A standard version difference automatic comparison and revision impact analysis method and system

Abstract

The application provides a standard version difference automatic comparison and revision influence analysis method and system, relates to the technical field of automatic text generation, and comprises the following steps: obtaining an old version character sequence and a new version character sequence of a standard draft; constructing a draft position difference interval according to the old version character sequence and the new version character sequence; collecting a comment coordinate sequence and a comment suggestion text reviewed by a standard review expert; establishing a comment adjustment mapping sequence according to the comment coordinate sequence, the comment suggestion text and the draft position difference interval; determining a comment internal compliance judgment sequence through vector dot product comparison operation according to the comment adjustment mapping sequence; generating a comment external rationality judgment sequence according to the comment internal compliance judgment sequence; and generating a standard version difference automatic comparison and revision influence analysis text according to the comment external rationality judgment sequence. The application can significantly improve the objectivity, accuracy and standardization of the manuscript revision and checking work.

Description

Technical Field

[0001] This invention relates to the field of automatic text generation technology, and in particular to a method and system for automatic comparison and revision impact analysis of standard version differences. Background Technology

[0002] As industry standardization continues to advance, the compilation, revision, and review of various approval documents and industry standard texts are becoming increasingly routine, leading to a continuous increase in the frequency of document version iterations. During the entire lifecycle management of documents, it is inevitable that multiple sources of data will be generated, including old versions, revised versions, and expert review comments. This necessitates several core tasks, such as identifying differences between different versions, verifying the implementation of comments, and checking the rationality of self-revised content. Traditional manual review methods are ill-suited to the rapid comparison needs of large batches of multiple versions of documents, and are inefficient at identifying minor textual changes, implicit semantic adjustments, and the implementation of comments. Therefore, the industry urgently needs automated and intelligent methods for document version difference analysis and revision verification.

[0003] Most current mainstream document version comparison and revision review technologies rely on basic text matching algorithms for simple character comparison, which can only identify superficial visual changes such as text additions, deletions, replacements, and word order adjustments. Some optimized solutions can combine basic coordinate positioning to achieve a rough correspondence between annotations and the document, rely on keyword matching to simply determine the implementation status of annotation content, and rely on manual screening to distinguish between modifications within annotations and unannotated self-adjustments. These solutions can basically achieve basic document difference marking and simple revision statistics, and are widely used in routine document version proofreading scenarios.

[0004] However, most existing technologies are limited to shallow character comparison, making it difficult to uncover the hidden differences in deep semantics, technical terminology, and word and sentence distribution. They are prone to overlooking potential problems with logical deviations even without obvious textual changes. At the same time, the spatial matching accuracy between annotations and the manuscript is limited, which can easily lead to misaligned annotations and disordered associations. It is impossible to accurately define overlapping and related areas and independent modification areas, and there is a lack of unified quantitative judgment criteria. As a result, the overall verification and evaluation lacks objectivity and standardization. Summary of the Invention

[0005] To address the shortcomings of existing technologies, which are mostly limited to superficial character comparison and fail to uncover hidden differences in deep semantics, terminology, and word distribution, easily overlooking potential problems with logical deviations despite no obvious textual changes; and to improve the accuracy of spatial matching between annotations and manuscript differences, leading to misaligned annotations and disordered associations, the inability to accurately define overlapping and independently modified areas, and the lack of unified quantitative judgment criteria, resulting in a lack of objectivity and standardization in the overall review and evaluation, this invention provides a standard version difference automatic comparison and revision impact analysis method and system.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: The first aspect of this invention provides a method for automatic comparison and revision impact analysis of standard version differences, comprising: S1: Obtain the old and new character sequences of the standard draft for approval; S2: Construct the position difference range of the approval document based on the old and new character sequences; S3: Collect the annotation coordinate sequence and annotation suggestion text reviewed by the standard review experts; S4: Establish an annotation adjustment mapping sequence based on the annotation coordinate sequence, annotation suggestion text, and the position difference range of the submitted manuscript; S5: Adjust the mapping sequence according to the annotations, and determine the conformity judgment sequence within the annotations through vector dot product comparison operation; S6: Generate an external reasonableness judgment sequence based on the internal compliance judgment sequence; S7: Based on the reasonableness judgment sequence outside the annotations, generate standard version difference automatic comparison and revision impact analysis text.

[0007] A second aspect of this invention provides a standard version difference automatic comparison and revision impact analysis system, comprising: processor; The memory stores computer-readable instructions, which, when executed by the processor, implement the standard version difference automatic comparison and revision impact analysis method as described in the first aspect.

[0008] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the standard version difference automatic comparison and revision impact analysis method of the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, by acquiring the old and new character sequences and constructing the positional difference range of the submitted manuscript, the implicit differences in text word distribution, professional terminology, and internal logic can be effectively identified, overcoming the inherent limitations of traditional shallow character comparison. At the same time, the annotation coordinate sequence and annotation suggestion text are collected, and the annotation adjustment mapping sequence is constructed by combining the manuscript difference range. The internal compliance judgment of the annotation is completed by using vector dot product comparison, and the external rationality judgment sequence is further generated. The overlapping and related areas of annotations and the independent modification areas are accurately divided. A unified judgment standard is established based on quantitative comparison rules, which significantly improves the objectivity, accuracy, and standardization of the manuscript revision and verification work. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an automatic comparison and revision impact analysis method for standard version differences provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of the structure of a standard version difference automatic comparison and revision impact analysis system provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0014] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0015] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0016] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] Reference manual attached Figure 1 The diagram shows a flowchart of an automatic comparison and revision impact analysis method for standard version differences provided by an embodiment of the present invention.

[0019] This invention provides a method for automatic comparison and revision impact analysis of standard version differences. This method can be implemented using a device for automatic comparison and revision impact analysis of standard version differences, which can be a terminal or a server. The processing flow of this method may include the following steps: S1: Obtain the old and new character sequences of the standard draft for approval.

[0020] S2: Construct the position difference range of the approval document based on the old and new character sequences.

[0021] Optionally, the range of positional differences in the submitted documents may specifically include: the starting point of the difference, the ending point of the difference, and the magnitude of the change.

[0022] In one possible implementation, S2 specifically includes: S201: Perform feature encoding parsing and mapping on the old and new character sequences to obtain a vector space distribution topological hierarchical network.

