A script analysis method, device, equipment and storage medium

Abstract

Embodiments of the present application relate to a script analysis method, device, equipment and storage medium. The method comprises: generating script structure data of a script to be analyzed; performing cross-episode state analysis on a target entity node in the script structure data to obtain visual effect identification data; and performing visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain a visual effect analysis result. The technical solution of the embodiments of the present application can realize flexible and universal, accurate and efficient automatic script analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular relates to a script analysis method and device, equipment and a storage medium. BACKGROUND

[0002] In modern film and television production, visual effect production is an important technical means to improve the quality of works and the experience of watching. Before the start of a film and television project, the production team needs to analyze the visual effect content of the script in detail in order to formulate an accurate budget plan and production plan.

[0003] The traditional script visual effect content analysis process mainly relies on the script team to manually read the script to generate asset analysis tables, visual effect analysis tables and budget tables in the form of tables. This method is inefficient and time-consuming. With the development of artificial intelligence technology, script automatic analysis methods based on machine learning have gradually emerged. However, the existing script automatic analysis scheme is fixed in the way of applying machine learning technology, and it is difficult to provide accurate and efficient processing methods for the special text format of the script file and the complex semantic association between scenes. A large amount of manual post-correction work is still required, which cannot save labor and time costs and does not have practical value for application. SUMMARY

[0004] The present application provides a script analysis method, device, equipment and storage medium, which aims to realize flexible and universal, accurate and efficient automatic script analysis.

[0005] In a first aspect, the present application provides a script analysis method, comprising:

[0006] generating script structure data of a script to be analyzed;

[0007] performing cross-episode state analysis on a target entity node in the script structure data to obtain visual effect recognition data;

[0008] performing visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect recognition data to obtain a visual effect analysis result.

[0009] Optionally, before the script structure data of the script to be analyzed is generated, the method further comprises:

[0010] performing text layout analysis on an original script to obtain text image data;

[0011] extracting target text data from the text image data according to a preset script format;

[0012] performing paragraph optimization processing on the target text data to obtain the script to be analyzed.

[0013] Optionally, the generating the script structure data of the script to be analyzed comprises:

[0014] performing target structural processing on the script to be analyzed to obtain a structured script of each scene; wherein the structured script comprises target entity data and entity relationship data in the scene;

[0015] constructing a script knowledge graph according to the structured script to obtain the script structure data.

[0016] Optionally, the constructing a script knowledge graph according to the structured script to obtain the script structure data comprises:

[0017] creating the target entity node according to the target entity data;

[0018] constructing the target entity node into a target hierarchical graph according to the entity relationship data;

[0019] constructing each target hierarchical graph into the script knowledge graph according to the association relationship between each target entity node and each scene to obtain the script structure data.

[0020] Optionally, the performing cross-scene state analysis on the target entity node in the script structure data to obtain visual effect recognition data comprises:

[0021] obtaining the target entity data of each target entity node in each scene in sequence according to the scene time sequence;

[0022] identifying the target visual effect state of each target entity node according to the target entity data, and generating the visual effect recognition data for each target visual effect state.

[0023] Optionally, the performing visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect recognition data to obtain a visual effect analysis result comprises:

[0024] determining a target visual effect asset according to the script structure data and the visual effect recognition data;

[0025] establishing an association relationship between the target visual effect asset and a paragraph of the script to be analyzed;

[0026] generating the visual effect analysis result according to the association relationship between the target visual effect asset and the paragraph.

[0027] Optionally, the performing visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect recognition data to obtain a visual effect analysis result comprises:

[0028] inputting the script structure data and the visual effect identification data into a script analysis model;

[0029] obtaining the visual effect analysis result output by the script analysis model;

[0030] Before the visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data, the method further includes:

[0031] generating a model training sample according to a visual effect judgment knowledge base;

[0032] training a large language model according to the model training sample to obtain the script analysis model.

