Method and system for structuring decoding of enterprise strategic goals based on semantic space mapping

Abstract

The application provides a semantic space mapping-based enterprise strategy target structured decoding method and system, obtains standardized policy terms and numerical expressions through paragraph level and syntax analysis to generate unique semantic anchor points; utilizes context snapshots to realize anchor point positioning and migration of dynamic change areas of the text; combines a lightweight domain language model and multi-granularity similarity matching to determine semantic compensation and index derivation relationships, automatically complete the context chain of new anchor points, inherits the attributes of historical anchor points to form a longitudinal consistent incremental key index set, and synchronously updates the strategy target decoding result, so that the application can efficiently and accurately realize cross-version tracing and structured output of key indexes, and improve data integrity and semantic coherence in the strategy text evolution process.

Description

Technical Field

[0001] This invention relates to the field of structured semantic anchor tracking and indicator evolution management technology of strategic texts, and in particular to a method and system for structured decoding of enterprise strategic objectives based on semantic space mapping. Background Technology

[0002] In the field of structured decoding of enterprise strategic objectives, strategic texts are frequently iterated as business adjustments are made, and the key indicators added, changed, and deleted in these texts are becoming increasingly complex. Traditional methods typically employ full-text re-parsing, unified vectorization, or global knowledge graph reconstruction. Each version change requires re-semantic modeling and indicator extraction of the entire document, resulting in high computational overhead and high response latency, making it difficult to meet the real-time requirements of high-frequency update scenarios. Existing technologies lack a rapid evolution capture mechanism for local changes, making it impossible to locate the changed areas of key indicators without re-parsing the entire text. This leads to a large amount of redundant computation and low processing efficiency, restricting the practicality and scalability of dynamic monitoring of enterprise-level strategic objectives. In the process of incremental extraction across versions, existing solutions lack the ability to explicitly model semantic evolution, making it difficult to effectively maintain the semantic coherence of local contexts. When complex evolutions such as term replacement, condition expansion, or indicator derivation occur in strategic texts, knowledge graph or semantic retrieval methods are prone to node drift and logical chain breaks, leading to the omission of newly added key indicators and the misjudgment of the same indicators as new or missing due to minor adjustments in expression. The decoding results lack vertical consistency and historical inheritance relationships, making it impossible to accurately determine the evolution types such as semantic equivalent replacement, condition expansion, or indicator derivation, thereby affecting the comparability and accuracy of strategic target tracking. Furthermore, while technical approaches relying on large-scale model-based re-inference or complex question-and-answer screening possess some incremental processing capabilities, they heavily depend on global semantic computation and large-scale parameter calls. The process is opaque, lacks interpretability, and fails to provide clear audit trails and justifications for changes, thus failing to meet regulatory requirements for corporate compliance review and decision-making traceability. Existing solutions offer a rather crude attribution of semantic changes and lack a lightweight processing paradigm centered around highly stable semantic anchors, making it difficult to balance efficiency and reliability in the dynamic evolution of decoding results. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides a method and system for structured decoding of enterprise strategic objectives based on semantic space mapping.

[0004] The technical solution of this invention is implemented as follows: a structured decoding method for enterprise strategic objectives based on semantic space mapping, comprising: S1: Obtain the initial version of the strategic text, and based on the characteristics of high structural stability, low ambiguity and strong domain reference capability, identify standardized policy terms, indicator phrases with constraints and nested numerical expressions, and generate several semantic anchors with unique identifiers.

[0005] S2: Based on the semantic anchors and their paragraph levels, adjacent syntactic stems and preceding and succeeding logical connectors, construct a compact structured tuple containing context offsets and timestamps to generate an initial context snapshot.

[0006] S3: Obtain the dynamically evolved new version of the strategy text and locate the changed areas. Map the semantic anchors registered in the old version to the corresponding positions in the new version, and generate anchor migration results containing the mapping success status or mapping failure status.

[0007] S4: For the state of mapping failure in the anchor point migration result, call the lightweight domain-adaptive language model to perform a local scan on the preset character window around the semantic anchor point to generate a semantic summary to be compared.

[0008] S5: Based on the semantic summary to be compared and the context summary in the historical context snapshot, perform multi-granularity similarity matching of keyword co-occurrence, consistency of modification relationship and retention of logical connector, and generate semantic compensation conclusions that are determined to be semantic equivalent replacement, condition expansion or index derivation.

[0009] S6: Based on the semantic compensation conclusion, the newly added semantic anchor point is confirmed, a new context snapshot is automatically generated, and a bidirectional semantic association edge is established with the adjacent existing semantic anchor points to generate a context snapshot chain that supports cross-version backtracking.

[0010] S7: Based on the context snapshot chain, trace the nearest comparable historical anchor point of the newly added semantic anchor point upwards along the chain, inherit its semantic role definition, measurement caliber description and target constraints, and generate an incremental key indicator set with vertical consistency.

[0011] S8: Update the strategic objective decoding output using the incremental key indicator set, and complete the structured decoding of the key indicator changes in the dynamically evolving strategic text without re-parsening the full text.

[0012] The present invention also provides a structured decoding system for enterprise strategic objectives based on semantic space mapping, which uses the above-mentioned structured decoding method for enterprise strategic objectives based on semantic space mapping to perform structured decoding of enterprise strategic objectives.

[0013] The structured decoding method and system for enterprise strategic objectives based on semantic space mapping provided by this invention have the following beneficial effects: (1) The key indicator incremental extraction method proposed in this invention effectively overcomes the high computational overhead and low response efficiency caused by the reliance on full-text re-parsing in the traditional strategic goal decoding process. By introducing a semantic anchor mechanism, a unique identifier is established and its context snapshot is fixed only for the smallest semantic unit with high stability and strong domain reference capability. This allows the system to quickly locate the changed area without traversing the entire text when a new version arrives, and realizes intelligent migration and status determination of the anchor based on local comparison. This lightweight evolution perception architecture significantly reduces processing latency, avoids repeated document-level coding and reasoning processes, and greatly improves the response speed and resource utilization efficiency of indicator extraction. It is especially suitable for enterprise management environments with short strategic planning cycles and high revision frequency. (2) This invention achieves reliable maintenance of semantic coherence and vertical consistency of key indicators across versions by constructing a context snapshot chain and a semantic compensation mechanism. It solves the problems of information breakage, misjudgment and omission that are prone to occur in existing methods when facing complex semantic evolution such as term replacement, condition expansion or indicator derivation. In anchor lifecycle management, a multi-granularity similarity matching strategy is introduced. Combining the dimensions of keyword co-occurrence, consistency of modification relationship and retention of logical connectors, the semantic equivalent replacement and condition evolution path are accurately identified. For new anchors, a snapshot chain is formed by linking to existing anchors through bidirectional semantic association edges. This supports tracing back to the most recent comparable historical nodes and inheriting their measurement caliber, constraints and role definitions, thereby ensuring the comparability and continuity of key indicators in the time dimension. This mechanism not only enhances the robustness and accuracy of the system, but also provides a clear tracing path and an interpretable evolution record for subsequent target tracking and performance evaluation. (3) This invention abandons complex and uncontrollable technical paths such as relying on large-scale model re-inference, triple fusion, or parent-child document tree matching. Instead, it adopts a compact tuple storage and localized processing paradigm centered around semantic anchors, which has high interpretability, strong audit traceability, and good engineering feasibility. The entire process does not involve full document vector retrieval or large-scale parameter calls. All operations focus on predefined anchors and their neighborhood windows. The calculation boundaries are clear and the process is transparent, which facilitates the provision of complete behavior logs and change evidence in enterprise compliance review and decision tracing scenarios. At the same time, this design naturally supports modular expansion and domain adaptation. It can be quickly migrated to different industries or strategic systems by configuring anchor dictionary and context template without retraining the model or reconstructing the knowledge structure. In particular, in the process of supporting the layer-by-layer decoding of enterprise strategic goals from macro vision to specific KPIs, this method can continuously capture the dynamic evolution trajectory of indicators, realize the automatic updating and inheritance of key data without interrupting business flow, and effectively build an efficient, stable, and reliable incremental target decoding system. Attached Figure Description

[0014] Figure 1 This is a flowchart of the enterprise strategic goal structure decoding method based on semantic space mapping of the present invention; Figure 2 This is a sub-flowchart of the semantic space mapping-based structured decoding method for enterprise strategic objectives of the present invention; Figure 3 This is another sub-flowchart of the semantic space mapping-based structured decoding method for enterprise strategic objectives of the present invention. Detailed Implementation

