit. Indicates an explicit rebinding statement The title path code of the sentence unit. Indicates candidate named entity mentions The footnote association code of the sentence unit, Indicates an explicit rebinding statement The footnote association code of the sentence unit; S5-4: For those not falling within the range of valid text structure For pending entity records, retain the association merging result formed in S4. For pending entity records that do not meet the explicit rebinding condition or the association condition in step S4, output them as independent entities and set the status code to 2. S5-5: Output calibration result set ,in, This indicates the constructor for the result set. Indicates a record of an entity pending determination. The original surface form mentioned by the corresponding candidate named entities, This indicates the normalized surface form of candidate named entity references after normalization of capitalization, punctuation, and appellation. Indicates a record of an entity pending determination. The final corresponding canonical entity identifier, Indicates a record of an entity pending determination. The corresponding structural range identifier, Indicates a record of an entity pending determination. The block number to which it belongs Indicates a record of an entity pending determination. The corresponding candidate named entity references their position index in the full text. Indicates a record of an entity pending determination. The final status code, Indicates a record of an entity pending determination. The source of the correction is marked.

7. The named entity consistency correction method for long text block alignment as described in claim 5, characterized in that, In S4-1, the structural range compatibility coefficient The calculation formula is as follows: in, This indicates a function for compatibility with title paths. Indicates a candidate named entity reference in the previous text block. The title path code of the sentence unit. Indicates candidate named entity references for the current text block. The title path code of the sentence unit, when In the position of When it is on the subsequent title chain of the root, The value is 1 if it is set to 1, otherwise the value is 0. This indicates that the footnote association chain contains functions. Indicates a candidate named entity reference in the previous text block. The footnote association code of the sentence unit, Indicates candidate named entity references for the current text block. The footnote association code of the sentence unit, when Include When used as a prefix association chain The value is 1 if it is set to 1, otherwise the value is 0. Functions indicating equality of role-restricted identifiers. Indicates a candidate named entity reference in the previous text block. Role-restricted identifier code, Indicates candidate named entity references for the current text block. Role-restricted identifier code, when and If they are the same, then The value is 1 if it is set to 1, otherwise the value is 0. Indicates the function for stage code equality. Indicates a candidate named entity reference in the previous text block. The stage code of the sentence unit. Indicates candidate named entity references for the current text block. The stage code of the sentence unit, when and If they are the same The value is 1 if it is set to 1, otherwise the value is 0. Functions indicating equality of speech unit codes. Indicates a candidate named entity reference in the previous text block. The speech unit code of the sentence unit it belongs to. Indicates candidate named entity references for the current text block. The speech unit code of the sentence unit, when and If they are the same The value is 1 if it is not 1, otherwise the value is 0.

8. The named entity consistency correction method for long text block alignment as described in claim 5, characterized in that, In S4-5, the collision counting The calculation formula is as follows: in, This indicates an explicitly defined conflict indicator. It takes the value of 1 when the same alias is explicitly bound to different canonical entities within the comparison interval, and otherwise takes the value of 0. Indicates the number of title path conflict indicators, when and The value is 1 if the elements do not belong to each other's subsequent title chains, and 0 otherwise. Indicates a candidate named entity reference in the previous text block. The title path code of the sentence unit. Indicates candidate named entity references for the current text block. The title path code of the sentence unit; Indicates the phase conflict indicator quantity, when and If they are different, the value is 1; otherwise, the value is 0. Indicates a candidate named entity reference in the previous text block. The stage code of the sentence unit. Indicates candidate named entity references for the current text block. The stage code of the sentence unit it belongs to; Indicates the conflict indicator quantity of the speaking unit, when and If they are different, the value is 1; otherwise, the value is 0. Indicates a candidate named entity reference in the previous text block. The speech unit code of the sentence unit it belongs to. Indicates candidate named entity references for the current text block. The speech unit code of the sentence unit.

9. The named entity consistency correction method for long text block alignment as described in claim 6, characterized in that, In S5-2, the formula for calculating the end position of the effective text structure range interval is as follows: in, Indicates the starting position The next heading boundary at the same or higher level. Indicates the starting position The next stage involves switching positions. Indicates the starting position The next speaking unit will switch positions.

10. A named entity consistency correction system for long text block alignment, characterized in that, include: The text structure range parsing and structure boundary segmentation module is used to read long input text, perform text structure range parsing on the long input text, generate a sentence-by-sentence structure annotation sequence, determine the segmentation boundary based on the sentence-by-sentence structure annotation sequence, and generate text segments. The module for generating pending entity records and transmitting inter-block status is used to extract candidate named entity mentions in each text block, construct pending entity records for each candidate named entity mention, write the pending entity records into the inter-block status transmission data packet, and transmit the inter-block status transmission data packet between adjacent text blocks. The cross-block association determination and reverse update module is used to match the pending entity records in the current text block with the pending entity records passed from the previous text block. For pending entity records with conflicts, they are kept separate. For pending entity records without conflicts and that meet the association conditions, association merging is performed. For pending entity records without conflicts but that do not meet the association conditions, they are kept in a pending state. When an explicit rebinding statement appears in the following text of the current text block, the effective text structure range of the explicit rebinding statement is determined. Within the effective text structure range, the pending entity records that meet the reverse update conditions are reverse updated to determine the canonical entity identifier. For pending entity records that are not within the effective text structure range but have met the association conditions, their association merging results are retained. Finally, the correction result set is output.

Images