[0033] In a second aspect, an embodiment of the present application provides a script analysis device, including:

[0034] a structure generation module configured to generate script structure data of a script to be analyzed;

[0035] a visual effect identification module configured to perform cross-episode state analysis on a target entity node in the script structure data to obtain visual effect identification data;

[0036] an analysis result module configured to perform visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain a visual effect analysis result.

[0037] In a third aspect, an embodiment of the present application provides a script analysis device, including:

[0038] one or more processors;

[0039] a memory configured to store one or more programs;

[0040] When the one or more programs are executed by the one or more processors, the one or more processors implement the script analysis method provided by any embodiment of the present application.

[0041] In a fourth aspect, an embodiment of the present application provides a storage medium containing computer executable instructions, which, when executed by a computer processor, are used to perform the script analysis method provided by any embodiment of the present application.

[0042] The script analysis method, device, equipment and storage medium provided by the embodiment of the present application generate script structure data of a script to be analyzed, perform cross-episode state analysis on target entity nodes in the script structure data, obtain visual effect identification data, and perform visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data, so as to obtain a visual effect analysis result, thereby solving the problem that the prior art lacks document processing capability and scene correlation analysis capability, and realizing flexible, universal, accurate and efficient automatic script analysis. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 A flowchart of a script analysis method provided for the first embodiment of the present application;

[0044] Figure 2 A flowchart of a script analysis method provided for the second embodiment of the present application;

[0045] Figure 3 A flowchart of script standardization processing provided for the second embodiment of the present application;

[0046] Figure 4 A flowchart of generating visual effect identification data provided for the second embodiment of the present application;

[0047] Figure 5 A flowchart of script analysis model training and working provided for the second embodiment of the present application;

[0048] Figure 6 A flowchart of a script analysis method provided for the second embodiment of the present application;

[0049] Figure 7 A structural schematic diagram of a script analysis device provided for the third embodiment of the present application;

[0050] Figure 8 A structural schematic diagram of a script analysis device provided for the third embodiment of the present application; DETAILED DESCRIPTION

[0051] The present application will be further described below in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, but not all the structures.

[0052] Embodiment one

[0053] Figure 1A flowchart of a script analysis method provided for the first embodiment of the present application. The embodiment can be applied to automatically analyze the content and visual effect production requirements of a script. The method can be executed by a script analysis device, which can be implemented by hardware and / or software and generally integrated in an electronic device, such as a computer device. The method specifically includes:

[0054] Step 110: generating script structure data of the script to be analyzed.

[0055] The script to be analyzed can be any script file that needs to analyze its content and visual effect production requirements. The script structure data can be structured data that records the content of the script to be analyzed.

[0056] By identifying and analyzing the semantic structure features of the text content of the script to be analyzed, the script to be analyzed can be structured and sorted to obtain script structure data. Optionally, the script to be analyzed can be structured and sorted according to the scene structure in the content to obtain script structure data in units of scenes.

[0057] The script structure data can include but is not limited to all scene basic information, character information, location information, prop information, vehicle information, animal information, and / or event information contained in the script content, and present each entity and the relationship between them in the above information in a specific data structure. The script structure data can record information in a predetermined structure, which can specifically consist of nodes and associations between nodes. By identifying the entity concepts such as scenes, characters, props, vehicles, and / or animals in the above information in the script to be analyzed, data nodes for representing entities in the script structure data can be generated. By analyzing the associations between entities in the above information in the script to be analyzed, the associations between nodes in the script structure data can be set. Through structured sorting, the content of the script to be analyzed can be converted into script structure data with clear structure, providing a data basis for subsequent script and visual effect requirement analysis.

[0058] Optionally, the script to be analyzed can be a complete script, a specified scene, a paragraph, or a segment in the script, such as a newly modified part or an added part of the script, which can be determined according to script analysis requirements. By determining all or part of the complete script as the script to be analyzed, both full script analysis and local analysis can be provided. The complete script is analyzed for overall budget evaluation and resource planning in the early stage of the project, and the specified scene, specific paragraph, or selected segment of the script is analyzed for targeted analysis to meet the dynamic adjustment requirements of script modification, scene addition, etc., and to realize incremental updating.

