A layout element semantic constraint intelligent generation method

Abstract

The present application relates to computer layout generation, graphical interface typesetting and layout content organization technical field, specifically relates to a kind of layout element semantic constraint intelligent generation method.The method obtains the history layout state sequence and current layout generation input of to be generated layout, constructs history layout relationship atlas sequence, determines reference semantic binding relationship, authorized relationship change set and layout element semantic constraint set;Candidate layout state data is generated, abnormal relationship editing set, cross-time abnormal editing density and target relationship local subgraph are determined, and then the counterfactual binding vulnerability, competition subject isolation gap and competition attribution channel set are determined;Accordingly, candidate layout state data is isolated and regulated and iteratively generated, and target layout is output, to improve the stability of subject attachment relationship.

Description

Technical Field

[0001] This invention relates to the fields of computer layout generation, graphical interface typesetting, and layout content organization; more specifically, it relates to an intelligent generation method for semantic constraints of layout elements. Background Technology

[0002] As the types and quantities of digital content continue to increase, data dashboards, electronic reports, information cards, mobile interfaces, web pages, and promotional displays often need to be automatically generated or dynamically adjusted based on content length, number of elements, display size, and reading order. Existing layout generation technologies typically employ preset templates, layout rules, geometric constraints, or sample-based learning methods to handle the position, size, spacing, alignment, container boundaries, and overlap of layout elements. These methods are effective in preventing elements from going out of bounds, reducing element overlap, maintaining basic alignment, and adapting to different canvas sizes.

[0003] Some elements on a page layout do not exist independently. For example, captions need to be attached to corresponding images, unit elements need to be attached to corresponding numerical values ​​or data charts, risk warnings need to be attached to relevant information areas, and operation buttons need to be attached to corresponding operation objects. These subordinate page elements, in addition to geometric proximity, also have semantic relationships with the main page elements, such as content ownership, information hierarchy, reading order, container ownership, group boundaries, and visual ownership. When the page layout remains stable, these relationships can usually be maintained through template positions, fixed groupings, or preset associations.

[0004] In the event of asynchronous content updates, additions or deletions of layout elements, list reordering, canvas size changes, or iterative layout generation, the relative positions, container structures, and reading paths between layout elements may continuously change. While a subordinate layout element may still meet geometrically valid requirements such as non-overlapping and no boundary crossings, its visual position may gradually approach another main layout element, or it may form a stronger alignment, container sharing, or reading path continuity relationship with another main layout element. In this case, the subordinate layout element may still be geometrically valid, but its primary dependency relationship may shift from the original main layout element to other main layout elements, causing errors in units, captions, warnings, or the corresponding operation objects.

[0005] Existing technologies can typically examine geometric constraints or explicitly configured static relationships based on the current layout state, and some solutions can also evaluate the rationality of the layout based on element type and content information. However, for layouts that undergo multiple dynamic adjustments, judging solely based on the current layout state makes it difficult to distinguish between changes in authorization relationships resulting from normal content updates and unauthorized changes in subject dependency relationships. Furthermore, in situations where a candidate layout has not yet experienced obvious binding errors, but a slight positional disturbance can cause a reversal of subject dependency relationships, conventional single-layout evaluations also struggle to quantify the fragility of such semantic binding relationships.

[0006] Furthermore, when a subordinate element is adjacent to multiple main elements, different pieces of visual attribution evidence may collectively influence the subject attachment relationship. Some evidence strengthens the connection between the subordinate element and the original main elements, while other evidence strengthens its connection with competing main elements. Simply increasing the spacing between elements or adjusting a single alignment may increase unnecessary layout changes or may not address the main channels of evidence that lead to incorrect visual attribution.

[0007] Therefore, the intelligent generation of semantic constraints for page elements still needs to solve the following technical problems: how to identify cross-temporal abnormal relationship editing in candidate pages by combining historical page state sequences and changes in authorization relationships during the dynamic changes of the page; how to detect the potential binding flip risk of subordinate page elements relative to competing page elements while the geometric legality conditions remain valid; and how to implement isolation control with a small scope and clear direction on the candidate page state data based on counterfactual binding vulnerability, competitive entity isolation gap, and competitive attribution channel set to obtain target pages that meet the semantic constraints of page elements. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent method for generating semantic constraints of page elements.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligently generating semantic constraints for page layout elements, comprising: S1. Obtain the historical page state sequence and the current page generation input. Construct a historical page relationship graph sequence based on the historical page state sequence. Determine the baseline semantic binding relationship, authorization relationship change set, and page element semantic constraint set from the historical page relationship graph sequence. S2. Generate candidate page status data based on the current page output, and determine the abnormal relationship editing set, abnormal relationship editing energy, cross-temporal abnormal editing density, and target relationship local subgraph based on the candidate page status data, historical page relationship graph sequence, and authorized relationship change set; S3. Based on the target relationship local subgraph and candidate page status data, determine the subordinate page elements, the baseline main page elements, the candidate competing main page elements, the visual attribution evidence vector, and the counterfactual binding failure domain; S4. Determine the counterfactual binding vulnerability based on the counterfactual binding failure domain, and determine the subject's visual attribution confidence, target competing subject's layout elements, competing subject's isolation gap, and competing attribution channel set based on the visual attribution evidence vector; S5. Based on the cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive subject isolation gap, and competitive attribution channel set, isolate, regulate, and iteratively generate the candidate page state data until the candidate page state data meets the constraints in the page element semantic constraint set, and output the target page.

[0010] In one embodiment, the baseline semantic binding relationship includes the subject dependency relationship between subordinate page elements and baseline main page elements, and the set of semantic constraints for page elements includes subject dependency constraints, content ownership constraints, information hierarchy constraints, reading order constraints, container ownership constraints, grouping boundary constraints, visual ownership constraints, and geometric legal constraints.

[0011] In one embodiment, step S2 includes: generating candidate page state data based on the current page generation input, and constructing a candidate page relationship graph based on the candidate page state data; comparing the candidate page relationship graph with the historical page relationship graph sequence for relationship editing, and obtaining an abnormal relationship editing set using the authorized relationship change set; determining the abnormal relationship editing energy and cross-temporal abnormal editing density based on the abnormal relationship editing set, and determining the target relationship local subgraph based on the cross-temporal abnormal editing density.

[0012] In one embodiment, the candidate page layout relationship graph is constructed based on the page element set, element semantic roles, element content ownership identifiers, element geometric attributes, reading order, container boundaries, and grouping boundaries in the candidate page layout state data; the cross-temporal abnormal editing density is determined by the relationship abnormal editing energy, and the target relationship local subgraph is a local graph that has reached the preset abnormal condition of cross-temporal abnormal editing density and contains subordinate page elements.