[0015] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0016] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0017] like Figure 1 As shown, this invention provides a structured decoding method for enterprise strategic objectives based on semantic space mapping, specifically including: S1: Obtain the initial version of the strategic text, and based on the characteristics of high structural stability, low ambiguity and strong domain reference capability, identify standardized policy terms, indicator phrases with constraints and nested numerical expressions, and generate several semantic anchors with unique identifiers. S2: Based on the semantic anchors and their paragraph levels, adjacent syntactic stems and preceding and succeeding logical connectors, construct a compact structured tuple containing context offsets and timestamps to generate an initial context snapshot; S3: Obtain the dynamically evolved new version of the strategy text and locate the changed area, map the semantic anchors registered in the old version to the corresponding positions in the new version, and generate anchor migration results including mapping success status or mapping failure status. S4: For the state of mapping failure in the anchor point migration result, call the lightweight domain-adaptive language model to perform a local scan of the preset character window around the semantic anchor point to generate a semantic summary to be compared. S5: Based on the semantic summary to be compared and the context summary in the historical context snapshot, perform multi-granularity similarity matching of keyword co-occurrence, consistency of modification relationship and retention of logical connector, and generate semantic compensation conclusions that are determined to be semantic equivalent replacement, condition expansion or index derivation. S6: Based on the semantic compensation conclusion, the newly added semantic anchor point is confirmed, a new context snapshot is automatically generated and a bidirectional semantic association edge is established with the adjacent existing semantic anchor points to generate a context snapshot chain that supports cross-version backtracking; S7: Based on the context snapshot chain, trace the nearest comparable historical anchor point of the newly added semantic anchor point upwards along the chain, inherit its semantic role definition, measurement caliber description and target constraints, and generate an incremental key indicator set with vertical consistency. S8: Update the strategic objective decoding output using the incremental key indicator set, and complete the structured decoding of the key indicator changes in the dynamically evolving strategic text without re-parsening the full text.

[0018] Step S1: Obtain the initial version of the strategic text, and based on the characteristics of high structural stability, low ambiguity, and strong domain referential ability, identify standardized policy terms, constraint-bound indicator phrases, and nested numerical expressions to generate several semantic anchors with unique identifiers. Specifically, this includes: S1.1: Perform paragraph-level division and syntactic tree parsing on the initial version of the strategic text to extract the syntactic backbone units containing leading logical connectors, core predicate verbs, and subsequent modifiers, thereby obtaining a set of original semantic fragments with clearly defined boundaries; The initial version of the input strategic text is processed to perform paragraph hierarchical segmentation to identify the depth level number and parent node path structure of each paragraph, forming a hierarchical data set with a unique topological location from the original text. Dependency syntactic tree parsing is performed on sentences in the segmented paragraphs to extract the predicate verb as the core node. At the same time, the leading logical connectors appearing at the beginning of the sentence and the subsequent modifiers connected to the core predicate verb are identified to form a set of syntactic backbone structures. For each unit of the syntactic core structure set, grammatical boundary labeling is performed, and the start and end boundaries of the core predicate and its modification path are determined based on the dependency relationship, thereby fixing the minimum coverage of the original semantic fragment; Perform structural closure verification on the semantic segments with marked boundaries to ensure that the segments contain complete logical chains and modifying semantics, without syntactic breaks or missing data, and generate a set of verified original semantic segments. The original semantic fragment set is serialized and encapsulated, and the paragraph level number, core predicate verb, preceding logical connector and subsequent modifier are bound into indexable structured fragment objects for subsequent domain dictionary matching and pattern scanning. By using paragraph hierarchical division and syntactic tree parsing, the initial version of the strategic text is transformed into a set of original semantic fragments with clearly defined boundaries, thereby achieving stable decomposition and standardized description of the strategic text in terms of structural hierarchy and syntactic skeleton. For example, in the corporate annual strategic planning text, the input data contains 12 paragraphs, each containing an average of 4 sentences. The paragraph hierarchy division algorithm sets the maximum hierarchy depth to 5, and the parent node path identifier is represented in the format "Lx.yz", for example, the first paragraph of the second section of the first chapter is "L1.2.1". Dependency parsing uses a custom domain dependency model, identifying the predicate verb "elevate" as the core node, the sentence-initial logical connector "to achieve" as the preceding connector, and the subsequent modifier "to more than 5%" as the target constraint. Syntax boundary annotation locks the fragment's start position at character index 0 and the end position at character index 45, forming an original semantic fragment of length 46. Structural closure verification confirms that the logical chain in this fragment is continuous and unbroken. The serialization encapsulation module combines the fragment's attributes into a structured object {hierarchy:"L1.2.1", predicate:"elevate", preceding:"to achieve", succeeding:"to more than 5%"}. The results show that the method can generate 12×4=48 original semantic fragments with unique topological positions and complete syntactic skeletons, providing a clear and stable structural foundation for subsequent domain dictionary matching; S1.2: Based on the original set of semantic fragments, perform domain dictionary matching and regular pattern scanning to filter out indicator phrases that conform to the standardized policy terminology definition, have numerical constraints, and numerical expressions nested in fixed templates, thereby generating a list of candidate highly stable semantic units. Based on the original semantic fragment set output from step S1.1, the domain dictionary retrieval interface is called to load a structured term database containing standardized policy terms, key industry indicator phrases, and commonly used numerical templates. A highly stable feature matching mechanism is used to compare each input fragment to generate term matching records that conform to the dictionary definition. The regular expression compiler is used to construct a set of pattern expressions for numerical constraints, including unit system matching patterns, range expression patterns, and time limit patterns. The input fragment is scanned to capture index phrases that match the numerical constraints and to record their matching offset positions and matching group contents. For numeric expressions nested in fixed templates, the marked boundaries of the template structure are parsed, the numeric fields within the template are extracted, and their format is verified to conform to the domain standard numeric expression definition through pattern matching. The verified expressions are then encapsulated as independent candidate units. The domain dictionary matching records are cross-merged with the regular pattern scanning results, and fragments that do not simultaneously meet the semantic stability criteria and numerical constraints are removed to generate a temporary list of candidate high-stability semantic units. Perform uniqueness checks and format standardization on the temporary list, merge duplicate matches, unify the encoding rules for numerical values ​​and terms, and output a list of candidate highly stable semantic units with consistent structure. Through the above chain processing method, the original set of semantic fragments is transformed into a list of candidate highly stable semantic units that have been verified by both domain knowledge and patterns, thereby achieving stable capture of key semantic units and preparation of input for subsequent ambiguity evaluation. For example, in a corporate strategy text scenario, the original semantic fragment set contains three fragments extracted from syntactic analysis: "Increase R&D investment intensity to over 5% within three years," "Move forward the carbon emission peak target to 2030," and "Increase by 1.2 percentage points compared to the previous year." The domain dictionary contains standardized policy term entries "carbon emission peak" and "R&D investment intensity," and numerical template definitions include percentage expressions and percentage point expressions. When performing dictionary matching, both "carbon emission peak" and "R&D investment intensity" are matched in the fragments, generating matching records. Regular expression pattern scanning uses a fixed template to match percentage formats ≥0.0%, matching the numerical condition in "Increase R&D investment intensity to over 5% within three years" and the percentage point value in "Increase by 1.2 percentage points compared to the previous year." Nested numerical template matching locates the time constraint "Move forward to 2030" and conforms to the fixed template year pattern [1900-2100]. The filtering logic merges dictionary matches and regular expression results, retaining fragments that contain both domain terms and numerical conditions, and removing fragments without numerical constraints. Uniqueness checks and format standardization unify "1.2 percentage points" into the numerical unit "percentage point". The output list of candidate highly stable semantic units is {carbon emission peak @ 2030, R&D investment intensity @ >= 5%, growth @ 1.2 percentage points}, which provides consistent input for subsequent ambiguity quantification and domain reference intensity calculation. S1.3: Perform ambiguity quantification evaluation and domain reference strength calculation on the candidate high-stability semantic unit list to eliminate polysemous words and retain effective data items with low ambiguity and strong domain reference ability, thereby outputting a verified pure semantic anchor material set; S1.4: Based on the pure semantic anchor material set, perform global unique identifier hash encoding and metadata binding processing to establish a two-way mapping relationship from text content to system internal index, thereby generating a structured semantic anchor object carrying a unique identity. S1.5: Perform spatial location registration and attribute encapsulation processing on the structured semantic anchor object to solidify its paragraph level coordinates and adjacent syntactic environment information in the initial version of the strategic text, thereby finally outputting a semantic anchor with a unique identifier that can be used for migration comparison in subsequent versions.