[0023] Among them, the vector space distributed topology hierarchical network is a multi-level feature association mapping system built on the deep feature coding architecture. It takes the cleaned new and old version character sequence vector data as input, and completes the layer-by-layer mapping, non-linear transformation and dimensional unified representation of features through the input layer, multiple fully connected hidden layers and output layer. It can transform unstructured text content into high-dimensional dense feature vectors. Based on the network hierarchical topology relationship, it can characterize the spatial dependency features and contextual association logic between different characters and sentences, and completely retain the distributed feature information of text semantics, structure and content differences. It provides a standardized high-dimensional feature topology support carrier for subsequent matrix generation, vector difference comparison and text difference quantitative analysis.

[0024] S202: Generate the old and new versions of the text feature matrix by using a vector space distributed topological hierarchical network.

[0025] It should be noted that the approval draft text document of the Unified Character Encoding Standard Version 8 (UCVS 8) was directly accessed from the standard industry database. Invalid spaces and newline characters were removed from the document, generating an old character sequence of 5000 characters and a new character sequence of 5200 characters. These cleaned old and new character sequences were then input into the input layer of the feature encoding architecture. This input layer contains 512 neurons and receives 1D word vector data converted from characters. The input layer data is then passed to a feedforward neural network containing 6 hidden layers. The hidden layers are connected using fully connected logic. Each hidden layer contains 1024 neurons, and a linear rectified function is used as the activation function in each hidden layer to perform non-linear mapping processing on the signal, removing redundant feature signals with values ​​less than 0. The feature data processed by the hidden layers is finally passed to the output layer containing 768 neurons, which directly outputs a dense feature vector of 768 dimensions. During the training phase, the feature encoding architecture employs a stochastic gradient descent algorithm to update the network node weights. The learning rate is set to 0.001, the total number of iterations is set to 500, and the batch size is set to 32 samples. The mean squared error loss between the network output and the ground truth labeled matrix is ​​calculated using backpropagation. By extracting the feedback values ​​from the output layer of the feature encoding architecture, the old and new versions of the text feature matrix are obtained.

[0026] For example, for a test set containing 100 characters, the input layer divides it into 100 independent labeled nodes, each node corresponding to a 512-dimensional basic vector. After propagation through a fully connected layer of 6 hidden layers, the output layer generates a matrix set with 100 rows and 768 columns. Through the above matrix dimension transformation, the spatial dependencies of different characters are accurately mapped.

[0027] S203: Based on the old and new versions of the text feature matrix, perform feature level alignment difference comparison on the same row dimension of the matrix, extract the state offset magnitude between corresponding vector nodes, and obtain the row-level feature deviation map.

[0028] Among them, the line-level feature deviation map is based on the same-line feature vectors of the old and new versions of the text feature matrix. The vector offset amplitude is calculated line by line using the cosine similarity algorithm. Combined with the preset difference benchmark value, the high and low deviation intervals are divided. The visualized correlation distribution map is formed with the line number coordinate as the horizontal index and the state offset amplitude and deviation level as the correlation attributes. It can quantify the degree of feature difference of each line of text, intuitively present the semantic feature deviation of different text positions, and accurately distinguish between substantive content changes and invalid format changes.

[0029] Specifically, 1D row feature vectors with the same row index coordinates within the old and new text feature matrices are extracted. Cosine similarity comparison logic is then performed on these two 1D row feature vectors. Specifically, the sum of squares of all elements in the first matrix row feature vector is obtained and its square root is taken to obtain the first vector's magnitude. Similarly, the sum of squares of all elements in the second matrix row feature vector is obtained and its square root is taken to obtain the second vector's magnitude. The corresponding elements are multiplied and summed to obtain the vector dot product. The vector dot product is divided by the product of the first and second vector magnitudes to obtain the feature level alignment difference. Subtracting this feature level alignment difference from the number 1 derives the state offset magnitude between the corresponding vector nodes. A 2D correlation table between row coordinates and state offset magnitudes is established, generating a row-level feature deviation map.

[0030] For example, when the magnitude of the first vector is 2.0, the magnitude of the second vector is 3.0, and the vector dot product is 4.8, dividing 4.8 by the product of 2.0 and 3.0 (6.0) yields a feature level alignment difference of 0.8. Subtracting 0.8 from 1 gives a state offset of 0.2. A difference judgment benchmark of 0.15 is set. When the calculated state offset is greater than 0.15, the coordinate is considered to be deviating from the higher range; when the value is less than or equal to 0.15, the coordinate is considered to be deviating from the lower range. Through cross-comparison of vector magnitude and dot product, subtle semantic deviations between adjacent text are accurately quantified, eliminating feature noise caused by routine punctuation modifications.

[0031] It should be noted that each row number has a different state offset magnitude and deviation range classification. The first row has a state offset magnitude of 0.20, which corresponds to the higher deviation range. The second row has a state offset magnitude of 0.05, which is classified as the lower deviation range. The third row has a state offset magnitude of 0.18, which also belongs to the higher deviation range.

[0032] S204: Based on the row-level feature deviation map, retrieve the coordinates of the out-of-bounds row number.

[0033] S205: Aggregate the coordinates of out-of-bounds row numbers across consecutive row segment boundaries to obtain the coordinate distribution topology network structure.

[0034] The coordinate distribution topology network structure is based on the cross-boundary line number coordinates obtained through screening. It identifies the continuous correlation between adjacent coordinates and completes the aggregation of cross-line segment boundaries. It integrates the core parameters such as the start and end points and the magnitude of change of each continuous difference interval. Combined with the spatial correlation topology system constructed by the spatial position mapping relationship of the manuscript page, it can integrate discrete difference line coordinates into continuous and complete difference segments. It clearly represents the distribution location, range boundary and spatial arrangement logic of each difference region, eliminates the interference caused by fragmented discrete differences, and provides structured spatial topology support for accurately constructing the position difference interval of the submitted manuscript.

[0035] S206: Based on the coordinate distribution topology network structure, construct the position difference interval of the submitted documents.

[0036] It should be noted that the process iterates through all state offset magnitudes within the row-level feature deviation map, comparing each offset magnitude with a set difference benchmark of 0.15. Nodes with state offset magnitudes greater than 0.15 are retrieved and extracted, defined as out-of-bounds row number coordinates. For all out-of-bounds row number coordinates, a spatial adjacency identification operation is performed. It is determined whether the difference between any two out-of-bounds row number coordinates is equal to 1. If the difference is equal to 1, they are grouped into the same continuous cross-row segment, and the starting and ending row numbers of this cross-row segment are extracted as segment boundary aggregation parameters. By mapping the absolute distance relationship between the above segment boundary aggregation parameters and the spatial location of the approval document, a coordinate distribution topology network structure is constructed. Based on this network structure, continuous starting and ending row ranges are output, yielding the positional difference interval of the approval document.