[0013] In one embodiment, in step S3, candidate competing subject page elements are determined by subject page elements that have visual affiliation competition with subordinate page elements in the candidate page relationship graph; an ordered counterfactual perturbation trajectory and a set of counterfactual perturbation pages are generated based on the visual affiliation evidence vector; perturbation pages that satisfy geometric legality constraints are screened from the set of counterfactual perturbation pages; when the subject dependency relationship in the perturbation page is flipped from the baseline subject page element to the candidate competing subject page element, the corresponding perturbation page is recorded as a counterfactual binding failure sample; and a counterfactual binding failure domain is formed from the counterfactual binding failure samples.

[0014] In one embodiment, the visual attribution evidence vector includes at least two evidence channels selected from normalized spacing, alignment consistency, container sharing, reading path continuity, and group boundary transparency.

[0015] In one embodiment, the counterfactual binding vulnerability is determined as follows: the failure domain measure, minimum failure radius, and flip duration coefficient are calculated based on the counterfactual binding failure domain; the failure domain measure, minimum failure radius, flip duration coefficient, and cross-temporal anomaly edit density are nonlinearly coupled to obtain the counterfactual binding vulnerability.

[0016] In one embodiment, the subject visual attribution confidence of subordinate page elements relative to the baseline subject page element and candidate competing subject page elements is calculated based on the visual attribution evidence vector, and the subject attribution confidence difference is determined based on the subject visual attribution confidence. Candidate competing subject page elements whose subject attribution confidence difference reaches the preset competition condition are determined as target competing subject page elements, and the competing subject isolation gap is determined based on the subject visual attribution confidence of the target competing subject page element, the subject visual attribution confidence of the baseline subject page element, and the preset safety interval.

[0017] In one embodiment, the competitive attribution channel set is formed as follows: each evidence channel in the visual attribution evidence vector is subjected to single-channel suppression, the contribution value of each evidence channel to the isolation gap of the competing subject is calculated, and the evidence channels are selected from largest to smallest according to the contribution value to form the competitive attribution channel set.

[0018] In one embodiment, step S5 includes: when the counterfactual binding vulnerability reaches a preset vulnerability condition or the competitive subject isolation gap reaches a preset isolation condition, determining a minimum isolation control vector based on the cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive subject isolation gap, and competitive attribution channel set; dynamically controlling and iteratively generating candidate page state data based on the minimum isolation control vector until the constraints in the page element semantic constraint set meet the preset output conditions, and outputting the target page.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a historical page relationship graph sequence through a historical page state sequence, and performs cross-temporal relationship analysis on candidate page state data by combining a set of baseline semantic binding relationships and authorized relationship changes. This can distinguish between normal page adjustments and unauthorized semantic binding changes, improve the accuracy of identifying abnormal subject dependency relationships under conditions of asynchronous content updates, page element additions and deletions, list reordering, and canvas size changes, and reduce the possibility of misjudging normal relationship changes. By locating the page area with visual attribution competition through the local subgraph of the target relationship, and combining the counterfactual binding failure domain, counterfactual binding vulnerability, subject visual attribution confidence and competitive subject isolation gap, risk assessment is performed on candidate pages that have not yet had obvious subject attachment errors but are close to the binding flip boundary. It can identify page states that are geometrically legal but semantically unstable, and determine the target competing subject page elements and the set of competing attribution channels that cause erroneous visual attribution. Based on cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive subject isolation gap, and competitive attribution channel set, isolation control and iterative generation are implemented on candidate page state data. This can concentrate the control scope on page elements and evidence channels that affect subject dependency relationships, reduce page changes unrelated to semantic binding risks, and improve the stability of subject dependency relationships between subordinate page elements and benchmark subject page elements while ensuring that all constraints in the semantic constraint set of page elements are satisfied. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the overall process of a method for intelligently generating semantic constraints for page elements according to the present invention. Figure 2 This is a flowchart illustrating the determination of the historical layout relationship graph sequence, the baseline semantic binding relationship, the set of authorization relationship changes, and the set of semantic constraints for layout elements in this invention. Figure 3 This is a flowchart illustrating the determination of counterfactual binding vulnerability in this invention. Figure 4 This is a flowchart illustrating the determination of the visual attribution confidence of the main body, the layout elements of the target competing body, and the isolation gap of the competing body in this invention. Detailed Implementation

[0021] Reference Figure 1 A method for intelligently generating semantic constraints for page layout elements, comprising: Reference Figure 2 S1. Obtain the historical page state sequence and the current page generation input. Construct a historical page relationship graph sequence based on the historical page state sequence, and determine the baseline semantic binding relationship, authorization relationship change set, and page element semantic constraint set from the historical page relationship graph sequence. The historical page state sequence records the existence state, semantic role, content ownership, geometric attributes, and organizational relationship between page elements at different time nodes. The historical page relationship graph sequence organizes the scattered page states at each time node into a continuous relationship structure with element nodes and relationship connections. The baseline semantic binding relationship defines the normal subject dependency relationship between subordinate page elements and the baseline main page elements. The authorized relationship change set delineates the relationship changes that are allowed due to content updates, element additions or deletions, list sorting, or page size changes. The page element semantic constraint set forms the subject dependency, content affiliation, information hierarchy, reading order, container affiliation, grouping boundaries, visual affiliation, and geometric legality requirements that candidate pages must adhere to. Through the historical reference established in this step, normal page changes and unauthorized semantic binding changes have a clear distinguishing boundary.

