A method and system for pagination of film and television scripts based on natural language processing

Abstract

This invention discloses a method and system for pagination of film and television scripts based on natural language processing, belonging to the field of data processing technology. The invention extracts script elements through a multimodal semantic parsing module, combines a spatiotemporal continuity weighting algorithm with a dynamic compensation mechanism to generate a structured dataset; constructs a pagination decision matrix using four-dimensional decision factors, achieves intelligent conflict handling through a three-level arbitration mechanism, and outputs paginated scripts with timestamps and anomaly detection tags; a change impact analysis engine based on a knowledge graph is linked with an improved MLP decision model, driving real-time recalculation through confidence thresholds; finally, a local optimization scheme is generated by combining an impact assessment matrix and three-dimensional difference analysis, and multi-terminal pagination packages are dynamically rendered according to scene requirements; this reduces the fragmentation rate of key scenes, shortens the production cycle, and reduces version incident rates, while improving the efficiency of script modification and rearrangement, the matching degree of film and television shooting shots and dialogue, and the space for stage movement.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and specifically relates to a method and system for pagination and scheduling of film and television scripts based on natural language processing. Background Technology

[0002] Script pagination is a core aspect of film and television production, but traditional methods heavily rely on manual experience. Fixed pagination in paper scripts often disrupts continuous scenes (such as interruptions in a surgical emergency scene across pages) or high-density dialogue, leading to emotional discontinuities in actors' performances and errors in camera choreography. Script revisions typically trigger the rearrangement of related pages, and manual verification is time-consuming and takes up a significant portion of the production cycle (e.g., deleting a supporting character requires manually checking all related scenes). Theater performances require enlarged fonts to allow for movement, and film and television shooting requires binding to storyboards; existing tools cannot dynamically generate multi-scene adaptation solutions. Existing solutions are limited to basic text editing (e.g., Final Draft only supports rule-based pagination) or general typesetting systems (Adobe InDesign lacks semantic analysis), failing to meet the film and television industry's demand for intelligent dynamic pagination and real-time change response. There is an urgent need for intelligent technological breakthroughs that integrate semantic parsing, dynamic decision-making, and multi-terminal rendering.

[0003] Existing technologies suffer from problems such as low efficiency in multi-version collaboration, lack of cross-departmental requirement adaptation, and fragmented narrative coherence due to static pagination. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the problems in related technologies, this invention provides a method for pagination and scheduling of film and television scripts based on natural language processing, thereby overcoming the aforementioned technical problems existing in the prior art.

[0006] (II) Technical Solution

[0007] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0008] S1. Perform multimodal semantic parsing on film and television scripts to obtain a structured script dataset;

[0009] S2. Using a pagination decision matrix and a three-level arbitration mechanism for pagination conflicts, the structured script dataset is dynamically paginated to obtain a paginated script dataset.

[0010] S3. Construct a script knowledge graph based on the paginated script dataset; combine user modification instruction data with the script knowledge graph to obtain a change impact report through change impact analysis.

[0011] S4. Improve the MLP model to obtain an improved MLP model; train and optimize the improved MLP model by combining historical script modification records with decision instruction labels with optimization algorithms to obtain a script decision model.

[0012] Based on the change impact report and the script decision model, real-time decision instructions with confidence levels are obtained; and a unified management dataset is generated based on the real-time decision instructions with confidence levels.

[0013] S5. Based on the unified management dataset, an optimized pagination scheme package is obtained by constructing a version change impact assessment matrix, designing a differential version management algorithm, and optimizing local re-pagination.

[0014] S6. Based on performance requirements, generate a requirement-parameter mapping table; based on the optimized pagination scheme package and the requirement-parameter mapping table, obtain the final pagination package;

[0015] This invention utilizes an intelligent script processing system to transform the original script into a structured dataset through multimodal semantic parsing. It then generates paginated scripts that preserve narrative coherence based on a dynamic pagination arbitration mechanism. A knowledge graph is constructed to analyze the impact of changes, and an improved MLP decision model is used to output confidence-driven overall data. Finally, a version optimization engine and a requirement adaptation module generate multi-scene-specific pagination packages. A spatiotemporal continuity algorithm reduces the fragmentation of key scenes, and the knowledge graph-linked decision engine improves the efficiency of script modification and rearrangement. Furthermore, a dynamic rendering mechanism meets the diverse needs of theaters, film and television production, and dubbing, shortening the production cycle and eliminating version chaos incidents.

[0016] Preferably, step S1 includes the following steps:

[0017] S11. Collect and analyze the key element types in the industry standards of historical film and television scripts to obtain an element type set; the element type set includes five categories: scene, character, dialogue, action description, and prop;

[0018] S12. Based on the element type set, the key elements in the original script file are identified through a regular expression engine to generate an original element set with position markers; the regular expression template includes a scene recognition template and a character dialogue recognition template.

[0019] S13. Construct a script-specific semantic analysis model; the script-specific semantic analysis model consists of a model architecture, an industry terminology library, and a weight calculation algorithm; the weight calculation algorithm includes a scene spatiotemporal continuity weight algorithm and a scene spatiotemporal continuity weight algorithm.

[0020] Based on the original set of elements with position tags, the scene spatiotemporal continuity weights and scene spatiotemporal continuity weights of the elements are calculated through a script-specific semantic analysis model to obtain the element weight set.

[0021] S14. Use a dynamic compensation algorithm that includes continuous scene anti-segmentation compensation and climax segment density compensation to correct the element weights and obtain the corrected element weight set.