[0037] For example, when the set of out-of-bounds line number coordinates includes lines 10, 11, 12, and 18, if the difference between lines 10 and 11 is 1, and the difference between lines 11 and 12 is also 1, then these three coordinates are aggregated into the first consecutive line-spanning segment, with a starting line number of 10 and an ending line number of 12. Line 18 then forms a separate second consecutive line-spanning segment with both starting and ending line numbers of 18. By calculating the difference between adjacent line number coordinates for continuity identification, isolated noise differences can be effectively eliminated, accurately locating large, continuous areas of actual text variation, and avoiding frequent fragmented text merging.

[0038] In this embodiment of the invention, the character sequences of the old and new versions are first standardized, cleaned, and uniformly encoded. Deep feature extraction is then performed using a multi-layer fully connected feature encoding architecture to generate a corresponding text feature matrix. The precise state offset magnitude is then quantified using cosine similarity calculation of the same row vectors. The deviation interval is divided by a preset benchmark value, and the coordinates of out-of-bounds line numbers are screened. Subsequently, the continuous out-of-bounds coordinates are aggregated into segment boundaries and topological structures are constructed. Finally, a standardized interval of positional differences in the approval document is formed. This allows for precise identification of subtle text changes from a semantic feature perspective, filtering out invalid noise interference such as punctuation adjustments, effectively eliminating fragmented and isolated differences, and accurately locating continuous and substantial text change areas. This enables quantitative comparison, continuous positioning, and structured division of differences between the old and new documents, significantly improving the accuracy, comprehensiveness, and automated processing efficiency of document difference identification. It provides regular and reliable interval data support for subsequent revision content verification, annotation association, and version difference analysis.

[0039] S3: Collect the annotation coordinate sequence and annotation suggestion text reviewed by the standard review experts.

[0040] S4: Establish an annotation adjustment mapping sequence based on the annotation coordinate sequence, annotation suggestion text, and the position difference range of the submitted manuscript.

[0041] The annotation adjustment mapping sequence specifically includes: source locator, destination locator, and mapping association key.

[0042] In one possible implementation, S4 specifically includes: S401: Perform spatial dimension feature mapping and alignment on the line number coordinates in the position difference interval of the submitted manuscript and the boundary coordinates in the annotation coordinate sequence to establish a coordinate comparison benchmark network.

[0043] Among them, the coordinate comparison benchmark network takes the line number pixel coordinates of the difference range of the submitted manuscript and the boundary pixel coordinates of the expert annotation as the core carrier. By calculating the Euclidean distance of pixels and combining it with the overlap judgment threshold to screen related overlapping nodes, the spatial coordinate association benchmark system established based on the spatial correspondence binding relationship between nodes can unify the spatial mapping rules between the manuscript text area and the annotation page area, and accurately establish a two-way correspondence between the revision difference position and the review annotation position.

[0044] Specifically, the optical character recognition component is activated to retrieve the expert review and annotation document stored on the local hard drive. The absolute pixel coordinates of the four vertices of the text box containing review comments are extracted to construct an expert review and annotation coordinate sequence. The Chinese characters within the text box are then extracted to construct the annotation suggestion text. All line number coordinates within the positional difference range of the submitted document are converted to absolute page height pixel coordinates. The difference between these height pixel coordinates and the absolute pixel coordinates of the vertices in the expert review and annotation coordinate sequence is calculated to determine the pixel space Euclidean distance. An overlap judgment benchmark of 15 pixels is set. The calculated pixel space Euclidean distance is compared with the overlap judgment benchmark. If the Euclidean distance is less than or equal to 15 pixels, the line number coordinates and annotation boundary coordinates corresponding to this distance are extracted as associated overlapping nodes. Based on the pointer correspondence within all the acquired associated overlapping nodes, a coordinate comparison benchmark network is established.

[0045] For example, a line number coordinate converted to an absolute height of 800 pixels, while the absolute height of the top vertex of an expert review annotation text box is 810 pixels. Subtracting 800 pixels from 810 pixels yields a difference of 10 pixels. The absolute value of this 10 pixels is less than the overlap threshold of 15 pixels. Therefore, the line number coordinates are bound to the annotation boundary coordinates and entered into the associated overlapping node set. By calculating the absolute distance at the pixel level and comparing it with the threshold value, the physical image space mapping relationship between line numbers in electronic documents and expert annotations is accurately bridged.

[0046] S402: By comparing coordinates with the baseline network, the range intersection state discrimination operation is performed on the row number coordinates and boundary coordinates to obtain the interval state classification set.

[0047] It should be noted that the coordinate comparison benchmark network calls all related overlapping node data, and performs a range intersection status identification operation for each node. Specifically, the calculation process involves obtaining the highest and lowest pixel boundary values ​​of the difference interval in the submitted document, and simultaneously obtaining the top and bottom coordinate values ​​of the expert review annotation coordinate sequence. The lowest pixel boundary value of the difference interval is compared with the top coordinate value of the annotation, and the maximum value is taken as the upper boundary of the intersection. The highest pixel boundary value of the difference interval is compared with the bottom coordinate value of the annotation, and the minimum value is taken as the lower boundary of the intersection. The difference between the lower boundary and the upper boundary is then calculated. If the difference is greater than 0, it is determined to be a coordinate overlapping area, and the corresponding element is extracted as an element of the overlapping interval. If the difference is less than or equal to 0, it is determined to be an independent area with no coordinate overlap, and the corresponding element is extracted as an element of the independent interval without intersection. The elements of the overlapping interval and the elements of the independent interval without intersection are distributed and stored in two independent linear list data structures to establish an interval status classification set.

[0048] For example, the difference range is from a minimum of 400 pixels to a maximum of 600 pixels, the top of the annotation is 500 pixels and the bottom is 700 pixels. The upper boundary of the intersection is 500 pixels, and the lower boundary is 600 pixels. Subtracting 500 pixels from 600 pixels yields a difference of 100 pixels. Since 100 pixels is greater than 0, this is classified as an element with an overlapping intersection. Through this mechanism of cross-difference comparison at the upper and lower boundaries, the actual impact range of the difference between the revised comments and the original text is rigorously defined.

[0049] S403: Based on the overlapping interval elements and independent interval elements in the interval state classification set, integrate the annotation suggestion text and establish a mapping association hierarchical topology architecture.