[0022] The process involves obtaining the historical layout state sequence formed by the layout to be generated at three consecutive time nodes, as well as the current layout generation input. The historical layout state record at each time node includes the set of layout elements, stable element identity identifiers, element semantic roles, element information levels, element content ownership identifiers, element geometric attributes, element stacking order, text line break parameters, visibility status, reading order, container boundaries, and grouping boundaries. Based on the stable element identity identifiers, the process matches the same layout elements between adjacent time nodes, treating the layout elements as element nodes, and using subject dependency relationships, content ownership relationships, reading order relationships, container ownership relationships, grouping relationships, visual adjacency relationships, and geometric position relationships as relationship edges. The process arranges each element node and relationship edge according to the time node to construct a historical layout relationship graph sequence. For subordinate page elements in the historical page relationship graph sequence, select the main page elements that maintain the same content ownership identifier and the same subject attachment direction at two or more consecutive time points and are not covered by authorization relationship change records. These are identified as the baseline main page elements, and the subject attachment relationship between subordinate page elements and the baseline main page elements is identified as the baseline semantic binding relationship. Read the element addition records, element deletion records, content replacement records, list sorting records, container size adjustment records, and relationship change permission records carried by the current page generation input. Form an authorization relationship change set according to the element identity identifier, relationship type, change direction, and effective time range in the relationship change permission records. A set of semantic constraints for layout elements is established around the baseline semantic binding relationship. Among them, the subject dependency constraint limits the baseline subject layout element to which the subordinate layout element should be attached; the content affiliation constraint limits the consistency of content source between the subordinate layout element and the baseline subject layout element; the information hierarchy constraint limits the master-slave relationship between the two in the layout information hierarchy; the reading order constraint limits the reading position of the subordinate layout element relative to the baseline subject layout element; the container affiliation constraint limits the container to which the two belong; the group boundary constraint limits the permission conditions for crossing the group boundary; the visual affiliation constraint limits the affiliation relationship between the two in terms of spacing, alignment, reading path and boundary transparency; and the geometric legality constraint limits the layout elements to not overlap, not cross the boundary and have a minimum visible range. For example, the historical layout sequence includes data chart A, data chart B, percentage unit elements, and risk warning elements. At all three time points, the percentage unit elements and risk warning elements possess content ownership identifiers and subject dependency relationships pointing to data chart A. The current layout generation input records that data chart A and data chart B have adjusted positions due to changes in canvas width, but does not record permission for changes in subject dependency relationships. Therefore, the position adjustment is included in the set of permission relationship changes, and the subject dependency relationship between the percentage unit elements and risk warning elements and data chart A is still determined as the baseline semantic binding relationship. Even if the position adjustment causes the percentage unit elements to be geometrically closer to data chart B, the semantic constraint set of layout elements still retains its subject dependency constraints, content ownership constraints, and visual ownership constraints on data chart A, providing a stable reference for identifying cross-temporal semantic binding drift.

[0023] S2. Generate candidate page status data based on the current page generation input, and determine the abnormal relationship editing set, abnormal relationship editing energy, cross-temporal abnormal editing density, and target relationship local subgraph based on the candidate page status data, historical page relationship graph sequence, and authorized relationship change set; the candidate page status data records the actual status of page elements and element relationships in the current generation round, and compares the relationship editing between the candidate page relationship graph formed by the candidate page status data and the historical page relationship graph sequence; During the comparison process, relationship additions, deletions, and changes in relationship pointers that conform to the authorized relationship change set are excluded from the abnormal range. The remaining relationship changes are categorized into the abnormal relationship editing set. The abnormal relationship editing energy reflects the degree of change in abnormal relationship editing in terms of relationship type, semantic level, and scope of influence, while the cross-temporal abnormal editing density reflects the degree of concentration of abnormal relationship editing in the local page relationship structure. Regions where the cross-temporal abnormal editing density reaches the preset abnormality condition are identified as local subgraphs of the target relationship. The scope of the inspection converges from the entire page to local relationship structures where semantic binding drift may occur. This step can identify subject dependency relationship anomalies in candidate pages when geometric conditions such as position not exceeding boundaries and element not overlapping are met.

[0024] The system reads the layout content, canvas size, element addition information, element deletion information, content replacement information, and layout arrangement requirements recorded in the current layout generation input to generate candidate layout state data. This data includes a set of layout elements and the stable element identity identifiers, semantic roles, content ownership identifiers, geometric attributes, reading order, container boundaries, and grouping boundaries for each element. Layout elements are treated as element nodes. The stable element identity identifiers, semantic roles, content ownership identifiers, and geometric attributes are recorded in the corresponding element nodes. Subject dependency relationships, content ownership relationships, reading order relationships, container ownership relationships, grouping relationships, visual adjacency relationships, and geometric position relationships are recorded as relationship edges between element nodes, constructing a candidate layout relationship graph. Based on the stable element identity identifiers, element nodes in the candidate layout relationship graph are matched against corresponding element nodes in the historical layout relationship graph sequence across time. For relationship edges that persist in the historical page relationship graph sequence but disappear in the candidate page relationship graph, it is recorded as a relationship deletion edit. For relationship edges that newly appear in the candidate page relationship graph, it is recorded as a relationship addition edit. For cases where the relationship edge corresponding to a subordinate page element changes from the base page element to other page elements, it is recorded as a relationship pointing edit. For cases where the reading order, container affiliation, or group affiliation changes, it is recorded as a relationship attribute edit. Each relationship edit is matched item by item with the identity identifier, relationship type, change direction, and effective time range of stable elements in the authorized relationship change set. Relationship edits that match successfully are included in the allowed changes set, and relationship edits that do not match successfully form the abnormal relationship edit set. For each abnormal relationship edit in the abnormal relationship editing set, the relationship type weight is determined according to the relationship type, where the relationship type weight corresponding to subject dependency relationship and content affiliation relationship is higher than that corresponding to geometric position relationship; the semantic influence weight is determined according to the semantic role and information level of the elements involved in the abnormal relationship edit; the historical stability weight is determined according to the continuous maintenance time of the edited relationship in the historical page relationship graph sequence; the product of the relationship type weight, semantic influence weight and historical stability weight is determined as the relationship abnormal editing energy of the corresponding abnormal relationship edit. Centered on each subordinate page element, extract the element nodes, relationship edges, container boundaries, and grouping boundaries that have two-level relationships with that subordinate page element to form a relational subgraph. Divide the sum of the relational abnormal editing energies corresponding to each abnormal relationship edit within the relational subgraph by the number of editable relationship edges within the relational subgraph to obtain the transtemporal abnormal editing density of that relational subgraph. When the transtemporal abnormal editing density reaches a preset abnormality condition and the relational subgraph contains subordinate page elements, that relational subgraph is determined as the target relational subgraph. For example, the historical layout relationship graph sequence records that percentage unit elements are continuously attached to data chart A. The current layout generation input only authorizes data chart A and data chart B to adjust element geometric attributes due to changes in canvas width. However, the candidate layout status data arranges the percentage unit elements near data chart B, deletes the primary attachment relationship of the percentage unit elements pointing to data chart A in the candidate layout relationship graph, and simultaneously adds a primary attachment relationship of percentage unit elements pointing to data chart B. Since the set of authorized relationship changes does not include the deletion and addition of this primary attachment relationship, both relationship edits are included in the abnormal relationship edit set. Subject dependency relationships have a high relation type weight. The relationship between percentage unit elements and data chart A remains consistent across multiple historical time points, thus increasing the historical stability weight. Local subgraphs containing percentage unit elements, data chart A, data chart B, and adjacent container boundaries exhibit high cross-temporal anomaly editing density and are identified as target relational subgraphs when preset anomaly conditions are met. This processing preserves evidence of cross-temporal drift in subject dependency relationships on local layouts, even when no geometric errors such as element overlap or boundary crossing occur in the candidate layout.