[0059] Step 120: performing cross-scene state analysis on target entity nodes in the script structure data to obtain visual effect identification data.

[0060] The target entity node can be a data node in the script structure data for representing each entity in each scene. The cross-scene state analysis can be an operation of obtaining the state of each entity in each scene and analyzing the visual effect production requirements generated by the state. The visual effect identification data can be data marking that the entity represented by the target entity node in the script structure data generates a visual effect production requirement in the corresponding scene.

[0061] Based on the structured presentation of the content of the script to be analyzed according to the script structure data, the basic entities such as roles, props, vehicles, and animals in each scene of the script to be analyzed can be determined without a hierarchical structure, and the scene entities and their hierarchical structure in each scene, as well as the events in each scene and the influence of the events on the state of the entities and scenes can also be determined. Therefore, the cross-scene state analysis is performed on each target entity node in the script structure data, the state of the entity represented by any target entity node in each scene and the state change between scenes can be obtained, and the visual effect identification data is generated to record that the state change of the entity generates the corresponding visual effect production requirement.

[0062] Step 130, performing visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain a visual effect analysis result.

[0063] The visual effect demand analysis can be an operation of determining all visual effect production contents required in the script to be analyzed. The visual effect analysis result can include data recording all visual effect production contents required in the script to be analyzed.

[0064] According to the script structure data, the information of the scenes, the scenes in each scene, the events, and the basic entities contained in the content of the script to be analyzed can be determined, and in combination with the visual effect production requirements generated by each entity in each scene recorded by the visual effect identification data, the script to be analyzed can be analyzed, the specific production contents corresponding to all visual effect production requirements generated by the content of the script to be analyzed can be determined, and the visual effect analysis result can be generated.

[0065] The technical scheme of the embodiment generates the script structure data of the script to be analyzed, and performs cross-scene state analysis on the target entity nodes in the script structure data to obtain the visual effect identification data, thereby performing visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain the visual effect analysis result, solving the problem of lack of document processing capability and scene correlation analysis capability in the prior art, and realizing flexible and general, accurate and efficient automatic script analysis.

[0066] Embodiment two

[0067] Figure 2A flowchart of a script analysis method provided for the second embodiment of the present application, the embodiment further refines the above technical solution, and can further include, before generating the script structure data of the to-be-analyzed script, performing text layout analysis on the original script to obtain text image data; extracting target text data from the text image data according to a preset script format; and performing paragraph optimization processing on the target text data to obtain the to-be-analyzed script. The method specifically includes:

[0068] Step 210: performing text layout analysis on the original script to obtain text image data.

[0069] The original script can be a script document file in any format. The text layout analysis can be an operation of determining the region position of each part of the text content presented in each page of the original script. The text image data can be an image of the region of the text text presenting the script content in the original script.

[0070] The original script can generally include various file formats and text formats, for example, the original script can be a Word document file, a PDF file, and a scanned picture, and different original scripts can use different identifiers to express script chapters. The header and the footer can contain words that are not script text content, and the original script can also contain revision marks of the script. The traditional document processing technology lacks a general recognition method suitable for all scripts, for example, it is impossible to write a general string matching rule or regular expression suitable for all script formats, so it is impossible to directly recognize the correct and complete text content from the original script. Therefore, the text layout analysis can be performed on the original script to ensure that the text content in the text image data obtained therefrom is the text text of the script content.

[0071] Optionally, the text layout analysis on the original script to obtain the text image data can include: acquiring target page samples in the original script according to a preset sampling strategy; performing page layout analysis on each target page sample to obtain a layout analysis result; determining a page clipping strategy according to the layout analysis result; and acquiring the text image data in the original script according to the page clipping strategy.