[0022] S15. Generate a structured script dataset based on the corrected set of element weights;

[0023] This invention precisely defines five categories of script elements through industry standard analysis, uses a regular expression engine to quickly locate and mark scenes and character dialogue, combines a script-specific semantic model to calculate spatiotemporal continuity weights and dialogue semantic density weights, and then strengthens the protection of key narrative paragraphs through a dynamic compensation algorithm, ultimately generating a structured dataset with corrected weights. This significantly improves parsing efficiency and accuracy, reduces the error rate of industry terminology library element recognition, and the spatiotemporal continuity algorithm and climax compensation mechanism work together to reduce the risk of key scene fragmentation, laying a high-precision data foundation for subsequent dynamic pagination.

[0024] Preferably, step S2 includes the following steps:

[0025] S21. Based on the scene weights and dialogue weights in the structured script dataset, obtain the spatiotemporal continuity and semantic density;

[0026] Calculate the number of unique roles appearing on the current page to obtain the role concentration; monitor the deviation value of the number of rows on the page to obtain the page balance.

[0027] The spatiotemporal continuity, semantic density, role concentration, and page balance together constitute the four-dimensional decision factors.

[0028] Set the weight coefficients of the four-dimensional decision factors to obtain the weights of the four-dimensional factors; set the threshold rules for the weights of the four-dimensional factors to obtain the decision rule base.

[0029] A pagination decision matrix is ​​constructed using the four-dimensional decision, the weights of the four-dimensional factors, and the decision rule base.

[0030] S22. Design a three-level arbitration mechanism for pagination conflicts; based on the three-level arbitration mechanism for pagination conflicts and the pagination decision matrix, the pagination decision matrix is ​​used as a row number constraint, and the three-level arbitration mechanism for pagination conflicts is used as the basis for selecting pagination points. Three-level arbitration is performed on the structured script dataset to obtain an arbitration result set containing pagination points and conflict reports.

[0031] S23. Based on the pagination point selection criteria in the arbitration result set, calculate the page capacity of the structured script dataset, and obtain the pagination intermediate dataset through multi-role balancing processing and page balance verification.

[0032] The pagination intermediate dataset is processed using a timestamp engine and anomaly detection and labeling to obtain the pagination script dataset;

[0033] This invention constructs a pagination decision matrix using four-dimensional decision factors (spatiotemporal continuity, semantic density, role concentration, and page balance), and dynamically selects pagination points using a three-level arbitration mechanism (line compression / key scene conflict handling / cross-page marking). A multi-role balance verification and timestamp engine then generate a paginated script dataset with anomaly detection markers. This mechanism overcomes the limitations of traditional static pagination; the decision matrix weight threshold reduces the risk of key scene segmentation; the three-level arbitration intelligently compresses low-density paragraphs, reducing page line deviation; and anomaly detection eliminates shooting continuity accidents, providing reliable pagination data for multi-terminal output.

[0034] Preferably, step S3 includes the following steps:

[0035] S31. Based on the paginated script dataset, define the node type set and edge relationship construction rules of the knowledge graph; the node type set includes scenes, characters, props, and page numbers; based on the node type set and edge relationship construction rules of the knowledge graph, obtain the script knowledge graph through a graph generation algorithm;

[0036] S32. Collect user modification instruction data; based on the user modification instruction data and script knowledge graph, obtain the set of related elements through change impact analysis and related element location; obtain the related page numbers based on the set of related elements, perform pagination recalculation, and obtain the change impact report;

[0037] This invention achieves element-level connections by constructing a script knowledge graph. When a user inputs a modification command, the system accurately and quickly locates the related elements and intelligently re-paginates, improving the accuracy of change impact reports and reducing the time spent on re-sorting caused by script modifications.

[0038] Preferably, step S4 includes the following steps:

[0039] S41. Collect historical change impact reports to obtain historical script modification records; collect the manually marked modification impact levels corresponding to the historical change impact reports to obtain historical script modification records with decision labels;

[0040] Extract the modification type, scope of impact, and emotional changes data from historical script modification records with decision labels to obtain the characteristics of historical script modification records;

[0041] S42. Construct an MLP model; improve the MLP model to obtain an improved MLP model;

[0042] S43. Use the historical script modification record features to train the improved MLP model. During the training process, use an optimization algorithm to find the optimal weights of the improved MLP model and obtain the optimal solution. Use the optimal solution as the weights of the improved MLP model to obtain the script decision model.

[0043] S44. Extract the modification type, scope of impact, and emotional change data from the change impact report to obtain real-time script modification characteristics; input the real-time script modification characteristics into the script decision model to obtain decision instructions with confidence.

[0044] S45. Set a confidence threshold; when the confidence level is greater than the confidence threshold, based on the script knowledge graph, change impact report, and real-time decision instructions, recalculate the affected pages, update the knowledge graph relationships, and generate new version pagination data to obtain the overall management dataset; otherwise, use the original knowledge graph relationships and pagination script dataset as the overall management dataset.

[0045] This invention trains an improved MLP model using labeled historical modification records, captures temporal dependencies through GRU units, prevents network degradation through residual connections, and enhances robustness through k-sparse activation. After real-time changes to the input features, the model outputs decision instructions with confidence scores, intelligently updates the knowledge graph based on thresholds, and generates a comprehensive dataset. This engine achieves industrial-grade decision-making efficiency, and the confidence mechanism enables unattended processing of high-impact modifications, improving the response speed of recalculation. It also reduces the manual cost of script version management and the accident rate during filming.

[0046] Preferably, step S42 includes the following steps:

[0047] S421. Lightweight GRU units are inserted between the hidden layers of the MLP model to capture the temporal dependency characteristics of script modifications. When character deletion and scene merging occur consecutively, the GRU gating automatically increases the influence coefficient weight.