[0025] S3. Based on the target relationship subgraph and candidate layout state data, determine subordinate layout elements, baseline subject layout elements, candidate competing subject layout elements, visual attribution evidence vectors, and counterfactual binding failure domains; analyze subject dependency relationships and visual attribution competition relationships around the target relationship subgraph. Determine subordinate layout elements and their corresponding baseline subject layout elements based on the baseline semantic binding relationships, and select subject layout elements that compete visually with subordinate layout elements from the candidate layout relationship graph to form candidate competing subject layout elements; For subordinate page elements, baseline main page elements, and candidate competing main page elements, at least two evidence channels are extracted from normalized spacing, alignment consistency, container sharing, reading path continuity, and group boundary transparency to construct a visual attribution evidence vector. An ordered counterfactual perturbation trajectory and a set of counterfactual perturbation pages are generated along the evidence change direction corresponding to the visual attribution evidence vector. Within the perturbation pages that satisfy geometric legality constraints, counterfactual binding failure samples are identified where the subject dependency relationship is flipped from the baseline main page element to a candidate competing main page element. All counterfactual binding failure samples collectively form a counterfactual binding failure domain, reflecting the range of erroneous flipping of the current subject dependency relationship under different perturbation directions and amplitudes.

[0026] Read the element nodes, relationship edges, container boundaries, and grouping boundaries in the target relationship subgraph. Combine the element semantic roles and element content ownership identifiers in the candidate page status data to identify the element nodes that are responsible for unit descriptions, figure captions, risk warnings, or operation instructions and need to be attached to other page elements as subordinate page elements. Based on the baseline semantic binding relationship, find the element nodes that maintain a subject-dependent relationship with the subordinate page elements and identify the found element nodes as the baseline subject page elements. After excluding the baseline main page elements, extract the main page elements from the target relationship local subgraph that are in the same container or adjacent container as the subordinate page elements, are adjacent in reading order, have a normalized spacing lower than the preset distance threshold, or have horizontal or vertical alignment relationships. The main page elements that meet at least one of these conditions are identified as candidate competing main page elements. For subordinate page elements, the baseline main page element, and each candidate competing main page element, a visual attribution evidence vector is calculated. The normalized spacing is determined by dividing the minimum edge distance between the subordinate page element and the corresponding main page element by the diagonal length of the area jointly enclosed by the two. The alignment consistency is determined by the alignment axis with the smallest deviation among the left edge, right edge, horizontal center line, top edge, bottom edge, and vertical center line of the two. Container sharing is determined by the smallest container to which both belong and the difference in container hierarchy. A higher value is used when they belong to the same smallest container, and the value decreases as the difference in container hierarchy increases, according to a preset attenuation ratio. Reading path continuity is determined by the number of page elements located between the two in the reading order; the fewer the number of page elements, the higher the reading path continuity. Group boundary transparency is determined by the number of line segments connecting the geometric centers of the two crossing the group boundary and the boundary level of each group boundary. Each evidence channel is normalized to the same numerical range and arranged in a fixed channel order to form a visual attribution evidence vector. The perturbation direction is to weaken the visual attribution evidence of subordinate page elements relative to the benchmark main page elements and strengthen the visual attribution evidence of subordinate page elements relative to candidate competing main page elements. Under the condition that the page element content, stable element identity and element semantic role remain unchanged, the element geometric attributes, alignment relationship, reading position or group boundary relationship of subordinate page elements are adjusted according to the preset perturbation step size. Each perturbation state is arranged from small to large according to the change of visual attribution evidence vector, generating an ordered counterfactual perturbation trajectory, and mapping each perturbation state in the ordered counterfactual perturbation trajectory to the corresponding perturbation page, forming a set of counterfactual perturbation pages. Each counterfactual perturbation layout is examined for geometric legality constraints, retaining layouts without element overlap, boundary violations, invisible areas, or illegal container affiliation. For each retained perturbation layout, a layout relationship graph is reconstructed, and the subject visual affiliation confidence of subordinate layout elements relative to the baseline subject layout element and each candidate competing subject layout element is calculated. The value obtained by subtracting the normalized spacing from one is taken as the distance affiliation component. Alignment consistency, container sharing, reading path continuity, and group boundary transparency maintain their original numerical directions. The larger the value of each affiliation component, the higher the degree of visual affiliation between the subordinate layout element and the corresponding subject layout element. The set of evidence channel indices used in the calculation is denoted as […]. , No. The attribution component corresponding to each evidence channel is denoted as follows: The weight of evidence is denoted as Subordinate layout elements Compared to the main layout elements Subject visual attribution confidence Determined according to the following formula: in, ,and: The basic main page element is denoted as The main elements of the candidate competition page are denoted as The confidence difference in subject attribution between candidate competing subject page elements and the benchmark subject page elements. Determined according to the following formula: When a candidate competing main element achieves the highest visual confidence score among all main elements, and the corresponding confidence difference in main element attribution is not lower than the binding determination threshold... When the subject dependency relationship in the perturbed layout is flipped from the baseline subject layout element to the candidate competing subject layout element, the corresponding perturbed layout is identified as a counterfactual binding failure sample. When multiple subject layout elements achieve the same highest subject visual attribution confidence, if the baseline subject layout element is among the parallel subject layout elements, the original subject dependency relationship is maintained; if the baseline subject layout element is not among the parallel subject layout elements, the subject layout element with the longest continuous subject dependency relationship in the historical layout relationship graph sequence is selected.

[0027] Adjacent counterfactual binding failure samples are connected based on the difference in evidence channels between perturbation states. Counterfactual binding failure samples whose evidence channel difference does not exceed the corresponding counterfactual perturbation step size are grouped into the same continuous failure region, forming a counterfactual binding failure domain from all continuous failure regions. When there are no counterfactual binding failure samples in the counterfactual perturbation pattern set, the counterfactual binding failure domain is marked as empty, the failure domain measure is set to zero, the normalized minimum failure radius is set to one, the flipping persistence coefficient is set to zero, and the counterfactual binding vulnerability is determined according to a nonlinear coupling method. When there are no perturbation patterns satisfying geometrically valid constraints in an ordered counterfactual perturbation trajectory, this ordered counterfactual perturbation trajectory does not participate in the calculation of the failure domain measure, minimum failure radius, and flipping persistence coefficient.