[0019] Step S2: Based on the semantic anchors and their paragraph levels, adjacent syntactic stems, and preceding and succeeding logical connectors, construct a compact structured tuple containing context offsets and timestamps to generate an initial context snapshot. Specifically, this includes: S2.1: Perform hierarchical parsing on the original strategic text paragraph where the semantic anchor point is located to extract the paragraph depth number and parent node path identifier, and generate paragraph hierarchical index data with a unique topological position; The generated semantic anchor objects with unique identifiers and their corresponding paragraph texts are used as input. The paragraph-level parsing rule base is loaded to identify the paragraph delimiter type in advance. During the parsing process, the text segmentation operator is called to physically segment the original strategic text to form a paragraph segmentation data stream. The paragraph segmentation data stream is processed by depth calculation. A hierarchical depth mapping table is used to map each paragraph to a predefined depth numbering system. The parent node index of the paragraph in the whole text is exported as the parent node path identifier according to the tree structure path generation algorithm to ensure the unique topological location traceability of the paragraph in multiple versions of the text. The paragraph depth number and the parent node path identifier are cross-validated for consistency. The structural integrity check module is called to verify that each node in the path identifier can be matched in the current text structure tree. If a broken chain is detected, the missing node is repaired and filled in through the path reconstruction algorithm to ensure the integrity of the paragraph level index. The paragraph depth number and the repaired parent node path identifier are combined into a structured index tuple. The tuple is then encoded into a compact data format using a serialization encapsulation algorithm and bound to the corresponding semantic anchor unique identifier, thereby achieving a bidirectional searchable mapping between the anchor and the paragraph level position. By parsing and processing paragraph levels and generating path identifiers, the semantic anchor point location data from the previous step is transformed into paragraph level index data with unique topological locations, achieving the expected technical effect of precise paragraph location in subsequent syntactic analysis and context feature capture. For example, in the initial version of a corporate strategy text, a semantic anchor is located in paragraph 2 of the third level. The parsing rule base is loaded to identify that this paragraph is delimited by a newline character and a number, and its depth number is calculated to be 3. The parent node path identifier is output as "1-2" using a tree structure path generation algorithm, where "1" represents the root directory item number and "2" represents the second subdirectory item number. During consistency verification, the path identifier is found to be complete and without broken links. The depth number 3 and the path identifier "1-2" are directly combined into an index tuple and bound to the unique identifier of the semantic anchor "A104F". This tuple is serialized, encapsulated, and compressed to a length of 64 bytes, and bidirectional indexing between the semantic anchor and the paragraph level is achieved through binding. Subsequently, in S2.2, the system can quickly retrieve the paragraph level information where the "A104F" anchor is located, ensuring that syntactic analysis is performed at the precise paragraph position, significantly improving the accuracy of dependency syntax feature capture. S2.2: Based on the paragraph-level index data, perform dependency parsing on the sentence components around the semantic anchor to identify the predicate verb core and adjacent logical connectors that directly modify the semantic anchor, and generate a syntactic backbone feature vector representing the local grammatical structure. Based on paragraph-level index data input, the dependency parser is invoked to perform structured dependency relation parsing on sentence components around semantic anchors, generating an initial dependency relation graph containing subject-verb-object dependency paths and modifying components. Locate semantic anchor nodes in the dependency graph and retrieve all predicate verb nodes that directly depend on these nodes. Select core predicate verbs according to the dependency edge type to construct the core predicate part of the syntactic trunk. A connector recognition algorithm is executed on the vicinity of the predicate core node. Based on the matching of lexical features and the logical template library, the causal, progressive, and adversative logical connector nodes of the preceding and following parts are extracted as the logical extension of the syntactic trunk. The predicate core node and logical connector node are encoded in the order of dependency relationship to form a serialized syntactic skeleton chain, and the part-of-speech tags and semantic categories of the nodes are combined to generate syntactic backbone feature vectors. A vectorized encoder is used to map the grammatical skeleton chain to a fixed-length numerical feature representation, ensuring that the feature vectors can be used for subsequent context offset calculation and metadata encapsulation. Through the above dependency parsing and feature encoding processing, the paragraph-level index data of the previous step is transformed into syntactic backbone feature vectors that represent local syntactic structures, thereby realizing a quantifiable representation of the logical structure of semantic anchors in the context. For example, in a paragraph of a corporate strategy text, with a paragraph level number of 3.2 and a parent node path identifier of "1-3", the semantic anchor is "R&D investment intensity". The dependency parser uses a maximum entropy model to identify the direct predicate of this anchor as "improve", and the logical connectors as the preceding "to achieve" and the following "and maintain". The predicate core node index in the dependency graph is V5, the preceding connector node index is C1, and the following connector node index is C9. A skeleton chain [C1→V5→C9] is generated according to the dependency path order, with part-of-speech tags [CONJ, VERB, CONJ] and semantic categories [causal, action, progression]. The encoder maps the skeleton chain to a 32-dimensional feature vector, where the 5th dimension corresponds to the semantic strength of the predicate core, and the 12th dimension corresponds to the logical type distribution of the following connectors. In the offset calculation, this feature vector is used as input, and the predicate core strength is calculated using the following formula: ; in For the core strength of the predicate, This is the predicate weight coefficient. These are the component values ​​of the feature vector corresponding to the predicate. This represents the paragraph level depth value. The calculation result is used for offset quantization in subsequent context snapshots. Verification shows that this processing can significantly improve the accuracy of anchor semantic structure recognition, and the output syntactic backbone feature vector maintains high consistency in global alignment. S2.3: Calculate the distance difference between the start and end positions of the semantic anchor point within the paragraph and the first character of the paragraph using the syntactic backbone feature vector, so as to quantify the relative coordinates of the anchor point in the local context and generate the context space offset parameter. S2.4: Obtain standard timestamp information based on the release time of the current strategic text version, and serialize and encapsulate the context space offset parameter, syntactic backbone feature vector and paragraph level index data to form an initial structured tuple object with spatiotemporal attributes; S2.5: Perform a hash digest operation on the initial structured tuple object and bind a version identifier to generate an immutable and fast-retrieval-enabled initial context snapshot, completing the transformation from dynamic text to static baseline data units.