[0048] Construct input layer, residual block, and output layer pathways. For low-frequency, high-risk "addition or deletion of key features" events, residual connections ensure that key features reach the output layer directly.

[0049] S422. The hidden layer uses the k-sparse activation function to force the number of neurons activated in each layer, and prioritizes the retention of neurons sensitive to film and television features. An attention module is added before the output layer to dynamically allocate feature weights.

[0050] This invention captures temporal dependencies by inserting GRU units, ensures key features reach the output layer through residual connections, and dynamically allocates weights by combining k-sparse activation and attention mechanisms, thereby reducing prediction bias for low-frequency, high-risk events and improving model robustness.

[0051] Preferably, the optimal weights of the improved MLP model are found using an optimization algorithm during the training process in S43 to obtain the optimal solution, including the following steps:

[0052] S431. Construct a particle set; use each particle set in the particle set as the weight of each different improved MLP model; set the maximum number of iterations;

[0053] S432. Set the training accuracy and resource consumption of the improved MLP model, and set the fitness function according to the training accuracy and resource consumption.

[0054] S433. Start the iteration. In each iteration, the fitness function selects, crosses, and mutates particles in the particle set to obtain the particle set after the operation.

[0055] S434. Repeat S433 until the maximum number of iterations is reached, then stop iterating and obtain the optimal solution.

[0056] This invention employs a particle swarm optimization algorithm, which iteratively selects particles (weight combinations) through a fitness function (balancing training accuracy and resource consumption), and finds the optimal solution through selection-crossover-mutation operations. This algorithm improves model prediction accuracy, reduces inference latency, and decreases resource consumption, ensuring high-concurrency response requirements in various scenarios.

[0057] Preferably, step S5 includes the following steps:

[0058] S51. Based on the historical modification records in the centralized management dataset, perform modification type classification statistics to obtain modification type data; the modification type data includes adding or deleting characters, adjusting scenes, and modifying lines; based on the modification type data, calculate the influence coefficient through statistical analysis;

[0059] Based on the modification type data and the impact coefficient, the proportion mapping is recalculated to obtain the version change impact assessment matrix;

[0060] S52. Based on the data before and after script modification in the centralized management dataset, the comparison results data are obtained through text layer comparison, semantic layer analysis and graph layer tracing.

[0061] S53. Based on the impact assessment matrix and comparison results data, local optimized pagination data is obtained by modifying the regional positioning, impact radius calculation and re-pagination interval delineation.

[0062] Based on locally optimized pagination data and unmodified original regional data, an optimized pagination scheme package is obtained through a marking strategy and scheme synthesis operation;

[0063] This invention quantifies the impact of modifications through an impact assessment matrix, and accurately locates the changed areas by combining three-dimensional differential analysis of the text layer, semantic layer, and graph layer. It only performs local optimization on the re-pagination intervals and finally synthesizes a complete solution package. This engine compresses the traditional global re-sorting range, improves recalculation efficiency, and reduces the shooting connection error rate, achieving efficient version iteration at the film and television industry level.

[0064] Preferably, step S6 includes the following steps:

[0065] S61. Based on the performance scene configuration file and historical performance data provided by the production company, obtain the performance scene configuration data; extract the features and requirements parameters of theater mode, film and television shooting and dubbing scene from the performance scene configuration data to obtain requirement classification and standardization data.

[0066] Based on demand classification and standardized data, priority assessment and demand set generation are performed to obtain a structured demand set;

[0067] S62. Collect terminal device configuration files and extract features to obtain terminal features; based on terminal device features and a unified management dataset, obtain a device adaptation rule set through a dynamic rule generation algorithm and detailed explanation of rule parameters;

[0068] Define core mapping rules; based on the structured requirement set and the device adaptation rule set, perform dynamic parameter mapping through the core mapping rules to obtain the requirement-parameter mapping table;

[0069] S63. Based on the knowledge graph and paginated script dataset in the overall management dataset, perform character movement modeling and multi-dimensional analysis to obtain a character analysis report;

[0070] Based on the requirement-parameter mapping table, the optimized pagination solution package, and the role analysis report, the requirements are injected to obtain the final pagination package.

[0071] This invention generates a dynamic parameter mapping table by using a structured requirement set and a device adaptation rule set, and injects an optimized pagination scheme package into the character movement analysis report to output a final pagination package for multiple scenarios. This scheme achieves closed-loop output across the entire chain, improves the matching degree of storyboard marking, shot dialogue, and theatrical version movement space in film and television shooting, and reduces the dubbing re-recording rate.

[0072] A film and television script pagination system based on natural language processing is used to implement the above-mentioned film and television script pagination method based on natural language processing. It includes a script structure parsing module, a dynamic pagination arbitration module, a knowledge graph construction and change analysis module, an intelligent decision engine module, a version optimization engine module, and a pagination package delivery adaptation module.

[0073] The script structure parsing module performs multimodal semantic parsing on the input original film and television scripts. By analyzing the element type set defined by industry standards, it uses a regular expression engine to identify key elements and generate an original element set with position tags. Based on the constructed script-specific semantic analysis model, it calculates the spatiotemporal continuity and semantic density weights. After being corrected by a dynamic compensation algorithm, it finally outputs a structured script dataset.

[0074] The dynamic pagination arbitration module is based on a structured script dataset and dynamically selects the pagination point through a pagination decision matrix and a three-level arbitration mechanism; it also generates a paginated script dataset by combining a timestamp engine and anomaly detection.

[0075] The knowledge graph construction and change analysis module constructs a script knowledge graph from the paginated script dataset, defines nodes and relationship rules, performs change impact analysis by associating user modification instructions with the graph, locates affected elements and recalculates associated page numbers, and outputs a change impact report containing pagination change decisions.