[0028] For example, a percentage unit element is attached to data chart A in a candidate layout, and data chart B is located to the right of the percentage unit element and has a high degree of horizontal alignment with it. Data chart B is identified as a candidate competing main layout element. The percentage unit element is moved in a direction closer to data chart B and away from data chart A, with a perturbation step of two layout coordinate units, while keeping the percentage unit element from going out of the container boundary and from overlapping with other layout elements. When moving to the sixth page coordinate unit, the subject attachment relationship corresponding to the percentage unit element still points to data chart A; when moving to the eighth page coordinate unit, the subject attachment relationship shifts to data chart B, and the corresponding perturbed page is identified as a counterfactual binding failure sample. Multiple counterfactual binding failure samples obtained along the same perturbation direction are adjacent to each other in the perturbation state, and the continuous region formed by these counterfactual binding failure samples constitutes the counterfactual binding failure domain corresponding to the percentage unit element.

[0029] Reference Figure 3 , Figure 4 S4. Determine the vulnerability of counterfactual binding based on the counterfactual binding failure domain, and determine the confidence of the subject's visual attribution, the target competing subject's layout elements, the competing subject's isolation gap, and the set of competing attribution channels based on the visual attribution evidence vector; quantify the semantic binding flip risk and the degree of visual attribution competition. The failure domain measure, minimum failure radius, and flip duration coefficient are calculated based on the counterfactual binding failure domain. The failure domain measure reflects the failure range formed by the counterfactual binding failure sample; the minimum failure radius reflects the minimum disturbance required for the current layout state to reach the subject dependency relationship flip boundary; and the flip duration coefficient reflects the degree of preservation of the subject dependency relationship in continuous disturbance states after the flip. The counterfactual binding vulnerability is obtained by nonlinearly coupling the failure domain measure, minimum failure radius, flip duration coefficient, and cross-temporal abnormal editing density. Based on the visual attribution evidence vector, the subject visual attribution confidence scores of subordinate page elements relative to the baseline subject page element and candidate competing subject page elements are calculated respectively, and the subject attribution confidence difference is determined based on the two types of subject visual attribution confidence scores. Candidate competing subject page elements whose subject attribution confidence differences reach the preset competition conditions are determined as target competing subject page elements. The competing subject isolation gap is determined by combining the subject visual attribution confidence scores of the target competing subject page elements, the subject visual attribution confidence scores of the baseline subject page elements, and the preset safety interval. Single-channel suppression is applied to each evidence channel in the visual attribution evidence vector. The contribution value of each evidence channel is calculated based on the change in the isolation gap between competing subjects before and after suppression. Evidence channels are selected from largest to smallest contribution value to form a set of competing attribution channels. This step not only determines whether the subject dependency relationship is already erroneous, but also reveals the vulnerable state that is close to the flip boundary but has not yet been erroneous, and locates the main evidence channels that cause visual attribution competition.

[0030] Read the counterfactual binding failure domain formed in step S3, count the number of counterfactual binding failure samples within the counterfactual binding failure domain and the number of perturbed pages in the counterfactual perturbed page set that satisfy geometrically valid constraints, and determine the ratio of the number of counterfactual binding failure samples to the number of perturbed pages that satisfy geometrically valid constraints as the failure domain measure; use the visual attribution evidence vector corresponding to the candidate page state data as the original evidence point, and calculate the weighted Euclidean distance between the original evidence point and the evidence point corresponding to each counterfactual binding failure sample, wherein the weight of each evidence channel is determined according to the stability of the corresponding evidence channel in the historical page relationship graph sequence, and determine the minimum value among all weighted Euclidean distances as the minimum failure radius; Along each ordered counterfactual perturbation trajectory, find the perturbation position where the subject dependency relationship is first flipped from the baseline subject page element to the candidate competing subject page element. Count the number of consecutive perturbation pages that remain flipped after the perturbation position, and calculate the proportion of the number of consecutive perturbation pages to the number of perturbation pages that satisfy the geometric legal constraints after the perturbation position. This yields the trajectory duration value of the corresponding ordered counterfactual perturbation trajectory. The trajectory persistence values ​​of each ordered counterfactual perturbation trajectory are weighted and averaged according to the number of counterfactual binding failure samples to obtain the flipping persistence coefficient. The failure domain measure, minimum failure radius, flipping persistence coefficient, and cross-temporal anomaly editing density are normalized to a numerical range of zero to one. The normalized minimum failure radius is reversed, and one minus the normalized minimum failure radius is taken as the radius vulnerable component. The counterfactual binding vulnerability is determined by subtracting the product of the failure domain measure complement, the radius vulnerability component complement, the flip duration coefficient complement, and the transtemporal anomaly editing density complement. This nonlinear coupling method improves the counterfactual binding vulnerability when the failure domain is large, the flip boundary distance is short, the flip state is maintained for a long time, or the anomaly relationship editing is concentrated, while retaining the coupling increment generated when multiple risk components increase simultaneously. For the visual attribution evidence vector between subordinate page elements and the baseline main page elements, and the visual attribution evidence vector between subordinate page elements and each candidate competing main page element, the value obtained by subtracting the normalized spacing from one is used as the distance attribution component, and alignment consistency, container sharing, reading path continuity, and group boundary transparency are used as the remaining attribution components. The evidence weights are configured according to the attribution stability of each evidence channel in the historical page state sequence, and the weighted sum of each attribution component is performed. The main visual attribution confidence between zero and one is obtained through numerical range compression. The subject attribution confidence score is obtained by subtracting the subject visual attribution confidence score of the benchmark subject page element from the subject visual attribution confidence score of the candidate competing subject page element. When the subject attribution confidence score reaches the preset competition condition, the corresponding candidate competing subject page element is included in the competing subject set, and the candidate competing subject page element with the highest subject visual attribution confidence score in the competing subject set is determined as the target competing subject page element. The competing subject isolation gap is determined by subtracting the subject visual attribution confidence score of the benchmark subject page element from the sum of the subject visual attribution confidence score of the target competing subject page element and the preset safety interval. If the calculation result is less than zero, the competing subject isolation gap is set to zero. For each evidence channel in the visual attribution evidence vector, the attribution component of the current evidence channel is replaced with a preset neutral value, while keeping the other evidence channels unchanged. The subject visual attribution confidence and competitive subject isolation gap of the target competing subject layout elements and the baseline subject layout elements are recalculated. The competitive subject isolation gap before suppression is subtracted from the competitive subject isolation gap after suppression to obtain the contribution value of the current evidence channel to the competitive subject isolation gap. Evidence channels with a contribution value greater than zero are selected, arranged in descending order of contribution value, and evidence channels whose cumulative contribution value reaches a preset proportion of all positive contribution values ​​are selected from the first in the list to form a set of competitive attribution channels. The parameters involved in this implementation are determined based on the historical page state sequence of the page to be generated and confirmed page samples with the same page type, element semantic roles, and container hierarchy as the page to be generated. Confirmed page samples include stable page samples where subject dependency relationships are maintained correctly and abnormal page samples where subject dependency relationships have been incorrectly flipped. Each confirmed page sample records the historical page relationship graph, the state of page element semantic constraints, the density of cross-temporal abnormal edits, counterfactual binding vulnerability, subject attribution confidence difference, and competing subject isolation gap.