[0020] like Figure 2 As shown, step S3 involves: acquiring the dynamically evolved new version of the strategic text and locating the changed areas; mapping the semantic anchors registered in the old version to the corresponding positions in the new version; and generating anchor migration results containing either a successful mapping status or a failed mapping status. Specifically, this includes: S3.1: Calculate paragraph-level hash fingerprints for the dynamically evolved new version of the strategic text, and perform differential comparison processing with the paragraph-level hash fingerprints of the initial version of the strategic text to generate a set of coordinates of the changed areas that identify the locations of text additions, deletions, and modifications. Based on the generated initial context snapshot and paragraph-level index data, the dynamically evolved new version of the strategy text is used as input, and paragraph-level structural feature quantification is performed to support version difference detection. The new strategic text is segmented into paragraphs according to predefined hierarchical parsing rules, and the depth number, parent node path identifier, and character sequence of each paragraph are extracted. Standardized encoding operators are used to perform uniform encoding format conversion on paragraph character sequences, removing redundant spaces and invisible characters to ensure consistency of hash calculations; A secure hash algorithm is used to generate hash fingerprints for each segment encoding sequence. For segments with segment depth and parent node path identifiers, a composite hash key containing hierarchical information is constructed to enhance the local stability of the fingerprint. The new paragraph-level hash fingerprint set is compared with the initial paragraph-level hash fingerprint set by differential comparison. During the comparison, the consistency of hash values ​​is checked sequentially according to the paragraph depth-first and parent node path matching strategies. For paragraphs with mismatched hash values ​​in the detection results, add, delete, and modify markers are created, and a set of coordinates of the changed area is generated by distinguishing the marker types of newly added paragraphs, deleted paragraphs, and paragraphs with modified content. The set of coordinates of the changed area generated by differential comparison transforms the result of the previous step into precise spatial positioning data for local text extraction and semantic anchor mapping, achieving the expected technical effect of low-latency version difference awareness. For example, in one embodiment, the new strategic text is divided into 12 paragraphs, each with a depth number and a parent node path identifier. The encoding conversion uses the UTF-8 standard format, removing all tabs and consecutive spaces. The paragraph hash fingerprint is generated using the SHA-256 algorithm, where the character sequence of each paragraph and its level identifier are concatenated to form the input, resulting in a 256-bit hash value. During differential alignment, the text is traversed sequentially from depth 0 to 3, comparing the hash values ​​with paragraphs of the same depth in the initial version. Newly added paragraphs have no matches in the initial set, deleted paragraphs have no matches in the new version set, and modified paragraphs have different hash values ​​but the same level path. For the coordinate set of version difference markers, the coordinate range for newly added paragraphs is from the start position 0 to the end position 85, for deleted paragraphs it is from the start position 102 to the end position 164, and for modified paragraphs it is from the start position 200 to the end position 275. In S3.2, local text extraction and semantic anchor index matching will be performed on this set. During the verification process, the accuracy of version difference identification is significantly improved, ensuring that the system can still quickly and accurately locate the changed area under high-frequency strategic text iteration. S3.2: Extract local text fragments based on the set of coordinates of the changed region, and perform feature vector matching operation on the local text fragments using a pre-built semantic anchor index library to generate a candidate semantic anchor mapping list; S3.3: Perform unique identifier consistency verification and context offset tolerance verification on each candidate in the candidate semantic anchor mapping list to generate an anchor mapping record to be confirmed containing precise location information; For the candidate semantic anchor mapping list output by step S3.2, load the anchor index data containing the unique identifier field and the context offset parameter as the input object for the verification process; Perform a hash value comparison operation between the unique identifier of each candidate and the corresponding identifier in the old version of the registered semantic anchor index, and use a binary bit-by-bit matching method to determine the consistency of the unique identifier and record the verification status. For candidates that pass the verification, extract the character start and end positions calculated in the new version of the strategic text, construct a context offset data group relative to the corresponding paragraph level, and perform difference calculation with the offset parameters stored in the old version of the context snapshot. Using a tolerance verification algorithm, the offset difference is compared with a preset tolerance threshold. The threshold is set based on the average character length of the paragraph and the stability weight of the syntactic core. If the difference is within the threshold range, it is marked as a precise alignment state. The offset difference is calculated using the following formula: ; in, This is the offset of the old context snapshot. For the new context offset, The absolute value of the difference. The unique identifier consistency check status and the context offset tolerance check result are fused to generate a pending anchor point mapping record containing precise location information and check status markers; The output of this step is structured anchor point mapping record data, which realizes the dual confirmation of the uniqueness and spatial alignment of candidate mapping items, and provides accurate input for the generation of anchor point migration results in S3.4; For example, in the version migration detection of corporate strategy text, the candidate semantic anchor "carbon intensity reduced by 5%" has an offset parameter of 120 characters in the old version and 124 characters in the new version, and the unique identifier hash value matches perfectly. The tolerance threshold is set to 6 characters, calculated based on an average paragraph length of 300 characters and a stability weight of 0.02. The difference of 4 characters is less than the tolerance threshold, and the position alignment status is determined to be passed. The anchor mapping record to be confirmed generated after merging the verification status contains the identifier "ANCHOR_00123", version coordinates "paragraph 2_sentence 5", offset 124, and alignment status TRUE. Verification shows that this record will directly enter the mapping success branch in S3.4, thus significantly improving the version migration comparison accuracy and low-latency response performance. S3.4: Based on the verification pass status or verification failure status of the anchor point mapping record to be confirmed, associate the semantic anchor points registered in the old version with the new version text coordinates or mark them as unmatched objects to generate anchor point migration results containing mapping success status or mapping failure status. Based on the verification pass status and verification failure status of the anchor point mapping record to be confirmed as input conditions, the unique identifier mapping result and the context offset tolerance verification result are loaded into the version mapping processing unit. The verified anchor mapping records will be bound to a position. The unique identifier of the semantic anchor object registered in the old version will be linked to the paragraph level coordinates and character start and end positions of the new strategic text to solidify its spatial positioning in the new text. The anchor point mapping records that fail the verification are marked, the status identifier generation algorithm is called to assign them unmatched status labels, and the labels are bound to unique identifiers and historical context snapshots to ensure that they can be identified by subsequent compensation analysis during the evolution process; Batch serialization processing is performed on the set of anchor points with successful mapping status, and the version identifier, spatial coordinate parameters, timestamp information and context association metadata are encapsulated into structured migration objects to support fast retrieval and cross-version association; Perform a difference segmentation operation on the set of anchor points in the mapping failure state, and divide them into the unmatched object pool according to paragraph level and semantic category, so that the local scanning module can be called in a targeted manner in the subsequent semantic compensation step; By using location binding and status marking, the mapping records to be confirmed generated in the previous step are transformed into anchor migration results containing mapping success or mapping failure status, so as to achieve the structured association or difference marking effect of old semantic anchors in the new strategic text. For example, in an enterprise strategic planning system, the initial version of the text contains 30 registered semantic anchors. After differential comparison of the paragraph-level hash fingerprints of the new and old versions of the text, 12 mapping records to be confirmed are obtained. Among them, 8 pass the unique identifier consistency check, the context offset tolerance is set to ±15 characters, and the version identifier encoding rule is based on SHA-256 hash. The version mapping processing unit binds the 8 verified anchor objects to the coordinates of the new version of the text. For example, if the new position of an anchor has a paragraph-level index of "3-2", a starting offset of 120 characters, and an ending offset of 145 characters, a migration object is generated containing... The context offset parameter. Four anchors that fail validation are assigned the status label "NM" and stored in the unmatched object pool, pre-classified according to semantic categories such as "policy terminology" and "indicator phrase," for use in calling the local scanning module. In system retrieval tests, successfully mapped objects return spatial location results within 20 milliseconds, while unmapped objects significantly improve the completeness of new indicator identification during the compensation analysis phase, ensuring low latency and high interpretability in the incremental extraction process.