[0076] The intelligent decision engine module captures temporal dependencies by inserting GRU units, adds residual connections to prevent network degradation, and uses the k-sparse activation function to improve robustness, resulting in an improved MLP model. It trains the model using labeled historical modification records, generates a script decision model through optimization algorithms, and outputs decision instructions with confidence scores by combining real-time change features, driving knowledge graph updates and pagination recalculation to generate a comprehensive management dataset.

[0077] The version optimization engine module constructs a version change impact assessment matrix based on the overall management dataset, locates the modified area through differential analysis of the text layer, semantic layer, and graph layer; combines the impact radius calculation to delineate the re-pagination interval, synthesizes the locally optimized data and the unmodified original data, and finally packages and outputs the optimized pagination scheme package.

[0078] The pagination package delivery and adaptation module generates a requirement-parameter mapping table by parsing performance requirements and resolves parameter conflicts through weighted decision-making. Combining the optimized pagination scheme package and the role analysis report, it generates the final pagination package, which includes scene-specific outputs such as theater flow diagrams, film and television storyboard marking scripts, and dubbing continuous dialogue scripts.

[0079] (III) Beneficial Effects

[0080] The present invention has the following beneficial effects:

[0081] This invention upgrades the film and television industry from a workshop-style operation that relies on human experience to an intelligent production line centered on semantic parsing, dynamic decision-making, and multimodal output by ensuring narrative integrity, achieving industrial-grade collaborative efficiency, and enabling intelligent adaptation across all scenarios.

[0082] This invention effectively solves the narrative fragmentation problem caused by script pagination through multimodal semantic parsing and dynamic pagination arbitration mechanism; based on scene spatiotemporal continuity weight algorithm and climax paragraph density compensation, the system automatically identifies and protects key narrative paragraphs, reduces cross-page interruption rate, improves the continuity of actors' emotional performance, and ensures that the intensity of dramatic conflict carried by each page meets the director's expectations.

[0083] This invention significantly improves the efficiency of multi-version collaboration by combining knowledge graphs and intelligent decision engines. When script modifications occur, the change impact analysis module quickly and accurately locates related elements, and the script decision model triggers local re-pagination with high confidence, solving the problem of time-consuming manual verification. Film and television production teams can synchronize the latest pagination data in real time, reducing shooting accidents caused by version chaos.

[0084] The final pagination package generated by this invention is deeply adapted to the needs of multiple scenarios in the film and television industry; for theater performances, the font line spacing is dynamically enlarged and character movement markers are embedded, increasing the visible distance of the lines when stage actors move; storyboard markers are bound to film and television shooting, improving the matching degree between camera movement and the rhythm of the lines; the dubbing-specific version eliminates the problem of emotional discontinuity caused by pauses across pages by forcing continuous pagination of lines, thus reducing the recording re-recording rate.

[0085] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0086] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0087] Figure 1 This is a flowchart illustrating a film and television script pagination method based on natural language processing according to the present invention.

[0088] Figure 2 This is a schematic diagram of a film and television script coordination and pagination system based on natural language processing according to the present invention. Detailed Implementation

[0089] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0090] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "top", "middle", "inner", etc., which indicate orientation or positional relationship, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the invention.

[0091] Example 1:

[0092] Please see Figure 1 This invention discloses a method for pagination and scheduling of film and television scripts based on natural language processing, comprising the following steps:

[0093] S1. Perform multimodal semantic parsing on film and television scripts to obtain a structured script dataset;

[0094] S1 includes the following steps:

[0095] S11. Collect and analyze key element types in industry standards of historical film and television scripts (such as Hollywood standards and Chinese broadcasting formats) to obtain an element type set; the element type set includes five categories: scene, character, dialogue, action description, and prop; the scene includes three elements: "interior / exterior" identifier, location, and time; the character includes all-caps names and dialogue status; the prop includes physical objects that can move across scenes.

[0096] S12. Based on the element type set, the key elements in the original script file are identified through a regular expression engine, generating an original element set with position markers; the regular expression template includes a scene recognition template and a character dialogue recognition template; the scene recognition template, for example, transforms "Interior, hospital operating room, night" into scene type="Interior" loc="Hospital operating room" time="Night"; the character dialogue recognition template, for example, transforms "JOHN (angrily) must leave immediately" into <character id="JOHN" state="angrily">dialogue content<character>;

[0097] S13. Construct a script-specific semantic analysis model (ScriptBERT); the script-specific semantic analysis model consists of a model architecture, an industry terminology database, and a weight calculation algorithm; the model architecture includes an input layer, an embedding layer, a 12-layer Transformer, and an output layer; the input layer is used for script text segmentation sequences, and the embedding layer is used for enhanced embedding of industry terms; the weight calculation algorithm includes a scene spatiotemporal continuity weight algorithm and a scene spatiotemporal continuity weight algorithm.

[0098] Based on the original set of elements with location tags, the scene spatiotemporal continuity weight and scene spatiotemporal continuity weight of the elements are calculated by the script-specific semantic analysis model to obtain the element weight set; the scene spatiotemporal continuity weight is calculated based on the number of shared characters, the total number of characters, the number of identical props and the total number of props; the dialogue semantic density weight includes the calculation of emotional intensity and action complexity.

[0099] S14. Use a dynamic compensation algorithm that includes continuous scene anti-segmentation compensation and climax segment density compensation to correct the element weights and obtain the corrected element weight set; when the element weight > 0.7, use continuous scene anti-segmentation compensation, and when the element weight > 0.8, use climax segment density compensation.