[0031] The weights of relationship types are determined based on the level of influence of different relationship types on subject dependency relationships. The weights of relationship types corresponding to subject dependency relationships and content attribution relationships are higher than those corresponding to reading order relationships, container attribution relationships, grouping relationships, visual adjacency relationships, and geometric position relationships. The semantic influence weights are determined based on the semantic roles and information levels of the elements involved in editing abnormal relationships. The historical layout relationship graph sequence includes... Each time point, and . No. The number of times that the relationship corresponding to the abnormal relationship edit maintains the same relationship type, relationship pointer, and relationship attributes at adjacent time points is denoted as . Historically stable weights Determined according to the following formula: The relation type weight, semantic influence weight, and historical stability weight are all limited to a numerical range of zero to one and remain unchanged during a single target layout generation.

[0032] No. The proportion of evidence channels that maintain their original subject attachment relationships and whose corresponding attribution component changes do not exceed the channel stability tolerance in the stable sample is denoted as follows: When the present tense is true: Weight of evidence Determined according to the following formula: all When all values ​​are zero, all evidence channels use the same evidence weight. The channel stability tolerance is the larger of the median absolute deviation of the change in the assigned component of the corresponding evidence channel in the stable layout sample and the change in the smallest resolvable channel.

[0033] When calculating the weighted Euclidean distance corresponding to the minimum failure radius, each evidence channel uses an evidence weight. The minimum failure radius is normalized based on the maximum weighted Euclidean distance between the original evidence point and the corresponding evidence point of the perturbation face satisfying geometric legal constraints within the current counterfactual perturbation range. If the normalization result exceeds one, it is rounded down to one. If there are counterfactually bound failure samples and the maximum weighted Euclidean distance is zero, the normalized minimum failure radius is set to zero. The cross-temporal anomaly editing density is determined by the ratio of the sum of the anomaly editing energies of each relation within the relational local subgraph to the number of editable relation edges. If the ratio exceeds one, it is rounded down to one. Both the normalized minimum failure radius and the cross-temporal anomaly editing density are within the numerical range of zero to one.

[0034] The threshold corresponding to the preset abnormal condition is denoted as The threshold corresponding to the preset vulnerability condition is denoted as The threshold corresponding to the preset isolation condition is denoted as The threshold corresponding to the preset competition condition is denoted as The binding determination threshold is denoted as And satisfy: Each threshold is selected within the historical value range of the corresponding indicator, with the minimum sum of the number of stable layout samples judged as abnormal and the number of abnormal layout samples judged as stable serving as the basis for setting the threshold; when there are multiple thresholds that achieve the same judgment result, the lower threshold is selected.

[0035] The preset safety interval is denoted as The preset safety interval is determined based on the lower limit of the difference between the confidence level of the subject visual attribution of the baseline subject element in the stable layout sample and the confidence level of the subject visual attribution of the highest candidate competing subject element. The preset neutral value used for single-channel suppression is the median of the attribution components of the corresponding evidence channel in the stable layout sample. The preset cumulative contribution ratio is greater than zero and not greater than one, and is the minimum ratio at which the subject attachment relationships in the stable layout sample remain correct after forming the competing attribution channel set according to the corresponding ratio.

[0036] The counterfactual perturbation step size is determined based on the minimum resolvable adjustment amount of the corresponding adjustable layout attribute, and the counterfactual perturbation range does not exceed the adjustable range of the corresponding layout element under geometric legality constraints and semantic constraints of the layout element. The perturbation amount corresponding to the risk change sensitivity is consistent with the counterfactual perturbation step size of the corresponding adjustable layout attribute. The preset maximum number of iterations is determined based on the allowed layout generation processing time and the single-round isolation control processing time.

[0037] If the historical page state sequence is insufficient to determine the aforementioned parameters, the page template parameters of the same page type are read; if no corresponding page template parameters exist, they are determined based on page samples with manually confirmed subject dependency relationships. During the generation of a target page, all weights, thresholds, preset safety intervals, preset neutral values, preset cumulative contribution ratios, counterfactual perturbation step size, perturbation amount, and preset maximum number of iterations remain unchanged.

[0038] When no candidate competitor page elements satisfy the preset competition conditions exist, the target competitor page elements are marked as empty, the competitor isolation gap is set to zero, the competition attribution channel set is marked as empty, and no isolation control is performed for the competitor isolation gap. When the counterfactual binding vulnerability reaches the preset vulnerability condition, the page elements and relationship types involved are determined based on the abnormal relationship editing set in the target relationship local subgraph, and the position, spacing, alignment axis position, reading order position, container inner margin, or group boundary position of the corresponding page elements are included in the adjustable page attributes. When no adjustable page attributes exist, the feasible vector set is marked as empty.

[0039] When there are target competing entity layout elements but the contribution value of each evidence channel to the competitive entity isolation gap is not greater than zero, all evidence channels in the visual attribution evidence vector are included in the competitive attribution channel set, and the control direction that cannot reduce the competitive entity isolation gap is excluded by the constraint check of the candidate isolation control vector.