[0021] like Figure 3 As shown, step S4: For the mapping failure state in the anchor point migration result, a lightweight domain-adaptive language model is invoked to perform a local scan of the preset character window around the semantic anchor point to generate a semantic summary to be compared. Specifically, this includes: S4.1: Obtain the semantic anchor identifier of the mapping failure state and its position coordinates in the old version of the strategy text, calculate the corresponding local scanning region boundary in the new version of the strategy text based on the preset context window radius parameter, and generate a set of local text window coordinates containing the start offset and the end offset. S4.2: Based on the set of local text window coordinates, extract the original character sequence fragments from the new version of the strategic text, and use the text cleaning operator to perform noise filtering and format standardization processing on the original character sequence fragments to generate a clean local text data block with a unified encoding format; S4.3: Load the pre-trained lightweight domain-adaptive language model weight parameters, inject the clean local text data block as an input prompt sequence into the model encoder layer, perform forward propagation inference to capture the semantic feature vector representation in the local context, and generate a high-dimensional context-aware embedding tensor. The clean local text data blocks generated by text cleaning are subjected to input prompt sequence construction operation. The character sequence is encoded into a fixed-length embedding input matrix according to the preset domain semantic grouping rules to ensure that the lightweight domain-adaptive language model can accurately receive local context instructions. The embedded input matrix is ​​loaded into the input buffer area of ​​the model encoder layer, and the pre-trained weight parameters are called to initialize the weights of each multi-head self-attention unit and feedforward network of the encoder, so as to maintain the sensitivity and stability of the model in the domain context. In the encoder layer, the forward propagation computation process is triggered, and multi-head self-attention matching, residual connection and layer normalization are executed in sequence to capture the semantic dependencies across clauses and phrases in local text data blocks. The output sequence vector processed by multi-layer encoder stacking is fed into the high-dimensional feature aggregation module, and the embedding tensor with context awareness is generated by matrix multiplication and nonlinear activation function joint operation. We employ an embedding tensor storage operation to bind high-dimensional context-aware embedding tensors to corresponding semantic anchor identifiers for use in subsequent attention mechanism aggregation and semantic summarization generation. By loading a pre-trained lightweight domain-adaptive language model and performing the above processing, pure local text data blocks are transformed into high-dimensional context-aware embedding tensors that can represent contextual semantic relationships, thereby achieving compensatory feature capture effects after anchor semantic transfer failure. For example, in the enterprise strategic goal decoding system, a local scan segment within ±200 characters of the center position of the semantic anchor point where the mapping failed is selected from the new version of the strategic text. After removing format specifiers and meaningless markers by a text cleaning operator, a character sequence of length 512 is formed. The character sequence is mapped to a 512×256 input matrix using domain word vector encoding rules. Lightweight domain-adaptive language model encoder parameters are loaded. This encoder model contains four stacked layers, each with eight self-attention heads and a feedforward network with a parameter dimension of 256. During the forward propagation phase, the dot product correlation of the query, key, and value matrices is calculated and normalized within each self-attention head to obtain the correlation coefficient between any two character positions in the sequence. Subsequently, the sequence is processed by the non-linear activation function ReLU and residual connections to achieve feature preservation and gradient stabilization. The output sequence vector set of the encoder layer is subjected to mean pooling and concatenation operations to generate a 256-dimensional high-dimensional context-aware embedding tensor. In this embodiment, the embedded tensor is used to extract local semantic information of "R&D investment intensity" and its update conditions through an attention mechanism in subsequent steps. This process significantly improves the accuracy of anchor point compensation identification in the real-time update task of the system and controls the update response delay to the millisecond level. S4.4: Based on the high-dimensional context-aware embedding tensor, key semantic information is aggregated through an attention mechanism, and the aggregated semantic information is compressed and reconstructed using the sequence generation strategy of the decoder head to generate a candidate semantic summary sequence covering core indicator elements and logical constraint relationships. The high-dimensional context perception is embedded in the tensor input attention mechanism processing unit to calculate the correlation score matrix between each component vector of the embedded tensor and the global context reference vector. The relevance score matrix is ​​used to perform weight normalization to form an attention weight distribution for weighted aggregation; Based on the attention weight distribution, the vectors of the embedded tensor are weighted and summed to generate an aggregated set of semantic feature vectors. The aggregated set of semantic feature vectors is injected into the sequence generation module at the head of the decoder, and the candidate sequence of semantic segments is output step by step according to the set sequence generation strategy. A compression and reconstruction method is used to remove redundancy and increase information density in the candidate semantic fragment sequence, forming a candidate semantic summary sequence that covers core indicator elements and logical constraint relationships; By combining attention mechanisms and decoders, the high-dimensional embedding tensor from the previous step is transformed into a compact and semantically focused candidate summary sequence, enabling precise extraction and structured expression of key indicator information in the local context of the strategy text. For example, for the high-dimensional context-aware embedding tensor of the input, the global context reference vector is set to 64-dimensional units, and a scaled dot product attention mechanism is used when calculating the relevance score matrix, the formula of which is: ; in This is the correlation score matrix. For querying the matrix, The key matrix, with superscripts This indicates the transpose operation. Let be the dimension of the key vector. The resulting relevance score matrix is ​​normalized using Softmax to obtain the attention weight distribution. With a dimension of 64, the weights stabilize between 0.01 and 0.25. This weight is then used to perform a weighted sum operation on the embedding tensor: ; in, For value matrices, This is the aggregated semantic feature set. The decoder employs a Beam Search strategy with a beam width of 5, generating candidate semantic fragment sequences ranging from 8 to 15 tags in length. During the compression and reconstruction stage, low-frequency redundant modifications are removed through syntactic pattern matching, maintaining the integrity of the semantic chain. In performance verification, the candidate semantic summary sequences can completely retain core directive indicators such as "R&D investment intensity" and "increase to over 5%", accurately summarizing their contextual logical relationships. This significantly improves matching accuracy and completeness in subsequent multi-granularity similarity matching, while reducing processing latency. S4.5: Perform structured verification and redundancy removal operations on the candidate semantic summary sequence, filter non-critical descriptive information according to the domain terminology dictionary, and finally generate a standardized semantic summary object to be compared for subsequent multi-granularity similarity matching.

[0022] Step S5: Based on the semantic summary to be compared and the context summary in the historical context snapshot, perform multi-granularity similarity matching of keyword co-occurrence, consistency of modification relations, and retention of logical connectors to generate semantic compensation conclusions that are determined to be semantically equivalent substitutions, conditional expansions, or index derivations. Specifically, this includes: S5.1: Perform word segmentation and part-of-speech tagging on the semantic summary to be compared and the context summary in the historical context snapshot to extract the set of high-frequency domain terms and the sequence of functional words, thereby obtaining a standardized list of lexical units as the input basis for subsequent co-occurrence analysis; S5.2: Perform sliding window co-occurrence statistical calculation based on the standardized vocabulary unit list to construct a local context association matrix and filter out significant co-occurrence feature pairs, thereby obtaining a keyword co-occurrence metric to characterize the stability of term distribution in context; An input index structure for co-occurrence analysis is constructed based on the standardized vocabulary unit list to ensure that the relative positions of high-frequency domain terms and functional words can be quickly retrieved in subsequent operations. A sliding traversal algorithm with a preset window size is applied to the input index structure to perform a co-occurrence count accumulation operation on each word unit pair within the window, generating a local co-occurrence count table; The local co-occurrence count table obtained during the sliding traversal is serialized according to the starting position of the window and mapped to a two-dimensional matrix coordinate system to obtain a local context association matrix that represents the degree of mutual association between terms in the local context. A saliency screening operation is performed on the local context association matrix, and a set of significant co-occurrence feature pairs that meet the conditions is extracted by using dual constraints of co-occurrence frequency threshold and association strength threshold set in the domain. Normalization is performed on the set of significant co-occurrence feature pairs, and the co-occurrence frequency of each feature pair is divided by its total occurrence in the global text to generate a keyword co-occurrence metric for quantifying the stability of term distribution in context. By using sliding window co-occurrence statistics and significance screening, the standardized vocabulary unit list from the previous step is transformed into a keyword co-occurrence metric that can be used for subsequent syntactic consistency analysis, thereby achieving an accurate representation of the stability of term distribution in context. For example, in the updated version of the company's annual strategy report, the extracted standardized vocabulary unit list includes core terms such as "peak carbon emissions," "energy structure optimization," and "increase to over 5%" as well as a set of functional words. The sliding window size is set to 5 vocabulary units. When traversing the input index structure, the co-occurrence frequency of any two terms is counted within each window. For example, "peak carbon emissions" and "energy structure optimization" co-occur 12 times across multiple windows. After establishing a local contextual association matrix, the association strength of this pair of features in the matrix is ​​0.86. A co-occurrence frequency threshold of 8 times and an association strength threshold of 0.75 are set. Only feature pairs that meet both threshold conditions are retained, resulting in a set of significant feature pairs including "peak carbon emissions – energy structure optimization" and "energy structure optimization – increase to over 5%." Normalization is then performed on this set using the following formula: ; in, The co-occurrence metric for the i-th and j-th keywords is... This represents the cumulative number of times a feature pair co-occurs across all windows. This indicates the total number of times a single term appears in the entire text. For words Total number of occurrences in the old and new summaries For words After normalization, the co-occurrence metric for "peak carbon emissions – energy structure optimization" was 0.43 in both the old and new abstracts. This metric was used to verify the stability of the metric constraints in subsequent consistency matching of modification relations. The application effect was that it significantly improved the recognition accuracy of the co-occurrence pattern of core terms in the context, effectively supporting the accuracy of semantic compensation determination. S5.3: The keyword co-occurrence quantification index is used to drive the dependency parsing algorithm to reconstruct the structure of the original summary fragment, so as to identify the subject-verb-object core skeleton and the modifier paths of attributive, adverbial and complement, thereby obtaining the consistency matching score of the modification relationship to evaluate the completeness of the index constraints. S5.4: Based on the consistency matching score of the modified relationship and the predefined logical template library, perform logical connector retention verification to detect the existence status of causal, progressive or adversative relationships, thereby obtaining the logical connector retention coefficient to determine the breakage or continuation of the semantic logical chain; Based on the consistency matching score of the modification relation, an initial set of conditions for input logic judgment is constructed, including the causal path identifier, progressive path identifier and adversative path identifier of the dependency syntax tree, and the corresponding logical connector pattern in the predefined logical template library is loaded as the template to be compared. The input modification relation consistency matching score is matrix-mapped with the association rules of each logical connector pattern. The pattern matching operator is called to search for logical connector fragments that fully or partially match the template rules in the semantic summary to be compared, and a logical connector matching state vector is generated. The logical connection matching state vector is grouped according to the type of connector. The matching frequency of each type of connector (causal, progressive, and adversative) is multiplied with the consistency matching score of the modification relationship to quantify the retention strength of that type of connector in the semantic logical chain. The product value is scaled using a normalization operator to map the retention strength of different categories of connectors to a unified interval. The logical connector retention coefficient is then calculated using the following formula: ; in, This is the retention coefficient for logical connectors. Match the state value for the i-th logical connector. The consistency matching score for the i-th type of modification relation. Count the categories of logical connectors; The retention coefficient of logical connectors is compared with a preset threshold. If the coefficient is lower than the threshold, the semantic logical chain is determined to be broken. If it is higher than or equal to the threshold, the chain is determined to be continued. The determination result and the matching score are output together for subsequent multi-factor fusion decision-making. By using logical connector pattern matching, category grouping intensity quantification and normalization coefficient calculation, the modification relationship consistency matching score of the previous step is transformed into a logical connector retention coefficient that can indicate the continuation or breakage of the semantic logical chain, thereby achieving accurate quantification of the evolution of logical relationships in strategic texts. For example, in the scenario of updating corporate strategy text, the input semantic summary to be compared and the historical context summary are parsed using dependency parsing to obtain three types of path identifiers: causal, progressive, and adversative. The consistency matching scores for modification relations are 0.85, 0.76, and 0.64, respectively. The predefined logical template library includes causal connective patterns such as "therefore" and "so," progressive connective patterns such as "and" and "in addition," and adversative connective patterns such as "but" and "however." In the matching state vector, the causal match frequency is 2, the progressive match frequency is 1, and the adversative match frequency is 1. Multiplying the match frequency by the corresponding match score yields the causal strength of 1.70, the progressive strength of 0.76, and the adversative strength of 0.64, which, after normalization to the interval [0,1], are 0.85, 0.38, and 0.32, respectively. The logical connective retention coefficient is calculated using the formula: ,in =3, summing yields a total intensity of 2.0, with a coefficient value of 0.667. Compared with the preset threshold of 0.60, the logical chain is determined to continue, and the output coefficient 0.667 and the state "continue" are used for subsequent multi-factor fusion decision-making. Verification results show that this coefficient can significantly improve the accuracy of semantic evolution classification. S5.5: Based on the keyword co-occurrence quantification index, the consistency matching score of the modification relation, and the retention coefficient of the logical connector, perform multi-factor weighted fusion decision to map to the preset semantic evolution classification rule set, thereby generating a semantic compensation conclusion that is judged as semantic equivalent replacement, condition extension, or index derivation as the final output result. Based on the calculated keyword co-occurrence metric, the consistency matching score of modification relation, and the retention coefficient of logical connector, the three are used as the input parameter matrix for multi-factor fusion decision-making. Normalization is performed on each input parameter to eliminate the influence of different units on the fusion result. The normalization method adopts the maximum and minimum value linear standardization to map all indicators to a unified numerical range. A multi-factor weighted summation model is constructed using weight vectors. The weight values ​​are configured based on domain expert experience and historical validation data. The fusion decision score is calculated using the following formula: ; in, To integrate decision scores, This is a normalized metric for keyword co-occurrence. The normalized consistency matching score for the modification relation is... This represents the normalized logical connector retention coefficient. These are the weight coefficients of the corresponding factors; Threshold discrimination processing is performed on the fusion decision score. The threshold set is set into three intervals according to the preset semantic evolution classification rule base, which correspond to semantic equivalent replacement, condition extension and indicator derivative types respectively. The intervals containing the fusion scores are mapped to the corresponding semantic evolution classification labels to form structured semantic compensation conclusion data units; The data unit is bound to a version identifier and encapsulated in metadata to enable it to be directly passed to the new anchor point confirmation process in subsequent steps. By using multi-factor weighted fusion and classification rule mapping, the multi-granularity matching results of the previous step are transformed into executable semantic compensation judgment data, thereby achieving accurate classification and compensation decisions for the semantic evolution types of newly added indicators. For example, in a scenario of incremental extraction of strategic text for a certain enterprise, the calculated keyword co-occurrence metric K=0.82, the modification relation consistency matching score R=0.75, the logical connector retention coefficient L=0.65, and the weight coefficients are configured as follows: After normalization, all parameters remain at their original values. Substituting these values ​​into the formula: ; The fusion decision score D = 0.755 was obtained. Based on a preset classification rule threshold set, if the score is ≥ 0.7 and < 0.85, it is determined to be a conditional extension type. This conclusion is bound to the current version identifier to generate a semantic compensation conclusion data unit and passed to the new anchor confirmation process. The effect of applying this embodiment is that the accuracy of the evolution type determination of the new indicator is significantly improved, the classification results are stable and highly interpretable, which helps maintain semantic consistency in the subsequent construction of the context snapshot chain and cross-version backtracking process.