[0050] The mapping and association hierarchical topology architecture is based on the interval classification results output by the coordinate comparison benchmark network. It combines the intersection and overlap relationship and independent distribution characteristics of the difference interval and the annotation area, integrates various annotation suggestion text contents, and completes the node hierarchical arrangement and chain splicing based on the sorting logic of the page's global pixel coordinates. It constructs a multi-node, hierarchical, and strongly associated topology association system, which can realize the hierarchical binding and orderly association between the manuscript difference interval, spatial coordinates, and annotation content, and clearly distinguish the mapping relationship between overlapping associated areas and independent non-associated areas.

[0051] S404: Based on the mapping association hierarchical topology, establish annotation adjustment mapping sequences.

[0052] It should be noted that the process involves reading overlapping interval elements and independent interval elements that do not intersect, stored within the interval state classification set, and extracting the global height pixel coordinate attribute of each element as the sorting key. Using the quicksort algorithm, all interval elements are processed in ascending order of their global height pixel coordinates, and the underlying node queue is concatenated and aggregated to construct a fixed-length 1D node queue. For each element node in the 1D node queue, the corresponding annotation suggestion text content is read based on its attached annotation text identifier. This text content is then appended as a string variable to the element node's data mounting area, and the annotation suggestion text content is simultaneously merged to construct a mapping association hierarchical topology. By outputting the complete linked list data structure of this topology, the annotation adjustment mapping sequence is obtained.

[0053] For example, if overlapping intervals have a height of 300 pixels and independent intervals have a height of 150 pixels, after comparison and sorting, the system places the independent interval with a height of 150 pixels at the top of the queue, followed immediately by the overlapping interval with a height of 300 pixels, concatenating them into the second node. The extracted Chinese string modified to "two-way lane" is then attached to the first node. By introducing global height coordinates for fast sorting and synchronized character attachment, logical conflicts caused by out-of-order annotations from different reviewers are eliminated.

[0054] In this embodiment of the invention, by combining pixel coordinate conversion, Euclidean distance calculation, and overlap threshold determination, spatial alignment of the coordinates of the difference intervals in the submitted manuscript and the coordinates of the annotation boundaries is achieved. Relying on boundary intersection comparison, the overlapping and related relationships between the annotation area and the text difference intervals are accurately identified. Then, by combining coordinate sorting and text mounting mechanisms, a hierarchical topology architecture is constructed and an annotation adjustment mapping sequence is generated. This accurately connects the physical space of the manuscript layout with the corresponding relationship between the revised annotations, effectively identifies the coverage area of ​​the annotations and text modification areas, and organizes the disordered and scattered review annotation content. It avoids annotation misalignment, disorder, and confusion of relationships, achieving one-to-one binding and structured association between difference intervals, position coordinates, and modification suggestions. This ensures accurate matching between annotation comments and manuscript difference content, improves the logic, regularity, and automated matching capabilities of manuscript revision association analysis, and provides orderly and complete mapping data support for subsequent accurate implementation of annotation modifications and version revision verification.

[0055] S5: Adjust the mapping sequence based on the annotations, and determine the conformity judgment sequence within the annotations through vector dot product comparison.

[0056] The specific compliance judgment sequence within the annotation includes: compliance items, violations of boundaries, and omissions / missing items.

[0057] In one possible implementation, S5 specifically includes: S501: Adjust the mapping sequence according to the annotations and retrieve overlapping interval elements in the mapping association hierarchical topology.

[0058] S502: Extract the feature representation entities corresponding to the new text feature matrix and annotation suggestion text in the overlapping interval elements.

[0059] Among them, the feature representation entity is a standardized high-dimensional vector representation formed by deep feature encoding and dimensional unification mapping of the new version of the revised text fragments and the annotation suggestion text through a dedicated neural network. It can transform the text content in natural language form into calculable dense feature data, deeply condense the semantic connotation, sentence structure and professional expression features of the text, unify the feature dimension specifications of texts from different sources, and can be directly used for vector dot product, similarity comparison and difference quantification calculation, providing a standardized digital feature carrier for accurately judging the degree of adoption of annotations and the matching of text content.

[0060] S503: Perform feature space vector distribution hierarchical mapping on the feature representation entity to generate a set of interval feature distribution states.

[0061] Specifically, all linked list nodes within the annotation adjustment mapping sequence are traversed, and internal overlapping interval elements marked with overlapping attributes are retrieved and filtered out by node type identifiers. Based on the row number index pointer stored within the overlapping interval element, the 768-dimensional row vector data corresponding to the row number in the new text feature matrix is ​​extracted. Simultaneously, the annotation suggestion text content attached to the overlapping interval element is input into a separate feature extraction convolutional neural network. This network contains one input layer that receives the text sequence, three convolutional hidden layers that use 3x3 convolutional kernels for sliding scanning, and a max-pooling mechanism after each hidden layer to reduce the feature dimension. Finally, a fully connected output layer generates a feature representation entity vector with the same 768-dimensional dimension. The extracted new text feature matrix row vector data and feature representation entity vectors are paired and combined according to the same dimension index, performing a feature space vector distribution hierarchical mapping to generate a set of interval feature distribution states containing paired vector data.

[0062] For example, if the 5th node is found to be an element within an overlapping region, the 768-dimensional feature vector of the 8th row of the new matrix it points to is retrieved. The annotation text attached to this node is then input into a convolutional neural network. After three layers of convolution processing to extract local sentence structure features, another 768-dimensional feature representation entity vector is output. The two are then packaged into a set unit containing 1536 values ​​for storage. Through local feature extraction by the convolutional kernel and pairing mapping with high-dimensional vectors, the deep semantic correspondence between the modified text and the original text is deeply mined.

[0063] S504: Based on the set of interval feature distribution states, perform a multi-dimensional comparison of the vector dot product space of the new text feature matrix and the feature representation entity to obtain the coordinate nodes.

[0064] S505: Perform state marking and filtering on coordinate nodes to establish a decision node network.

[0065] Among them, the decision node network is based on paired feature vectors in the set of interval feature distribution states. It obtains quantitative feature values ​​by comparing vector dot products, and completes node screening, state marking and invalid data removal by combining preset decision benchmark thresholds. It retains valid coordinate nodes that meet the out-of-bounds judgment conditions and integrates them according to the association logic to form a structured node set. It can centrally carry the matching and verification results of text and annotations, and standardize the identification, attributes and spatial coordinate information of out-of-bounds nodes. It provides a regular and reliable node network support for accurately extracting benchmark out-of-bounds coordinate sequences and carrying out annotation compliance judgment.