[0040] For example, the failure domain measure corresponding to the percentage unit element is 0.4, the normalized minimum failure radius is 0.2, the flip persistence coefficient is 0.75, the cross-temporal abnormal editing density is 0.5, and the counterfactual binding vulnerability obtained according to the aforementioned nonlinear coupling method is 0.985; the confidence level of the percentage unit element relative to the subject visual attribution of data chart A is 0.62, and the confidence level relative to the subject visual attribution of data chart B is 0.58, with a subject attribution confidence difference of -0.04. Under the preset competition condition that the subject attribution confidence difference is not less than -0.05, data chart B is determined as the target competing subject layout element. When the preset safety interval is 0.15, the competitive subject isolation gap is 0.11. After implementing single-channel suppression on the alignment consistency evidence channel, the competitive entity isolation gap decreased by 0.07. After implementing single-channel suppression on the normalized spacing evidence channel, the competitive entity isolation gap decreased by 0.03. After implementing single-channel suppression on the container sharing degree evidence channel, the competitive entity isolation gap decreased by 0.01. When the contribution values ​​are arranged and the preset cumulative ratio is reached, the alignment consistency evidence channel and the normalized spacing evidence channel are combined into a competitive attribution channel set.

[0041] S5. Based on the cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive subject isolation gap and competitive attribution channel set, isolate, regulate and iteratively generate the candidate page state data until the candidate page state data meets the constraints in the page element semantic constraint set, and output the target page. When the counterfactual binding vulnerability reaches a preset vulnerability condition or the competitive entity isolation gap reaches a preset isolation condition, a minimum isolation control vector is determined by comprehensively considering the cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive entity isolation gap, and competitive attribution channel set. The minimum isolation control vector limits the control direction, control magnitude, and evidence channels involved in the control of candidate page state data, and the control scope is concentrated on page relationships that cause incorrect visual attribution of subordinate page elements; After completing one round of isolation adjustment, new layouts are generated and relationships are checked based on the updated candidate layout status data. The cross-temporal abnormal editing density, counterfactual binding vulnerability, and competitive subject isolation gaps are recalculated. Each round of adjustment focuses on unmet semantic constraints of layout elements, avoiding reliance solely on geometrically valid conditions such as proximity, non-overlap, and no boundary violations as the basis for layout output. When the subject dependency relationship remains stable, the target competing subject's layout elements form sufficient visual isolation, and the constraints in the layout element semantic constraint set meet the preset output conditions, the layout corresponding to the current candidate layout status data is determined as the target layout.

[0042] The counterfactual binding vulnerability is compared with the preset vulnerability condition, and the competitive entity isolation gap is compared with the preset isolation condition. When the counterfactual binding vulnerability reaches the preset vulnerability condition or the competitive entity isolation gap reaches the preset isolation condition, the cross-temporal abnormal edit density, counterfactual binding vulnerability, competitive entity isolation gap, and competitive attribution channel set corresponding to the target relationship local subgraph are read.

[0043] Adjustable layout attributes participating in isolation control are determined based on the set of competition attribution channels. The normalized spacing evidence channel corresponds to the horizontal position, vertical position, and element spacing of subordinate layout elements; the alignment consistency evidence channel corresponds to the alignment axis position of subordinate layout elements relative to the baseline main layout elements and the target competing main layout elements; the container sharing evidence channel corresponds to the position and margin of subordinate layout elements within their respective containers; the reading path continuity evidence channel corresponds to the reading order position of subordinate layout elements; and the group boundary transparency evidence channel corresponds to the position of subordinate layout elements relative to the group boundary.

[0044] Adjustable layout attributes in candidate layout status data constitute an attribute vector. ,in, This represents the number of adjustable layout attributes with an adjustable range length greater than zero. For the first One adjustable layout attribute. Layout attributes with an adjustable range length of zero are not included in the attribute vector and isolation control vector. The perturbation amount of each adjustable layout attribute is denoted as The corresponding unit direction vector is denoted as ,and: Comprehensive risk function Determined according to the following formula: in, and Attribute vectors The corresponding cross-temporal anomalous editing density, counterfactual binding vulnerability, and competitive subject isolation gap.

[0045] No. Sensitivity to risk changes in adjustable layout attributes Determined according to the following formula: When the adjustable range boundary or semantic constraints of layout elements do not allow bidirectional perturbation, the risk change sensitivity is determined by the ratio of the change in the comprehensive risk function before and after unilateral perturbation in the allowed direction to the amount of perturbation. Candidate control directions should preferentially be those corresponding to the decreasing direction of the comprehensive risk function.

[0046] No. The adjustment component of each adjustable layout attribute is denoted as The control components are obtained by discretization within the corresponding adjustable range according to the counterfactual perturbation step size, and each control component constitutes an isolated control vector. Constraint checks are performed on the candidate page state data corresponding to the isolation control vectors. A feasible vector set is formed by identifying isolation control vectors that satisfy the following conditions: semantic constraints on page elements, cross-temporal abnormal editing density lower than a preset abnormality condition, counterfactual binding vulnerability lower than a preset vulnerability condition, and competitive entity isolation gap lower than a preset isolation condition. .

[0047] No. The adjustable range length of each adjustable layout attribute is denoted as . ,and: Weighted Euclidean length of the isolation control vector Determined according to the following formula: Minimum Isolation Control Vector Determined according to the following formula: When multiple isolation control vectors have the same weighted Euclidean length, the isolation control vectors with lower counterfactual binding vulnerability, lower competitive subject isolation gap, and lower cross-temporal abnormal edit density are selected in sequence.

[0048] When the feasible vector set is not empty, the corresponding adjustable page attributes in the candidate page status data are updated based on the minimum isolation control vector. A new candidate page relationship graph is generated based on the updated candidate page status data, and the abnormal relationship editing set, abnormal relationship editing energy, cross-temporal abnormal editing density, target relationship local subgraph, counterfactual binding failure domain, counterfactual binding vulnerability, competitive subject isolation gap, and competitive attribution channel set are redefined. If the preset output conditions are not met in the current round, a new minimum isolation control vector is determined based on the updated values, and the next round of dynamic control is implemented.

[0049] When the feasible vector set is empty, the adjustable layout attributes are expanded to other layout elements associated with the abnormal relationship editing set within the local subgraph of the target relationship, and a new feasible vector set is formed. If the expanded feasible vector set is still empty, or the number of iterations reaches the preset maximum number of iterations, the iteration generation of the current candidate layout is terminated, and layouts that violate the subject dependency constraint or geometric legality constraint are not output.