[0023] Step S6: Based on the semantic compensation conclusion, confirm the newly added semantic anchor point, automatically generate a new context snapshot, and establish a bidirectional semantic association edge with neighboring existing semantic anchor points to generate a context snapshot chain that supports cross-version backtracking. Specifically, this includes: S6.1: Perform local syntactic dependency analysis on the text fragments within the preset character window and the semantic anchor points confirmed as new by the semantic compensation conclusion to extract the structured syntactic backbone sequence containing the preceding logical connector, the core predicate verb and the subsequent constraint conditions; S6.2: Perform tuple encapsulation operation based on the structured syntactic backbone sequence, and integrate the unique identifier of the newly added semantic anchor, the paragraph level index, the timestamp mark and the context offset into a compact structured tuple to generate a new context snapshot object to be registered; S6.3: Use a spatial proximity calculation algorithm to perform neighborhood retrieval processing on the newly added context snapshot object and the existing historical context snapshot set, so as to filter out a list of neighboring existing semantic anchors that meet the preset threshold conditions in both the document physical location and semantic vector space. Based on the input conditions of the newly added context snapshot object and the historical context snapshot set, the execution environment and parameter settings for spatial proximity calculation are established to ensure that the physical location and semantic vector space simultaneously meet the preset threshold. Normalize the paragraph level coordinates and context offset of the newly added context snapshot object to convert the document's physical location information into a comparable uniform scale coordinate vector; Perform the same normalization transformation on all objects in the historical context snapshot collection, and build a physical location vector index library to support fast comparison; The distance between the newly added context snapshot object and the historical snapshot object in the physical location vector space is calculated using Euclidean distance, and the physical proximity coefficient is calculated using the following formula: ; in This is the physical proximity coefficient. To add a new snapshot paragraph level x-axis, The horizontal axis represents the historical snapshot paragraph level. To add a snapshot paragraph level y-axis, The vertical axis represents the paragraph hierarchy of historical snapshots; Based on the syntactic backbone feature vector of the newly added context snapshot object and the semantic feature vector of the historical snapshot object, the semantic proximity coefficient is calculated using the cosine similarity algorithm, and obtained through the following formula: ; in, This is the semantic proximity coefficient. and These are the semantic feature vectors of newly added and historical snapshot objects, respectively; The physical proximity coefficient and the semantic proximity coefficient are weighted and fused to generate a composite proximity value. A preset threshold is used to determine whether the physical proximity and semantic matching conditions are met simultaneously. Generate a list of neighboring semantic anchors for all eligible historical snapshot objects and output it for use in the next sub-step; By using spatial proximity calculation and dual threshold filtering, the newly added context snapshot object in the previous step is transformed into a list of neighboring existing semantic anchors that meet the requirements of cross-version semantic tracing, thereby achieving local association pre-filtering of the context snapshot chain. For example, in an enterprise strategic planning system, a newly added context snapshot object has paragraph level coordinates of (12, 5) and a context offset of 45 characters. In the historical context snapshot set, there is an object with paragraph level coordinates of (13, 6) and an offset of 40 characters. After normalization, the physical location vectors are (0.48, 0.20) and (0.52, 0.24) respectively, and the Euclidean distance yields a physical proximity coefficient of 0.056. The semantic feature vector of the newly added snapshot is [0.12, 0.45, 0.38], and the semantic feature vector of the historical snapshot is [0.11, 0.43, 0.40]. The cosine similarity yields a semantic proximity coefficient of 0.998. Weighted fusion is set with a physical weight of 0.4 and a semantic weight of 0.6, resulting in a composite proximity value of 0.624, which is higher than the preset threshold of 0.6. Therefore, the historical object is deemed to meet the criteria and is included in the list of neighboring existing semantic anchors. The processing results significantly improve matching accuracy and semantic coherence preservation in the subsequent construction of bidirectional semantic association edges; S6.4: Perform a bidirectional semantic association edge construction operation based on the list of existing neighboring semantic anchors, and establish a connection link containing pointing relationship, association strength weight and evolution type label between the newly added context snapshot object and each existing neighboring semantic anchor to generate a local topological association structure; S6.5: Perform full-chain list fusion processing based on the local topological association structure, embed the newly generated connection links into the global version evolution graph, and update to form a complete context snapshot chain that supports cross-version backtracking and maintains semantic coherence.