[0066] S506: Obtain the baseline out-of-bounds coordinate sequence by determining the node network.

[0067] It should be noted that the paired vector data stored within the interval feature distribution state set is called, and the row vectors corresponding to the new text feature matrix and the feature expression entity vectors are extracted. A multi-dimensional vector dot product comparison operation is then performed on these two. Specifically, the corresponding values ​​of the new feature vector and the feature expression entity vector in each of the 768 dimensions are multiplied one by one, and the product results are summed to obtain the comparison feedback feature value. A judgment benchmark value of 85.0 is set, and it is determined whether the obtained comparison feedback feature value is greater than this benchmark value. If the calculated value is greater than 85.0, the feature expression is determined to have exceeded the set judgment benchmark. The absolute position coordinates of the comparison feedback feature are extracted as coordinate nodes, and the attribute fields of these coordinate nodes are marked as reasonably retained. Coordinate nodes with values ​​less than or equal to 85.0 are removed, and a judgment node network containing all retained coordinate nodes is established. Finally, the benchmark out-of-bounds coordinate sequence is obtained.

[0068] For example, the new version's feature vector values ​​and the annotation feature vector values ​​are multiplied 768 times. After summing all the products, the comparison feedback feature value is calculated to be 92.5. Since 92.5 is greater than the judgment benchmark value of 85.0, the corresponding row number and pixel height coordinates are packaged and stored in the benchmark out-of-bounds coordinate sequence, and the status is marked as 1. Through the comparison of dot product summation in high-dimensional space, the degree of actual adoption and consistency of the new version of the text with the expert annotations is strictly quantified.

[0069] It should be noted that different element nodes correspond to corresponding comparison feedback feature values ​​and out-of-bounds judgment results. Among them, the comparison feedback feature value of node 1 is 92.5, and the judgment result is that it exceeds the judgment benchmark. The comparison feedback feature value of node 2 is 76.3, and the judgment result is that it does not exceed the judgment benchmark. The comparison feedback feature value of node 3 is 88.1, and it is also in the state of exceeding the judgment benchmark.

[0070] S507: Determine the compliance judgment sequence within the annotation based on the baseline out-of-bounds coordinate sequence.

[0071] Specifically, all retained coordinate nodes within the baseline out-of-bounds coordinate sequence are extracted. Simultaneously, elements from previously separated, non-intersecting independent intervals in the annotation adjustment mapping sequence are read, and node queue concatenation and reorganization operations are performed for both types of elements. The absolute page height pixel coordinates recorded by each node and element are extracted as the standard measurement benchmark, and merge sort logic is used to re-compare the absolute height pixel coordinate values ​​of all nodes and elements. Following ascending numerical order, the absolute positions of global elements are integrated and inserted one by one into an empty linked list structure to establish an ordered association architecture, thereby obtaining the annotation compliance judgment sequence.

[0072] For example, the baseline out-of-bounds coordinate sequence contains one reserved coordinate node with an absolute height of 550 pixels, and the independent interval element set contains two element nodes with absolute heights of 450 pixels and 650 pixels respectively. By comparing the pixel height values, the reorganization queue will first insert the 450-pixel node, then the 550-pixel node, and finally the 650-pixel node, thus forming a complete chain with three ordered nodes. Through cross-sequence merge sorting and reorganization of absolute height pixel coordinates, seamless temporal fusion of local verification results and globally independent information is achieved.

[0073] In this embodiment of the invention, overlapping interval elements are precisely screened based on the annotation adjustment mapping sequence. The feature vectors of the new version of the manuscript and the exclusive feature vectors of the annotation suggestion text are extracted respectively. Through unified dimension mapping and multi-dimensional comparison operation of vector dot product, the semantic fit between the new version of the content and the annotation comments is quantified. Combined with a fixed judgment benchmark, the out-of-bounds status is identified and invalid nodes are filtered. A standardized judgment node network is constructed and the benchmark out-of-bounds coordinate sequence is output. Then, independent interval elements are integrated and sorted based on coordinates to complete the global ordered reorganization. Finally, an annotation compliance judgment sequence covering compliance items, out-of-bounds items, and omission items is generated. It can accurately quantify the actual implementation effect of the manuscript revision content on expert annotations from a high-dimensional semantic level, avoid the limitations of surface text comparison, accurately distinguish different situations such as annotation adoption, rectification non-compliance, and annotation omission, realize the refined, standardized, and automated judgment of annotation compliance, and orderly integrate full-domain verification data to ensure that the judgment results are objective, comprehensive, and hierarchical. It provides rigorous data basis and judgment support for subsequent manuscript compliance review, rectification problem tracing, and revision quality assessment.

[0074] S6: Generate an external reasonableness judgment sequence based on the internal compliance judgment sequence.

[0075] The reasonableness judgment sequence for annotations specifically includes: routine polishing items, logical self-consistency items, and terminology standardization items.

[0076] In one possible implementation, S6 specifically includes: S601: Extract the elements of the internal independent intervals based on the conformity judgment sequence within the annotations.

[0077] S602: Retrieve the old and new character sequences based on the spatial mapping coordinates of independent interval elements.

[0078] S603: Perform word segmentation encoding and spatial feature transformation on the old character sequence to obtain the old word distribution state set.

[0079] It should be noted that the process involves traversing all ordered nodes within the conformity judgment sequence of the annotations, and extracting the internal independent interval elements marked as independent attributes by judging the state identifiers within each node. The spatial mapping coordinate parameters stored in these independent interval elements are read, and the corresponding old and new character sequence fragments are precisely retrieved from the local cache database based on the line number interval pointed to by these parameters. The extracted old character sequence fragments are input into the Chinese word segmentation component. Based on the maximum matching segmentation logic and a preset industry-standard professional dictionary, the long string is divided into an array of independent word groups. For each independent word group, a one-hot encoding logic transformation is performed, converting the word group's position index in the dictionary into a high-dimensional sparse vector composed of a large number of 0s and a single 1, thus obtaining the old word element distribution state set corresponding to that independent interval.