[0050] The preset output conditions include: the subject dependency relationship remains pointing to the baseline subject page element, the cross-temporal abnormal editing density is lower than the preset abnormal condition, the counterfactual binding vulnerability is lower than the preset vulnerability condition, the competitive subject isolation gap is lower than the preset isolation condition, and all constraints in the page element semantic constraint set are satisfied. For example, the counterfactual binding vulnerability corresponding to the percentage unit element is 0.985, the competitive subject isolation gap is 0.11, the preset vulnerability condition is 0.7, the preset isolation condition is 0.05, and the competitive attribution channel set includes the alignment consistency evidence channel and the normalized spacing evidence channel. After applying two page coordinate units of bidirectional perturbation to the horizontal and vertical positions of the percentage unit element, an isolation control vector is determined that moves eight page coordinate units towards data chart A and offsets two page coordinate units vertically. This isolation control vector strengthens the distance attribution relationship between the percentage unit element and data chart A, weakens the alignment relationship between the percentage unit element and data chart B, and its corresponding weighted Euclidean length is less than other isolation control vectors that meet the constraints, so it is determined as the minimum isolation control vector. After this round of dynamic control, the counterfactual binding vulnerability decreases to 0.43, the competitive subject isolation gap decreases to zero, the cross-temporal abnormal editing density decreases to 0.18, the semantic constraints of each page element meet the preset output conditions, and the page corresponding to the updated candidate page state data is determined as the target page.

[0051] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligently generating semantic constraints for page layout elements, characterized in that, include: S1. Obtain the historical page state sequence and the current page generation input. Construct a historical page relationship graph sequence based on the historical page state sequence. Determine the baseline semantic binding relationship, authorization relationship change set, and page element semantic constraint set from the historical page relationship graph sequence. S2. Generate candidate page status data based on the current page output, and determine the abnormal relationship editing set, abnormal relationship editing energy, cross-temporal abnormal editing density, and target relationship local subgraph based on the candidate page status data, historical page relationship graph sequence, and authorized relationship change set; S3. Based on the target relationship local subgraph and candidate page status data, determine the subordinate page elements, the baseline main page elements, the candidate competing main page elements, the visual attribution evidence vector, and the counterfactual binding failure domain; S4. Determine the counterfactual binding vulnerability based on the counterfactual binding failure domain, and determine the subject's visual attribution confidence, target competing subject's layout elements, competing subject's isolation gap, and competing attribution channel set based on the visual attribution evidence vector; S5. Based on the cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive subject isolation gap, and competitive attribution channel set, isolate, regulate, and iteratively generate the candidate page state data until the candidate page state data meets the constraints in the page element semantic constraint set, and output the target page.

2. The method for intelligent generation of semantic constraints for page elements according to claim 1, characterized in that, The baseline semantic binding relationship includes the subject dependency relationship between subordinate page elements and the baseline main page element. The set of semantic constraints for page elements includes subject dependency constraints, content ownership constraints, information hierarchy constraints, reading order constraints, container ownership constraints, group boundary constraints, visual ownership constraints, and geometric legal constraints.

3. The method for intelligent generation of semantic constraints for page elements according to claim 2, characterized in that, Step S2 includes: generating candidate page state data based on the current page generation input, and constructing a candidate page relationship graph based on the candidate page state data; comparing the candidate page relationship graph with the historical page relationship graph sequence for relationship editing, and obtaining an abnormal relationship editing set using the authorized relationship change set; determining the abnormal relationship editing energy and cross-temporal abnormal editing density based on the abnormal relationship editing set, and determining the target relationship local subgraph based on the cross-temporal abnormal editing density.

4. The method for intelligent generation of semantic constraints for page elements according to claim 3, characterized in that, The candidate page layout relationship graph is constructed based on the page element set, element semantic role, element content ownership identifier, element geometric attributes, reading order, container boundary, and grouping boundary in the candidate page layout status data; the cross-temporal abnormal editing density is determined by the relationship abnormal editing energy, and the target relationship local subgraph is a local graph that has reached the preset abnormal condition of cross-temporal abnormal editing density and contains subordinate page elements.

5. The method for intelligent generation of semantic constraints for page layout elements according to claim 1, characterized in that, In step S3, candidate competing subject page elements are determined by subject page elements that have visual affiliation competition with subordinate page elements in the candidate page relationship graph; an ordered counterfactual perturbation trajectory and a counterfactual perturbation page set are generated based on the visual affiliation evidence vector; In the counterfactual perturbation layout set, perturbation layouts that satisfy geometric legal constraints are selected; when the subject dependency relationship in the perturbation layout is flipped from the baseline subject layout element to the candidate competing subject layout element, the corresponding perturbation layout is recorded as a counterfactual binding failure sample. Counterfact binding failure domains are formed from counterfact binding failure samples.

6. The method for intelligent generation of semantic constraints for page elements according to claim 5, characterized in that, The visual attribution evidence vector includes at least two of the following evidence channels: normalized spacing, alignment consistency, container sharing, reading path continuity, and group boundary transparency.

7. The method for intelligent generation of semantic constraints for page elements according to claim 6, characterized in that, Counterfactual binding vulnerability is determined as follows: based on the counterfactual binding failure domain, the failure domain measure, minimum failure radius, and flip-over duration coefficient are calculated. By nonlinearly coupling the failure domain measure, minimum failure radius, flip duration coefficient, and transtemporal anomaly editing density, counterfactual binding vulnerability is obtained.

8. The method for intelligent generation of semantic constraints for page elements according to claim 1, characterized in that, Based on the visual attribution evidence vector, the subject visual attribution confidence of subordinate page elements relative to the benchmark subject page elements and candidate competing subject page elements are calculated respectively, and the subject attribution confidence difference is determined based on the subject visual attribution confidence. Candidate competing subject page elements whose subject attribution confidence difference reaches the preset competition conditions are identified as target competing subject page elements, and the competing subject isolation gap is determined based on the subject visual attribution confidence of the target competing subject page elements, the subject visual attribution confidence of the benchmark subject page elements, and the preset safety interval.

9. The method for intelligent generation of semantic constraints for page elements according to claim 8, characterized in that, The set of competing attribution channels is formed as follows: each evidence channel in the visual attribution evidence vector is subjected to single-channel suppression, the contribution value of each evidence channel to the isolation gap of the competing subjects is calculated, and the evidence channels are selected from large to small according to the contribution value to form the set of competing attribution channels.

10. A method for intelligently generating semantic constraints for page elements according to any one of claims 1-9, characterized in that, Step S5 includes: when the counterfactual binding vulnerability reaches the preset vulnerability condition or the competitive subject isolation gap reaches the preset isolation condition, determining the minimum isolation control vector based on the cross-temporal abnormal editing density, counterfactual binding vulnerability, competitive subject isolation gap, and competitive attribution channel set; dynamically controlling and iteratively generating the candidate page state data based on the minimum isolation control vector until the constraints in the page element semantic constraint set meet the preset output conditions, and outputting the target page.

Images