[0024] Step S7: Based on the context snapshot chain, trace upwards along the chain to the nearest comparable historical anchor point of the newly added semantic anchor point, inheriting its semantic role definition, measurement caliber description, and target constraints to generate an incremental key indicator set with vertical consistency. Specifically, this includes: S7.1: Traverse and retrieve the bidirectional semantic association edges in the context snapshot chain, and perform a reverse path search with the newly added semantic anchor point as the starting node to obtain the nearest comparable historical anchor point identifier with the highest semantic similarity to the current newly added semantic anchor point. The bidirectional semantic association edges of the updated context snapshot chain in the global version evolution graph are traversed and retrieved, and the unique identifier of the newly added semantic anchor point is used as the starting node input to the queue initialization module of the retrieval algorithm. The reverse path search operator is invoked to perform adjacency list access processing on the starting node, and the predecessor node identifiers are obtained in descending order of association strength weight. The semantic vector representation of each predecessor node is then loaded into the similarity calculation engine. Based on the similarity calculation engine, the cosine similarity operation is performed on the context semantic vector of the newly added semantic anchor point and the context semantic vector of each predecessor node. The original cosine value is corrected by using a weighted fusion strategy combined with the evolution type label weight coefficient to form a corrected similarity index matrix. The maximum value selection process is performed on the corrected similarity index matrix, and it is determined whether it meets the "comparable historical anchor point" condition by combining the preset similarity threshold. If it meets the condition, the node identifier is output to the result cache unit; otherwise, the reverse search path continues to backtrack until a historical anchor point that meets the condition is found. By using reverse path search and similarity threshold judgment, the association information of the newly added semantic anchors in the context snapshot chain of the previous step is transformed into the most recent comparable historical anchor identifier data, so as to accurately locate the starting point of the incremental key indicator inheritance chain. For example, in the evolution graph of cross-version strategic text, the context semantic vector of the newly added semantic anchor A is [0.12, 0.35, 0.56], and its reverse association edge connects to three historical anchors B, C, and D, with corresponding context semantic vectors of [0.15, 0.33, 0.57], [0.54, 0.12, 0.28], and [0.10, 0.36, 0.60], respectively. When performing cosine similarity calculation, a fusion weight coefficient of 0.8 is used to correct the edges labeled "conditional extension" in the evolution type. The similarity calculation formula is as follows: ,in, For the vector of the newly added anchor point A, Given the vectors of candidate historical anchor points, after weight adjustment, the similarity scores of anchor points B, C, and D are 0.998, 0.756, and 0.994, respectively. With a similarity threshold of 0.95, anchor point B is finally determined to be the most recent comparable historical anchor point. Its identifier is output and recorded in the result cache unit, thus achieving accurate locking of the historical inheritance starting point under the condition of multiple candidate associations. S7.2: Based on the most recent comparable historical anchor identifier, extract the corresponding historical snapshot tuple from the historical context snapshot library, parse the semantic role definition field, measurement caliber description field and target constraint field contained in the historical snapshot tuple, and generate a set of attributes to be inherited; S7.3: Use the domain adaptation rule engine to verify the consistency between the measurement caliber description field in the set of attributes to be inherited and the local context of the current new version of the strategy text. If a conflict in unit system or statistical range is detected, perform adaptive mapping transformation to generate a calibrated set of inherited attributes. When verifying the consistency between the measurement caliber description field in the set of attributes to be inherited and the local context of the current new version of the strategy text, the unit system, statistical range, time span and data collection method of the measurement caliber description field are parsed into standardized structured parameter entries. The structured parameter entries obtained from the parsing are mapped to unit format encoding, and then precisely matched with the indicator numerical units and symbolic markers in the local context of the current new version of the strategic text. The unit matching status flag is calculated by the difference detection operator. For entries with inconsistent unit matching status flags, the unit conversion rule table in the domain adaptation rule engine is invoked to perform adaptive mapping, using conversion coefficients. Adjust the original values ​​according to the mapped unit system, where The dimensional transformation coefficient for the corresponding unit; The statistical range field is parsed to extract the upper and lower limits into numerical pairs. These pairs are then compared with the statistical ranges corresponding to the indicators in the current local context to detect overlap. The matching coefficient is calculated based on the overlap ratio. For entries with a matching coefficient lower than a preset threshold, where The length of the overlapping interval. To determine the length of the union interval, the range adjustment strategy in the domain adaptation rule engine is invoked to expand or shrink the interval boundaries according to the statistical caliber of the new text, and the adjusted range parameters are generated. The time span field is standardized by date formatting, parsing it into start and end times, and then aligned with the execution period of the corresponding metric in the new text, using the period deviation value. To determine cycle consistency, the execution cycle corresponding to the deviation value is corrected using the cycle mapping table. Through the above-mentioned unit conversion, range adjustment and period correction processing methods, the set of attributes to be inherited in the previous step is transformed into a set of inherited attributes that has been adapted and calibrated by the domain, so as to maintain the consistency of the measurement caliber description and the semantic consistency of the context in the incremental extraction of key indicators across versions. For example, in a strategic planning scenario for an energy company, the measurement scope description field in the historical snapshot records the indicator "carbon emission intensity" in kgCO2 / MWh, covering the entire domestic industry, with a time span of 2020-2022. In the new strategic document, the unit of this indicator has been changed to tCO2 / GWh, the statistical scope has been narrowed to the domestic power industry, and the time span has been adjusted to 2021-2023. The system fails to match the unit format after parsing, and the domain adaptation rule engine uses the built-in unit conversion table to call the unit conversion coefficient 1000 and apply the conversion coefficient. The unit conversion from kg to t was completed, and the scale unification from MWh to GWh was achieved by using the power conversion factor 1, resulting in the new unit tCO2 / GWh. The statistical range matching coefficient was calculated as follows. (The historical statistics covered the entire domestic industry, encompassing 8 business areas; the new statistics cover the domestic power industry, belonging to 4 of those categories.) A value below the threshold of 0.7 triggers a scope adjustment strategy, narrowing the scope boundary from the entire industry to the power industry code. Time span deviation value. In 2021, the periodic mapping table was revised to 2023. After the above processing, the output calibrated inherited attribute set has the new text context parameters, the incremental key indicator decoding results show a significant improvement in longitudinal consistency, the verification results show that the differences in indicator units across versions are eliminated, and the statistical range and time span, after adjustment, have complete historical evolution information; S7.4: Bind the semantic role definition, metrological caliber description and target constraints in the calibrated inherited attribute set to the structured descriptor of the newly added semantic anchor point to form an enhanced semantic anchor point instance with complete historical evolution information; S7.5: Aggregate and encapsulate all enhanced semantic anchor instances that have completed attribute binding, extract key numerical expressions and logical constraints, generate an incremental set of key indicators with vertical consistency, and output it to the strategic goal decoding module.

[0025] Step S8: Update the strategic objective decoding output using the incremental key indicator set, and complete the structured decoding of key indicator changes in the dynamically evolving strategic text without re-parseing the full text. Specifically, this includes: S8.1: Obtain the newly added semantic anchors and their associated context snapshot chains from the incremental key indicator set, extract the semantic role definition data of the most recent comparable historical anchors based on the bidirectional semantic association edge traversal logic, and generate indicator metadata objects with vertical inheritance. S8.2: Perform structured mapping processing on the measurement caliber description and target constraints in the indicator metadata object, and use the version difference merging algorithm to logically align the numerical expression of the newly added semantic anchor with the historical benchmark value to generate an updated standardized indicator parameter set. S8.3: Based on the standardized indicator parameter set, perform a state refresh operation on the strategic target decoding engine, replace the corresponding static indicator entries in the original decoding output through an incremental injection mechanism, and generate a dynamic strategic target view containing change timestamps and evolution path markers; S8.4: Perform integrity verification and logical consistency verification on the dynamic strategic target view, use cross-version backtracking pointers to check the dependency relationship between the newly added indicators and the adjacent existing semantic anchors, and generate the final structured key indicator decoding result file; Perform a structured integrity check on the dynamic strategic target view that includes change timestamps and evolution path markers to compare the consistency of the indicator parameter set of the newly added semantic anchors with the historical baseline values ​​in terms of data type, number of fields and constraint format, to ensure that the indicator set is free of missing or abnormalities at the structural level. Consistency verification of the logical dependencies of each new indicator is performed. By using cross-version backtracking pointers to retrieve neighboring semantic anchors in the context snapshot chain, semantic role definitions, measurement caliber descriptions and target constraints are extracted, and a dependency comparison matrix is ​​constructed. Based on the dependency comparison matrix, the semantic logic verification operator is invoked to calculate the survival state coefficient of the target decoding logic chain, using the formula: ; in, Indicates the number of causal relationships retained. This indicates the number of items to retain in a progressive relationship. Indicates the number of broken relationships and determines the continuity of semantic logic chains; For indicator entries with a continuity coefficient below the threshold, perform dependency repair processing, inject key logical connectors and modifiers from historical anchors into the structure descriptor of the new indicator, and update its logical verification status. All indicator entries that pass the integrity and logical consistency checks are integrated into the final structured key indicator decoding result file, and a decoding version identifier is bound to them to achieve the determinism and traceability of the structured decoding output. By using cross-version backtracking pointer analysis and dependency repair, the results of the previous step are transformed into structured key indicator data with logical closed-loop characteristics and historical consistency, so as to achieve stable decoding of dynamic strategic goal view in multi-version evolution. For example, in a corporate strategic planning system, a new version of the strategic text adds the indicator "Increase R&D investment intensity to 5.5% by 2025". The corresponding context snapshot chain includes the historical anchor "Increase R&D investment intensity to 5% by 2023" and the adjacent anchor "Chemical emission peak". In the integrity check, it was detected that the numerical field type of the new indicator is consistent with the historical data, the number of fields matches, and no constraints are missing. In the logical consistency check, the cross-version backtracking pointer retrieved that this indicator depends on the historical anchor's measurement caliber "calculated as a percentage of annual operating revenue" and the target constraint "continuous growth". In the dependency comparison matrix, the number of causal relationships retained is C=8, the number of progressive relationships retained is E=3, the number of broken relationships is D=1, and the continuity coefficient is calculated as follows: = 0.917, which is higher than the threshold of 0.85 after verification, so no repair is needed. In the output file, this indicator entry is bound to the decoding version identifier "V2025Q1" and retains the cross-version evolution path mark, realizing the high accuracy and strong traceability of the strategic goal decoding module in real time.