[0100] S15. Generate a structured script dataset based on the corrected element weight set and store it in XML format to obtain the structured script dataset; the generation of the structured script dataset based on the corrected element weight set includes the following steps: create a root node, insert scene nodes into the root node according to the script sequence, insert characters and props into the scene nodes, insert lines under the characters, inject the corrected weight values ​​from the corrected element weight set, and add element position markers (row number + column number) to obtain the structured script dataset.

[0101] S2. Using a pagination decision matrix and a three-level arbitration mechanism for pagination conflicts, the structured script dataset is dynamically paginated to obtain a paginated script dataset.

[0102] S2 includes the following steps:

[0103] S21. Based on the scene weights and dialogue weights in the structured script dataset, obtain the spatiotemporal continuity and semantic density;

[0104] Calculate the number of unique roles appearing on the current page to obtain the role concentration; monitor the deviation value of the number of rows on the page to obtain the page balance.

[0105] The spatiotemporal continuity, semantic density, role concentration, and page balance together constitute the four-dimensional decision factors.

[0106] Set the weight coefficients of the four-dimensional decision factors to obtain the weights of the four-dimensional factors; set the threshold rules for the weights of the four-dimensional factors to obtain the decision rule base.

[0107] A pagination decision matrix is ​​constructed using the four-dimensional decision, the weights of the four-dimensional factors, and the decision rule base; the pagination decision matrix is ​​as follows.

[0108] | Decision Factors | Weighting Coefficients | Threshold Rules |

[0109] | Spatiotemporal continuity | 0.4 | Pagination is disabled when the weighting coefficient is >0.7 |

[0110] | Semantic density | 0.3 | Every 0.1 change affects 5 rows |

[0111] | Role Concentration | 0.2 | Triggers the balancing algorithm when ≥4 |

[0112] Page balance | 0.1 | Re-paginate when line count deviation > 8 lines |

[0113] S22. Design a three-level arbitration mechanism for pagination conflicts; the three-level arbitration mechanism for pagination conflicts includes primary arbitration, secondary arbitration, and final arbitration; the primary arbitration is to prioritize the compression of low-density paragraphs with a weight coefficient <0.4 when the number of rows exceeds the limit by deleting redundant action descriptions (such as "he slowly" → "he") and merging adjacent short lines; the secondary arbitration is to calculate the pagination point loss value when there is a conflict in a key scene (weight coefficient >0.7); the final arbitration is to insert a cross-page mark and generate a manual verification report;

[0114] Based on the three-level arbitration mechanism for pagination conflicts and the pagination decision matrix, the pagination decision matrix is ​​used as a row number constraint, and the three-level arbitration mechanism for pagination conflicts is used as the basis for selecting pagination points. The structured script dataset is subjected to three-level arbitration to obtain an arbitration result set containing pagination points and conflict reports.

[0115] S23. Based on the pagination point selection criteria in the arbitration result set, calculate the page capacity of the structured script dataset, and obtain the pagination intermediate dataset (with page break markers) through multi-role balancing processing and page balance verification.

[0116] The pagination intermediate dataset is processed using a timestamp engine and anomaly detection marking to obtain the pagination script dataset; the pagination script dataset is as follows.

[0117] Page number: 34 [00:52:15-00:53:08]

[0118] ├─ Scene: Interior - Operating Room - Night (Spatiotemporal Continuity W_s=0.85)

[0119] ├─ Character: Doctor A [5 lines of dialogue][Emotional density W_d=0.9]

[0120] Line 1 of dialogue: Quick, use hemostats (seriously) (0:52:15)

[0121] Line 2 of dialogue: Blood pressure is dropping (seriously) (0:52:20)

[0122] │...

[0123] └─ Props: Scalpel (P08) [Continued to P35];

[0124] The anomaly detection markers include cross-page characters, prop interruptions, and time errors; the paginated script dataset includes XML for machine-readable complete metadata → S3 (knowledge graph construction); PDF for standard printing format and on-set shooting; and interactive HTML, a webpage with hover tooltips for director preview.

[0125] S3. Construct a script knowledge graph based on the paginated script dataset; combine user modification instruction data with the script knowledge graph to obtain a change impact report through change impact analysis.

[0126] S3 includes the following steps:

[0127] S31. Based on the paginated script dataset, define the node type set and edge relationship construction rules for the knowledge graph; the node type set includes scenes, characters, props, and page numbers; the edge relationship rules are as follows.

[0128]

[0129] Based on the node type set and edge relationship construction rules of the knowledge graph, a script knowledge graph is obtained through a graph generation algorithm;

[0130] S32. Collect user modification instruction data; based on the user modification instruction data and the script knowledge graph, obtain a set of related elements through change impact analysis and related element location; obtain the related page numbers based on the related element set, perform pagination recalculation, and obtain a change impact report (including a list of recalculated page numbers).

[0131] S4. Improve the MLP model to obtain an improved MLP model; train and optimize the improved MLP model by combining historical script modification records with decision instruction labels with optimization algorithms to obtain a script decision model.

[0132] Based on the change impact report and the script decision model, real-time decision instructions with confidence levels are obtained; and a unified management dataset is generated based on the real-time decision instructions with confidence levels.

[0133] S41 includes the following steps:

[0134] S41. Collect historical change impact reports to obtain historical script modification records; collect the manually annotated modification impact levels corresponding to the historical change impact reports to obtain historical script modification records with decision labels; the impact level is between 0.1 and 1;

[0135] Extract the modification type, scope of impact, and emotional changes data from historical script modification records with decision labels to obtain the characteristics of historical script modification records;

[0136] S42. Construct an MLP model; improve the MLP model to obtain an improved MLP model;

[0137] S42 includes the following steps:

[0138] S421. Lightweight GRU units are inserted between the hidden layers of the MLP model to capture the temporal dependency characteristics of script modification. When character deletion and scene merging occur consecutively, the GRU gate automatically increases the weight of the influence coefficient, solving the defect of traditional MLP ignoring the correlation of events.