[0080] For example, extracting the old character sequence with starting line number 25 and ending line number 26, and segmenting it into two independent word groups, "bridge" and "construction," assuming the total dictionary length is 10000, with "bridge" at position 500 and "construction" at position 600, generates a vector of length 10000, where position 500 is assigned a value of 1 and the rest are 0. A second vector is then generated, where position 600 is assigned a value of 1 and the rest are 0. Packaging these vectors together yields the corresponding word distribution state set. By using maximum matching segmentation and one-hot encoding based on an industry-standard professional dictionary, a fundamental transition from discrete text information to rigorous mathematical expression is facilitated.

[0081] S604: Based on the old version of the lexical distribution state set, perform lexical space mapping transformation on the new version of the character sequence.

[0082] S605: Perform a cross-entropy hierarchical distribution difference comparison on the nodes in the new character sequence and the old word distribution state set to obtain a cross-entropy state difference map.

[0083] Among them, the cross-entropy state difference map takes the word vectors after word segmentation and encoding of the old and new versions of character sequences as the analysis object. It relies on cross-entropy operation to quantify the distribution deviation of words, professional terms and sentence expressions. Combined with the global difference distribution set formed by associating and binding the position coordinates of each word, it can deeply explore the implicit feature differences brought about by word replacement and expression adjustment in the information distribution dimension, and intuitively reflect the degree of change of the unannotated modified content in independent intervals.

[0084] Specifically, the generated old version of the lexical distribution state set is read, and the extracted new version of the character sequence fragment is simultaneously input into the same Chinese word segmentation processing component. A new version of one-hot encoded sparse vector set is generated according to completely consistent maximum matching segmentation logic and dictionary matching rules, completing the lexical space mapping transformation operation to obtain the transformed representation entity. The specific calculation process involves extracting the non-zero position indices in the new version of the sparse vector, retrieving the probability distribution value at the corresponding position in the old version of the sparse vector, taking the natural logarithm of this probability distribution value and multiplying it by -1 to obtain the cross-entropy loss data at that position, summarizing the cross-entropy loss data of all nodes to extract the deviation magnitude of the node state distribution, and obtaining the cross-entropy state difference map.

[0085] For example, after the new version of the sequence word transformation, a sparse vector is obtained with a value of 1 at position 800. Searching the old version of the sparse vector reveals that its probability smoothing value at position 800 is 0.2. Calculating the natural logarithm of 0.2 yields approximately -1.6. Multiplying this by -1 gives a cross-entropy value of 1.6, which is used as the deviation magnitude of the node state distribution. This deviation magnitude, along with the word position identifier, is stored in a dictionary set to construct a difference graph. Through the information-theoretic cross-entropy loss operation logic, the deep distribution difference signal brought about by the lexical changes between the old and new versions is effectively amplified.

[0086] S606: Retrieve spatial coordinates based on the cross-entropy state difference map.

[0087] S607: Perform bottom-level node sequence splicing on spatial coordinates to generate a reasonableness judgment sequence for annotations.

[0088] It should be noted that the deviation values ​​of the state distribution of all nodes recorded within the cross-entropy state difference map are analyzed, and a judgment boundary benchmark value of 1.2 is set. For each node's state distribution deviation value, a cross-boundary verification operation is performed against this benchmark value of 1.2. Specifically, the deviation value is compared to see if it is greater than the judgment boundary benchmark value of 1.2. If the comparison result shows that the deviation value is less than or equal to 1.2, it means that the change between the old and new versions is a routine rhetorical adjustment, and it is judged not to have crossed the judgment boundary benchmark, and the original page height spatial coordinates corresponding to the node are retained. If the deviation value is greater than 1.2, it means that a significant semantic deviation has occurred, and it is removed and not retained. For all spatial coordinates that pass the verification and are retained, the underlying node sequence is spliced ​​in ascending order of value to generate the final annotation external rationality judgment sequence.

[0089] For example, three nodes with deviation values ​​of 0.8, 1.5, and 1.0 were retrieved. 0.8 was compared to 1.2; since it was less than 1.2, its corresponding spatial coordinate of 350 pixels was retained. 1.5 was compared to 1.2; since it was greater than 1.2, it was discarded. 1.0 was compared to 1.2; since it was less than 1.2, its corresponding spatial coordinate of 420 pixels was retained. The 350-pixel and 420-pixel coordinate nodes were then concatenated into a linked list of length 2 in order of size. A quantitative checkpoint using a cross-entropy threshold was used to eliminate redundant changes with no substantial impact, retaining only the core adjustment coordinates.

[0090] In this embodiment of the invention, independent interval elements are accurately extracted from the compliance judgment sequence within the annotations. Character sequence fragments from the old and new versions are located using spatial mapping coordinates. The old text's lexical features are transformed through standardized word segmentation, professional dictionary matching, and one-hot encoding. Then, the new text undergoes the same rule-based lexical spatial mapping. Cross-entropy hierarchical difference comparison quantifies the changes in word distribution and semantic expression. Combined with preset judgment boundary benchmarks, compliance coordinate screening and invalid semantic deviation nodes are eliminated. Finally, based on the ordered splicing of spatial coordinates, an external annotation rationality judgment sequence is generated, including routine polishing items, logical consistency items, and terminology unification items. This sequence can specifically identify self-adjusted content in the manuscript that is not within the scope of expert annotations. A quantitative model accurately distinguishes between reasonable optimization changes and substantial semantic changes, effectively filtering out invalid revision interference. It objectively identifies the compliance and rationality of self-modified content in the manuscript, improves the full-dimensional manuscript review system, and achieves a two-way combination of compliance verification within annotations and rationality judgment outside annotations. This comprehensively enhances the completeness, scientific rigor, and refined analysis level of the manuscript revision review.

[0091] S7: Based on the reasonableness judgment sequence outside the annotations, generate standard version difference automatic comparison and revision impact analysis text.

[0092] Optionally, the text for automatic comparison of standard version differences and analysis of the impact of revisions specifically includes: a statement of the reasons for the revision, a record of review and debate, and a finalized conclusion.

[0093] In one possible implementation, S7 specifically includes: S701: Perform joint selection and aggregation of overlapping interval elements and independent interval elements on the underlying spatial mapping architecture of the annotation rationality judgment sequence to establish a global correlation topology structure for interval coordinate distribution.

[0094] S702: Get the aggregated set of entities in the entire range.