[0026] The present invention also provides a structured decoding system for enterprise strategic objectives based on semantic space mapping, which uses the above-mentioned structured decoding method for enterprise strategic objectives based on semantic space mapping to perform structured decoding of enterprise strategic objectives.

[0027] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0028] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A structured decoding method for enterprise strategic objectives based on semantic space mapping, characterized in that, Includes the following steps: S1: Obtain the initial version of the strategic text, extract the original semantic fragments of the initial version of the strategic text, and filter standardized policy terms, numerically constrained indicator phrases, and nested numerical expressions from the original semantic fragments to generate semantic anchors; S2: Based on the semantic anchors and their paragraph levels, adjacent syntactic stems and preceding and succeeding logical connectors, construct compact structured tuples, and generate an initial context snapshot based on the compact structured tuples; S3: Obtain the dynamically evolved new version of the strategy text and locate the changed areas, map the semantic anchors registered in the old version to the corresponding positions in the new version, and generate anchor migration results; S4: For the state of mapping failure in the anchor point migration result, call the domain-adaptive language model to perform a local scan of the preset character window around the semantic anchor point to generate a semantic summary to be compared. S5: Based on the semantic summary to be compared and the context summary in the historical context snapshot, perform multi-granularity similarity matching to generate semantic compensation conclusions, specifically including: The semantic summary to be compared and the context summary in the historical context snapshot are processed by word segmentation and part-of-speech tagging, and the high-frequency domain term set and functional word sequence are extracted to obtain a standardized lexical unit list. Based on the standardized vocabulary unit list, a sliding window co-occurrence statistical calculation is performed to construct a local context association matrix and filter out significant co-occurrence feature pairs to obtain a keyword co-occurrence metric. The original abstract fragment is restructured using the keyword co-occurrence metric to identify the subject-verb-object core skeleton and the modifier paths of modifiers and complements, and to obtain the consistency matching score of the modification relationship. Based on the consistency matching score of the modified relationship and the predefined logical template library, the retention rate of logical connectors is checked to detect the existence status of causal, progressive or adversative relationships and obtain the retention rate coefficient of logical connectors. Based on the keyword co-occurrence quantification index, the modification relation consistency matching score, and the logical connector retention coefficient, a multi-factor weighted fusion decision is performed to generate a semantic compensation conclusion. S6: Based on the semantic compensation conclusion, the newly added semantic anchor point is confirmed, and a new context snapshot is automatically generated and a bidirectional semantic association edge is established with the adjacent existing semantic anchor points to generate a context snapshot chain; S7: Based on the context snapshot chain, trace the nearest comparable historical anchor point of the newly added semantic anchor point upwards along the chain, inherit its semantic role definition, measurement caliber description and target constraints, and generate an incremental set of key indicators; S8: Update the strategic objective decoding output using the aforementioned incremental key indicator set, and complete the structured decoding of key indicator changes in the dynamically evolving strategic text without re-parseing the full text. Specifically, this includes: Obtain the newly added semantic anchors and their associated context snapshot chains from the incremental key indicator set, extract the semantic role definition data of the most recent comparable historical anchors based on the bidirectional semantic association edge traversal logic, and generate indicator metadata objects. The measurement scope description and target constraints in the indicator metadata object are structured and mapped, and the numerical expressions of the newly added semantic anchors are logically aligned with the historical benchmark values ​​to generate an updated set of standardized indicator parameters. Based on the standardized indicator parameter set, the state refresh operation of the strategic target decoding engine is performed, and the corresponding static indicator entries in the original decoding output are replaced by an incremental injection mechanism to generate a dynamic strategic target view. The dynamic strategic target view is subjected to integrity verification and logical consistency verification. Cross-version backtracking pointers are used to check the dependency relationship between newly added indicators and adjacent existing semantic anchors, and the final structured key indicator decoding result file is generated.

2. The structured decoding method for enterprise strategic objectives based on semantic space mapping according to claim 1, characterized in that, The extraction of the original semantic fragments specifically involves: performing paragraph-level division and syntactic tree parsing on the initial version of the strategic text, extracting the syntactic core units containing leading logical connectors, core predicate verbs, and subsequent modifying components, thereby obtaining a set of original semantic fragments.

3. The structured decoding method for enterprise strategic objectives based on semantic space mapping according to claim 1, characterized in that, The compact structured tuple contains a context offset and a timestamp.

4. The structured decoding method for enterprise strategic objectives based on semantic space mapping according to claim 1, characterized in that, S3 specifically includes: The paragraph-level hash fingerprint of the dynamically evolved new version of the strategic text is calculated and compared with the paragraph-level hash fingerprint of the initial version of the strategic text to generate a set of coordinates of the changed area. Based on the set of coordinates of the changed region, local text fragments are extracted, and feature vector matching operations are performed on the local text fragments using a pre-built semantic anchor index library to generate a candidate semantic anchor mapping list. Perform unique identifier consistency verification and context offset tolerance verification on each candidate in the candidate semantic anchor mapping list to generate an anchor mapping record to be confirmed. Based on the verification pass or fail status of the anchor point mapping record to be confirmed, the semantic anchor points registered in the old version are associated with the new version text coordinates or marked as unmatched objects, generating anchor point migration results.

5. The structured decoding method for enterprise strategic objectives based on semantic space mapping according to claim 4, characterized in that, The anchor point migration result includes a mapping success status or a mapping failure status.

6. The structured decoding method for enterprise strategic objectives based on semantic space mapping according to claim 1, characterized in that, S4 specifically includes: Obtain the semantic anchor identifier of the mapping failure state and its position coordinates in the old version of the strategy text. Calculate the corresponding local scanning region boundary in the new version of the strategy text based on the preset context window radius parameter, and generate a set of local text window coordinates. Based on the set of local text window coordinates, the original character sequence fragments are extracted from the new version of the strategic text. Noise filtering and format standardization are performed on the original character sequence fragments to generate clean local text data blocks. Load the pre-trained domain-adaptive language model weight parameters, inject the clean local text data block as an input prompt sequence into the encoder layer of the domain-adaptive language model, perform forward propagation inference, and generate a high-dimensional context-aware embedding tensor. Based on the high-dimensional context-aware embedding tensor, key semantic information is aggregated through an attention mechanism, and the aggregated semantic information is compressed and reconstructed using the sequence generation strategy of the decoder head to generate a candidate semantic summary sequence. The candidate semantic summary sequence is subjected to structured verification and redundancy removal operations. Non-critical descriptive information is filtered based on the domain terminology dictionary to generate standardized semantic summary objects to be compared.

7. The structured decoding method for enterprise strategic objectives based on semantic space mapping according to claim 6, characterized in that, The local text window coordinate set includes a start offset and an end offset.

8. A structured decoding system for enterprise strategic objectives based on semantic space mapping, characterized in that: The enterprise strategic objectives are structured and decoded using the semantic space mapping-based enterprise strategic objective structure decoding method described in any one of claims 1-7.

Images