[0139] By constructing input layer, residual block, and output layer pathways, deep network degradation is avoided. For low-frequency, high-risk "addition or deletion of key features" events, residual connections ensure that key features reach the output layer directly, reducing prediction bias by 35%.

[0140] S422. The hidden layer uses the k-sparse activation function, which forces only 20% of neurons in each layer to be activated, prioritizing the retention of neurons sensitive to film and television features, suppressing irrelevant noise, and improving the robustness of the model; an attention module is added before the output layer to dynamically allocate feature weights.

[0141] S43. Use the historical script modification record features to train the improved MLP model. During the training process, use an optimization algorithm to find the optimal weights of the improved MLP model and obtain the optimal solution. Use the optimal solution as the weights of the improved MLP model to obtain the script decision model.

[0142] The optimal weights for the improved MLP model are found using an optimization algorithm during the training process in S43, and the optimal solution is obtained by the following steps:

[0143] S431. Construct a particle set; use each particle set in the particle set as the weight of each different improved MLP model; set the maximum number of iterations;

[0144] S432. Set the training accuracy of the improved MLP model to be... a And the resource consumption of the improved MLP model is b ,according to a and b The fitness of each particle in the aforementioned particle combination is set as follows: c The calculation formula is as follows:

[0145] ;

[0146] S433. Start the iteration. In each iteration, select, cross over and mutate the particles in the particle set to obtain the particle set after the operation.

[0147] S434. Repeat S433. When the maximum number of iterations is reached, stop iterating and obtain the optimal solution.

[0148] S44. Extract the modification type, scope of impact, and emotional change data from the change impact report to obtain real-time script modification characteristics; input the real-time script modification characteristics into the script decision model to obtain decision instructions with confidence.

[0149] S45. Set a confidence threshold. When the confidence level is greater than the threshold, based on the script knowledge graph, change impact report, and real-time decision instructions, recalculate the affected pages, update the knowledge graph relationships, and generate new version pagination data to obtain a unified management dataset. Otherwise, use the original knowledge graph relationships and pagination script dataset as the unified management dataset. For decision instructions with confidence levels, if the confidence level is greater than 0.9, the S2 engine is automatically called to recalculate the associated pages, avoiding manual intervention and improving efficiency by 40 times. When the confidence level is between 0.7 and 0.9, three optimization schemes are generated for selection to balance automation and manual control. When the confidence level is less than 0.7, only the change is recorded without triggering recalculation.

[0150] S5. Based on the unified management dataset, an optimized pagination scheme package is obtained by constructing a version change impact assessment matrix, designing a differential version management algorithm, and optimizing local re-pagination.

[0151] S51. Based on the historical modification records in the unified management dataset, perform modification type classification statistics to obtain modification type data; the modification type data includes adding or deleting characters, adjusting scenes, and modifying lines; based on the modification type data, calculate the influence coefficient through statistical analysis; for example, adding or deleting characters affects an average of 38 related elements, with a coefficient of 0.9; adjusting scenes affects an average of 12 elements, with a coefficient of 0.7; modifying lines affects an average of 3 elements, with a coefficient of 0.3;

[0152] Based on the modified data type and impact coefficient, a recalculated proportional mapping is performed to obtain the version change impact assessment matrix; as follows.

[0153]

[0154] S52. Based on the data before and after script modification in the centralized management dataset, comparison results are obtained through text-level comparison, semantic-level analysis, and graph-level tracing. The text-level comparison uses the Longest Common Subsequence (LCS) algorithm to locate the difference rows and marks them as: newly added content (green) and deleted content (red). The semantic-level analysis compares the changes in element weights and marks them as significant scene changes and changes in emotional density, respectively. The graph-level tracing queries the knowledge graph to obtain the props associated with the deleted character and the spatiotemporal associations of the modified scene.

[0155] S53. Based on the impact assessment matrix and comparison results data, local optimized pagination data is obtained by modifying the regional positioning, impact radius calculation and re-pagination interval delineation.

[0156] Based on locally optimized pagination data and unmodified original regional data, an optimized pagination scheme package is obtained through a marking strategy and scheme synthesis operation;

[0157] S6. Based on performance requirements, generate a requirement-parameter mapping table; based on the optimized pagination scheme package and the requirement-parameter mapping table, obtain the final pagination package;

[0158] S6 includes the following steps:

[0159] S61. Based on the performance scene configuration file and historical performance data provided by the production company, obtain the performance scene configuration data; extract the features and requirement parameters of theater mode, film and television shooting, and dubbing scene from the performance scene configuration data to obtain requirement classification and standardized data; theater mode features, such as actors needing larger line spacing for easier reading during stage movement, require parameters such as increasing line spacing by 15% and enlarging the font to 14pt; film and television shooting features, such as needing to associate with storyboards to guide camera movement, require parameters such as adding storyboard markers shot ref="S103"; dubbing scene features, such as maintaining dialogue continuity to avoid page breaks, require parameters such as forcing continuous pagination of dialogue for the same character.

[0160] Based on demand classification and standardized data, priority assessment and demand set generation are performed to obtain a structured demand set;

[0161] S62. Collect terminal device configuration files and extract features to obtain terminal features; based on terminal device features and a unified management dataset, obtain a device adaptation rule set through a dynamic rule generation algorithm and detailed explanation of rule parameters;

[0162] Define core mapping rules; based on the structured requirement set and the device adaptation rule set, perform dynamic parameter mapping through the core mapping rules to obtain the requirement-parameter mapping table; if there are conflicts, obtain the requirement-parameter mapping table through weighted decision-making.