[0095] It should be noted that all node objects within the rationality judgment sequence outside the annotations are extracted as the basic data pool. For the underlying spatial mapping architecture of the sequence, a joint selection and aggregation process of overlapping and independent interval elements is performed. Specifically, this involves extracting all valid judgment coordinate sets remaining within the baseline out-of-bounds coordinate sequence, performing a union operation with the basic data pool, removing duplicate absolute height pixel coordinate values, and retaining all unique individual height coordinates. Based on the merged and deduplicated list of individual height coordinates, the corresponding hidden layer mapping node entities are extracted according to the memory pointer addresses bound within each coordinate. All extracted hidden layer mapping node entities are then reconnected with each other according to their absolute height coordinate values, from smallest to largest, establishing a globally related topology structure for the interval coordinate distribution. Finally, a global interval entity aggregation set containing all relevant modified and retained fragments is output.

[0096] For example, the basic data pool contains two entity nodes with height coordinates of 350 pixels and 420 pixels, while the valid judgment coordinate set contains two nodes with height coordinates of 420 pixels and 550 pixels. These two sets are merged, and the extra 420-pixel coordinate is removed, resulting in three unique coordinates: 350 pixels, 420 pixels, and 550 pixels. Then, forward and backward memory pointers are reassigned to the entity nodes attached to these three coordinates to construct a singly linked list. Through the set union deduplication and relinking mechanism, a globally unified closed-loop management system is achieved for all review annotation modifications and self-correction adjustments.

[0097] S703: Establish a joint topology network of text nodes based on the global association topology structure of the global interval entity aggregation set and interval coordinate distribution.

[0098] S704: Generate a set of multi-source text-related elements by combining text nodes with a topological network.

[0099] Specifically, the system calls upon all entity node data within the global interval entity aggregation set, using the absolute spatial coordinates of the nodes as the exclusive retrieval key. This data is then input into the local traffic standard document archive query interface to retrieve the old version character sequence text, the new version character sequence text, and the initial annotations and suggestions from experts for the corresponding fragments. Multi-source data binding and mapping processing is then performed on the extracted three entity information items. Specifically, a composite data structure with three text attribute fields and one coordinate attribute field is instantiated. The old version sequence, the new version sequence, and the annotations and suggestions are assigned to their respective attribute fields, and the absolute position coordinates are assigned to the coordinate attribute field. All generated composite data structures are stored within the graph database system. A joint topology network of text nodes is established, with the absolute position coordinates considered as the core vertices of the graph database, ultimately generating a multi-source text association element set containing multi-dimensional information.

[0100] For example, based on the node coordinates with a height of 350 pixels, the database retrieves three independent text information entries: the original width of the old version, the modified width of the new version, and the suggested width adjustment for the annotation. A single structure is automatically created, and these three entries are encapsulated into their corresponding fields. The 350-pixel value is set as the key index vertex number in the graph database for storage. This multi-field binding mechanism of the composite structure avoids data gaps or loss of traceability during multi-version collaborative editing and review processes.

[0101] It should be noted that in this multi-source text element association structure table, different absolute position coordinates correspond to old version character sequence records and annotation suggestion text records, respectively. Specifically, the old version character sequence record corresponding to the 350 pixel position is the original construction width parameter, and the corresponding annotation suggestion is to increase the road width; the old version character sequence record corresponding to the 420 pixel position is the design load-bearing weight standard, and the corresponding annotation suggestion is to correct it to the latest standard; the old version character sequence record corresponding to the 550 pixel position is the estimated service life, and the corresponding annotation suggestion is to verify the fatigue test data.

[0102] S705: Construct a continuous text underlying closed-loop serial topology architecture based on the multi-source text association element set.

[0103] S706: Based on the underlying closed-loop serial topology of continuous text, generate standard version difference automatic comparison and revision impact analysis text.

[0104] It should be noted that the process involves extracting all core vertex data from the graph database within the multi-source text association feature set, and reading the bound mapping sequence feature entities within the graph database. The new character sequence attribute field value stored within each feature entity is obtained as the target reconstructed text block. The global absolute position pixel coordinates of the target mapping space attached to each vertex are extracted as the sorting criterion. Following the ascending order of these absolute position coordinate values, the memory address of the last character of each target reconstructed text block and the memory address of the first character of the next adjacent target reconstructed text block are extracted sequentially. A coherent splicing and fusion operation is performed between adjacent character memory addresses, automatically inserting standard paragraph line breaks and paragraph indentation control characters at the splicing point. After completing the splicing loop for all text blocks, a continuous text bottom-level closed-loop concatenated topology is constructed, directly outputting a standard version difference automatic comparison and revision impact analysis text file that conforms to industry typesetting standards.

[0105] For example, consider text block A located at 350 pixels, text block B located at 420 pixels, and text block C located at 550 pixels. The sorting logic places text block A first. The system retrieves the memory address after the period at the end of text block A, forcibly attaches it to the pointer of the first character of text block B, and automatically inserts a tab character containing two full-width spaces in between for concatenation. Then, the connection ports between text blocks B and C are processed in the same way, ultimately integrating them into a standardized, continuous document. Through seamless forced concatenation at the underlying memory address level and synchronized insertion of typesetting characters, the cost of manual typesetting intervention after the final document is completed is reduced.

[0106] In this embodiment of the invention, the interval elements in the rationality judgment sequence outside the annotations are jointly selected and aggregated. A global relational topology structure is constructed through set union deduplication, coordinate sorting, and node linking. All valid revision-related nodes are integrated to form a global interval entity aggregation set. Then, based on coordinate retrieval, the new and old versions of the text, annotation suggestions, and other multi-source information are associated and structured binding is completed. A joint topology network of text nodes and a continuous text closed-loop serial architecture are constructed. Finally, a version difference comparison and revision impact analysis text that conforms to industry standards is generated. It can realize the full-domain integration and closed-loop management of all revision content of the manuscript (including internal annotation rectification and external annotation self-adjustment), clearly associate multiple versions of text, annotation suggestions, and revision positions, automatically complete text splicing and standardized typesetting, and comprehensively present the overall revision, review opinions, and final draft basis. It greatly reduces the workload of manual integration, typesetting, and source tracing, avoids the problems of multi-source data discontinuity, source loss, and typesetting chaos, and provides comprehensive, orderly, and traceable standardized analysis text support for manuscript revision review, source tracing verification, and final draft archiving, thereby improving the efficiency and standardization of manuscript editing and review.

[0107] Reference manual attached Figure 2The diagram shows a structural schematic of a standard version difference automatic comparison and revision impact analysis system provided by the present invention.

[0108] This invention also provides a standard version difference automatic comparison and revision impact analysis system 20, applied to the above-mentioned standard version difference automatic comparison and revision impact analysis method, including: Processor 201.