[0163] S63. Based on the knowledge graph and paginated script dataset in the overall management dataset, perform character movement modeling and multi-dimensional analysis to obtain a character analysis report;

[0164] Based on the requirement-parameter mapping table, the optimized pagination solution package, and the role analysis report, requirements are injected to obtain the final pagination package; the final pagination package is as follows.

[0165] The final pagination of "A Movie"

[0166] ├─# Theater Mode Only

[0167] │├─ # Take away the position marker script

[0168] │└─# Character Movement Image Gallery

[0169] ├─ # For film and television shooting purposes

[0170] │├─ # Script with storyboard markings

[0171] │└─ # Storyboard - Page Number Reference Table

[0172] └─ # For use in dubbing scenarios

[0173] ├─ # Continuous Dialogue Script

[0174] └─ # Emotional Tips Manual

[0175] Example 2:

[0176] Please see Figure 2 A film and television script pagination system based on natural language processing is used to implement the above-mentioned film and television script pagination method based on natural language processing. It includes a script structure parsing module, a dynamic pagination arbitration module, a knowledge graph construction and change analysis module, an intelligent decision engine module, a version optimization engine module, and a pagination package delivery adaptation module.

[0177] The script structure parsing module performs multimodal semantic parsing on the input original film and television scripts. By analyzing the element type set defined by industry standards, it uses a regular expression engine to identify key elements and generate an original element set with position tags. Based on the constructed script-specific semantic analysis model, it calculates the spatiotemporal continuity and semantic density weights. After being corrected by a dynamic compensation algorithm, it finally outputs a structured script dataset.

[0178] The dynamic pagination arbitration module is based on a structured script dataset and dynamically selects the pagination point through a pagination decision matrix and a three-level arbitration mechanism; it also generates a paginated script dataset by combining a timestamp engine and anomaly detection.

[0179] The knowledge graph construction and change analysis module constructs a script knowledge graph from the paginated script dataset, defines nodes and relationship rules, performs change impact analysis by associating user modification instructions with the graph, locates affected elements and recalculates associated page numbers, and outputs a change impact report containing pagination change decisions.

[0180] The intelligent decision engine module captures temporal dependencies by inserting GRU units, adds residual connections to prevent network degradation, and uses the k-sparse activation function to improve robustness, resulting in an improved MLP model. It trains the model using labeled historical modification records, generates a script decision model through optimization algorithms, and outputs decision instructions with confidence scores by combining real-time change features, driving knowledge graph updates and pagination recalculation to generate a comprehensive management dataset.

[0181] The version optimization engine module constructs a version change impact assessment matrix based on the overall management dataset, locates the modified area through differential analysis of the text layer, semantic layer, and graph layer; combines the impact radius calculation to delineate the re-pagination interval, synthesizes the locally optimized data and the unmodified original data, and finally packages and outputs the optimized pagination scheme package.

[0182] The pagination package delivery and adaptation module generates a requirement-parameter mapping table by parsing performance requirements and resolves parameter conflicts through weighted decision-making. Combining the optimized pagination scheme package and the role analysis report, it generates the final pagination package, which includes scene-specific outputs such as theater flow diagrams, film and television storyboard marking scripts, and dubbing continuous dialogue scripts.

[0183] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0184] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for pagination and organization of film and television scripts based on natural language processing, characterized in that, Includes the following steps: S1. Perform multimodal semantic parsing on film and television scripts to obtain a structured script dataset; S2. Using a pagination decision matrix and a three-level arbitration mechanism for pagination conflicts, the structured script dataset is dynamically paginated to obtain a paginated script dataset. S2 includes the following steps: S21. Based on the scene weights and dialogue weights in the structured script dataset, obtain the spatiotemporal continuity and semantic density; Calculate the number of unique roles appearing on the current page to obtain the role concentration; monitor the deviation value of the number of rows on the page to obtain the page balance. The spatiotemporal continuity, semantic density, role concentration, and page balance together constitute the four-dimensional decision factors. Set the weight coefficients of the four-dimensional decision factors to obtain the weights of the four-dimensional factors; set the threshold rules for the weights of the four-dimensional factors to obtain the decision rule base. A pagination decision matrix is ​​constructed using the four-dimensional decision, the weights of the four-dimensional factors, and the decision rule base. S22. Design a three-level arbitration mechanism for pagination conflicts; based on the three-level arbitration mechanism for pagination conflicts and the pagination decision matrix, perform three-level arbitration on the structured script dataset to obtain an arbitration result set containing pagination points and conflict reports; S23. Based on the pagination point selection criteria in the arbitration result set, calculate the page capacity of the structured script dataset, and obtain the pagination intermediate dataset through multi-role balancing processing and page balance verification. The pagination intermediate dataset is processed using a timestamp engine and anomaly detection and labeling to obtain the pagination script dataset; S3. Construct a script knowledge graph based on the paginated script dataset; combine user modification instruction data with the script knowledge graph to obtain a change impact report through change impact analysis. S4. Improve the MLP model to obtain an improved MLP model; train and optimize the improved MLP model by combining historical script modification records with decision instruction labels with optimization algorithms to obtain a script decision model. Based on the change impact report and the script decision model, real-time decision instructions with confidence levels are obtained; and a unified management dataset is generated based on the real-time decision instructions with confidence levels. The steps in S4 to improve the MLP model and obtain the improved MLP model include: Lightweight GRU units are inserted between the hidden layers of the MLP model to capture the temporal dependency characteristics of script modifications. When character deletion and scene merging occur consecutively, the GRU gate automatically increases the weight of the influence coefficient. Construct input layer, residual block, and output layer paths; The hidden layer uses the k-sparse activation function, and an attention module is added before the output layer to dynamically allocate feature weights; S5. Based on the unified management dataset, an optimized pagination scheme package is obtained by constructing a version change impact assessment matrix, designing a differential version management algorithm, and optimizing local re-pagination. S5 includes the following steps: S51. Based on the historical modification records in the unified management dataset, perform modification type classification statistics to obtain modification type data; Based on the modification type data and the impact coefficient, the proportion mapping is recalculated to obtain the version change impact assessment matrix; S52. Based on the data before and after script modification in the centralized management dataset, the comparison results data are obtained through text layer comparison, semantic layer analysis and graph layer tracing. S53. Based on the impact assessment matrix and comparison results data, local optimized pagination data is obtained by modifying the regional positioning, impact radius calculation and re-pagination interval delineation. Based on locally optimized pagination data and unmodified original regional data, an optimized pagination scheme package is obtained through a marking strategy and scheme synthesis operation; S6. Based on the performance requirements, generate a requirement-parameter mapping table; based on the optimized pagination scheme package and the requirement-parameter mapping table, obtain the final pagination package.