[0109] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, they implement the standard version difference automatic comparison and revision impact analysis method as described in the method embodiment.

[0110] The standard version difference automatic comparison and revision impact analysis system 20 provided by the present invention can perform the above-mentioned standard version difference automatic comparison and revision impact analysis method and achieve the same or similar technical effects. To avoid duplication, the present invention will not elaborate further.

[0111] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0112] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0113] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0114] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0115] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0116] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0117] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0119] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0120] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0122] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the standard version difference automatic comparison and revision impact analysis method as described in the method embodiment.

[0124] The present invention provides a computer-readable storage medium that can realize the steps and effects of the standard version difference automatic comparison and revision impact analysis method of the above method embodiments. To avoid repetition, the present invention will not repeat them.

[0125] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0126] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0127] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0128] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0129] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for automatic comparison and revision impact analysis of standard version differences, characterized in that, include: S1: Obtain the old and new character sequences of the standard draft for approval; S2: Construct the position difference range of the approval document based on the old version character sequence and the new version character sequence; S3: Collect the annotation coordinate sequence and annotation suggestion text reviewed by the standard review experts; S4: Establish an annotation adjustment mapping sequence based on the annotation coordinate sequence, the annotation suggestion text, and the position difference range of the submitted document; S5: Adjust the mapping sequence according to the annotation, and determine the conformity judgment sequence within the annotation through vector dot product comparison operation; S6: Generate an external reasonableness judgment sequence based on the internal compliance judgment sequence of the annotation; S7: Based on the aforementioned annotation rationality judgment sequence, generate standard version difference automatic comparison and revision impact analysis text.

2. The method for automatic comparison and revision impact analysis of standard version differences according to claim 1, characterized in that, The specific range of positional differences in the submitted documents includes: the starting point of the difference, the ending point of the difference, and the magnitude of the change.

3. The method for automatic comparison and revision impact analysis of standard version differences according to claim 1, characterized in that, S2 specifically includes: S201: Perform feature encoding parsing and mapping on the old character sequence and the new character sequence to obtain a vector space distribution topological hierarchical network; S202: Generate the old version of the text feature matrix and the new version of the text feature matrix through the vector space distribution topology hierarchical network; S203: Based on the old version of the text feature matrix and the new version of the text feature matrix, perform feature level alignment difference comparison on the same row dimension of the matrix, extract the state offset amplitude between corresponding vector nodes, and obtain the row-level feature deviation map. S204: Based on the row-level feature deviation map, retrieve the coordinates of the out-of-bounds row number; S205: Aggregate the coordinates of the out-of-bounds row numbers across consecutive row segment boundaries to obtain a coordinate distribution topology network structure; S206: Based on the coordinate distribution topology network structure, construct the position difference interval of the submitted document.

4. The method for automatic comparison and revision impact analysis of standard version differences according to claim 1, characterized in that, S4 specifically includes: S401: Perform spatial dimension feature mapping and alignment on the line number coordinates in the position difference interval of the submitted manuscript and the boundary coordinates in the annotation coordinate sequence to establish a coordinate comparison benchmark network; S402: Using the coordinate comparison benchmark network, perform a range intersection state discrimination operation on the row number coordinates and the boundary coordinates to obtain an interval state classification set; S403: Based on the overlapping interval elements and independent interval elements in the interval state classification set, merge the annotation suggestion text and establish a mapping association hierarchical topology architecture; S404: Based on the mapping association hierarchical topology, establish the annotation adjustment mapping sequence.

5. The method for automatic comparison and revision impact analysis of standard version differences according to claim 4, characterized in that, S5 specifically includes: S501: Adjust the mapping sequence according to the annotation, and retrieve overlapping interval elements in the mapping association hierarchical topology; S502: Extract the new text feature matrix and the feature expression entity corresponding to the annotation suggestion text from the elements of the overlapping interval; S503: Perform feature space vector distribution hierarchical mapping on the feature representation entity to generate a set of interval feature distribution states; S504: Based on the set of interval feature distribution states, perform a multi-dimensional comparison of the vector dot product space between the new text feature matrix and the feature expression entity to obtain coordinate nodes; S505: Perform state marking and filtering retention on the coordinate nodes to establish a decision node network; S506: Obtain the baseline out-of-bounds coordinate sequence through the aforementioned decision node network; S507: Determine the annotation compliance determination sequence based on the baseline out-of-bounds coordinate sequence.

6. The method for automatic comparison and revision impact analysis of standard version differences according to claim 1, characterized in that, S6 specifically includes: S601: Extract the internal independent interval elements according to the conformity determination sequence in the annotation; S602: Retrieve the old character sequence and the new character sequence based on the spatial mapping coordinates of the independent interval elements; S603: Perform word segmentation encoding and spatial feature transformation on the old version character sequence to obtain the old version word element distribution state set; S604: Perform a word space mapping transformation on the new character sequence based on the old version word distribution state set; S605: Perform a cross-entropy hierarchical distribution difference comparison on the nodes in the converted new character sequence and the old word distribution state set to obtain a cross-entropy state difference map; S606: Retrieve spatial coordinates based on the cross-entropy state difference map; S607: Perform bottom-level node sequence splicing on the spatial coordinates to generate the annotation rationality judgment sequence.

7. The method for automatic comparison and revision impact analysis of standard version differences according to claim 1, characterized in that, The standard version difference automatic comparison and revision impact analysis text specifically includes: a statement of the reasons for the revision, a record of review and debate, and a finalized conclusion.

8. The method for automatic comparison and revision impact analysis of standard version differences according to claim 1, characterized in that, Specifically, S7 includes: S701: Perform joint selection and aggregation processing of overlapping interval elements and independent interval elements on the underlying spatial mapping architecture of the annotation rationality judgment sequence to establish a global correlation topology structure for interval coordinate distribution; S702: Obtain the aggregated set of entities across the entire region; S703: Based on the global interval entity aggregation set and the global association topology structure of the interval coordinate distribution, establish a joint topology network of text nodes; S704: Generate a multi-source text association element set through the joint topology network of the text nodes; S705: Based on the multi-source text association element set, construct a continuous text underlying closed-loop serial topology architecture; S706: Based on the underlying closed-loop serial topology of the continuous text, generate the standard version difference automatic comparison and revision impact analysis text.

9. A standard version difference automatic comparison and revision impact analysis system, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the standard version difference automatic comparison and revision impact analysis method as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the standard version difference automatic comparison and revision impact analysis method as described in any one of claims 1 to 8.

Images