2. The method for pagination and organization of film and television scripts based on natural language processing according to claim 1, characterized in that, S1 includes the following steps: S11. Collect and analyze the key element types in the industry standards of historical film and television scripts to obtain the element type set; S12. Based on the element type set, the key elements in the original script file are identified through the regular expression engine, and a set of original elements with position tags is generated. S13. Construct a script-specific semantic analysis model; Based on the original set of elements with position tags, the scene spatiotemporal continuity weights and scene spatiotemporal continuity weights of the elements are calculated through a script-specific semantic analysis model to obtain the element weight set. S14. Use a dynamic compensation algorithm to correct the element weights and obtain the corrected set of element weights. S15. Generate a structured script dataset based on the corrected set of element weights.

3. The method for pagination and organization of film and television scripts based on natural language processing according to claim 1, characterized in that, S3 includes the following steps: S31. Based on the paginated script dataset, define the node type set and edge relationship construction rules for the knowledge graph; obtain the script knowledge graph through a graph generation algorithm; S32. Collect user modification instruction data; based on the user modification instruction data and the script knowledge graph, obtain the set of related elements through change impact analysis and related element location; obtain the related page numbers based on the set of related elements, perform pagination recalculation, and obtain the change impact report.

4. The method for pagination and organization of film and television scripts based on natural language processing according to claim 1, characterized in that, S4 further includes the following steps: S41. Collect historical change impact reports to obtain historical script modification records; collect the manually marked modification impact levels corresponding to the historical change impact reports to obtain historical script modification records with decision labels; Extract the modification type, scope of impact, and emotional changes data from historical script modification records with decision labels to obtain the characteristics of historical script modification records; S43. Use the historical script modification record features to train the improved MLP model. During the training process, use an optimization algorithm to find the optimal weights of the improved MLP model and obtain the optimal solution. Use the optimal solution as the weights of the improved MLP model to obtain the script decision model. S44. Extract the modification type, scope of impact, and emotional change data from the change impact report to obtain real-time script modification characteristics; input the real-time script modification characteristics into the script decision model to obtain decision instructions with confidence. S45. Set a confidence threshold; when the confidence level is greater than the confidence threshold, based on the script knowledge graph, change impact report, and real-time decision instructions, recalculate the affected pages, update the knowledge graph relationships, and generate new version pagination data to obtain the overall management dataset; otherwise, use the original knowledge graph relationships and pagination script dataset as the overall management dataset.

5. The method for pagination and scheduling of film and television scripts based on natural language processing according to claim 4, characterized in that, The optimal weights for the improved MLP model are found using an optimization algorithm during the training process in S43, and the optimal solution is obtained by the following steps: S431. Construct a particle set; use each particle set in the particle set as the weight of each different improved MLP model; set the maximum number of iterations; S432. Set the training accuracy and resource consumption of the improved MLP model, and set the fitness function according to the training accuracy and resource consumption. S433. Start the iteration. In each iteration, the fitness function selects, crosses, and mutates particles in the particle set to obtain the particle set after the operation. S434. Repeat S433. When the maximum number of iterations is reached, stop iterating and obtain the optimal solution.

6. The method for pagination and organization of film and television scripts based on natural language processing according to claim 1, characterized in that, S6 includes the following steps: S61. Based on the performance scene configuration file and historical performance data provided by the production company, obtain the performance scene configuration data; extract the features and requirements parameters of theater mode, film and television shooting and dubbing scene from the performance scene configuration data to obtain requirement classification and standardization data. Based on demand classification and standardized data, priority assessment and demand set generation are performed to obtain a structured demand set; S62. Collect terminal device configuration files and extract features to obtain terminal features; based on terminal device features and a unified management dataset, obtain a device adaptation rule set through a dynamic rule generation algorithm and detailed explanation of rule parameters; Define core mapping rules; based on the structured requirement set and the device adaptation rule set, perform dynamic parameter mapping through the core mapping rules to obtain the requirement-parameter mapping table; S63. Based on the knowledge graph and paginated script dataset in the overall management dataset, perform character movement modeling and multi-dimensional analysis to obtain a character analysis report; Based on the requirement-parameter mapping table, the optimized pagination solution package, and the role analysis report, the requirements are injected to obtain the final pagination package.

7. A film and television script pagination system based on natural language processing, characterized in that, The system implements a film and television script pagination method based on natural language processing as described in any one of claims 1-6. The system includes a script structure parsing module, a dynamic pagination arbitration module, a knowledge graph construction and change analysis module, an intelligent decision engine module, a version optimization engine module, and a pagination package delivery adaptation module.

Images