[0072] The preset sampling strategy can be a strategy for acquiring a page sample representative of the page layout of the page recording the script text in the original script. The target page sample can be a page sample recording the script text in the original script selected according to the preset sampling strategy. The page layout analysis can be an operation of determining the page region position of the script text in the target page sample. The layout analysis result can be the page region position of the script text in the target page sample. The page clipping strategy can be a strategy for clipping the region of the script text from the complete page.

[0073] The complete original script includes the script content, and the beginning and end parts can also include pages recording cover, table of contents, and production information, etc. Therefore, a preset sampling strategy can be prepared in advance to ensure that the target page samples are obtained from the pages containing the script content in the original script. The preset sampling strategy can also limit the number of pages from which the target page samples are obtained, and can also disperse the positions of the target page samples in the original script to ensure that the layout analysis result obtained from the target page samples can represent the area position of the script content in all pages recording the script content in the original script. Optionally, according to the preset sampling strategy, 3-5 typical content pages in the middle part can be selected as target page samples by avoiding the preset number of pages in the beginning and end parts of the original script.

[0074] Through page layout analysis, the layout analysis result of the target page sample can be obtained according to the identification, layout features, etc. of each part of the text in the page image of the target page sample. Optionally, the YOLO (You Only Look Once) model can be used to perform page layout analysis on the target page sample. The YOLO model can identify the elements in the page image without being disturbed by the watermark, such as DocLayout-YOLO v10, which can identify the areas of three types of key elements on the picture, including title, plain text, and abandon. The abandon area includes the header, footer, page number, and other non-analyzed script content. The YOLO model can return the coordinates of the identified areas.

[0075] The layout analysis result includes the position of the script content in the target page sample, so the page cropping strategy can be determined according to the layout analysis result to obtain the page image of the script content in the original script page through image cropping.

[0076] Optionally, determining the page cropping strategy according to the layout analysis result can include: calculating the consistency score of the layout analysis result of each target page sample; and determining that the consistency score exceeds a preset consistency threshold, and calculating an optimized cropping template according to the layout analysis result.

[0077] The consistency score can be a data describing the consistency between the layout analysis results of the target page samples. The preset consistency threshold can be a consistency score of the layout analysis results of the target page samples, which is high enough to determine the area of the script content in the original script. The optimized cropping template can be a template for cropping the script content pages of the original script.

[0078] The consistency score can be calculated according to the region position, structure and boundary and other information described in the layout analysis result of each target page sample. If the consistency score exceeds a preset consistency threshold, it can be indicated that the consistency of the layout analysis result obtained according to the current obtained target page sample is high enough, and can be used as a reliable judgment result of the layout of the page recording the script text in the original script, so that the optimized clipping template can be calculated according to the current obtained layout analysis result. Optionally, if the consistency score does not exceed the preset consistency threshold, the target page sample can be reacquired, and the page layout analysis can be performed on each target page sample to obtain the layout analysis result, so as to obtain the layout analysis result with high enough consistency score as the basis of the page clipping strategy.

[0079] Optionally, the consistency score of the layout analysis result of each target page sample can be obtained according to the coordinates returned by the YOLO model, the boundary position variance, the content area overlap and the structure similarity between target page samples, and specifically can be calculated according to the formula C = w1 x variance + w2 x overlap + w3 x similarity, wherein variance is the boundary position variance, overlap is the content area overlap, similarity is the structure similarity, w1, w2 and w3 are weight coefficients, and can be determined as 0.4, 0.3 and 0.3 according to the experimental experience value.

[0080] Step 220, extracting target text data from the text image data according to a preset script format.

[0081] The preset script format can be a text format required by a predetermined script, for example, can include chapter title format, dialogue format and the like. The target text data can be data of text text recording clean and complete script content in the preset script format.

[0082] The text image data contains images of script text, and no longer contains other content images in the script. Therefore, the text content can be extracted from the text image data by using related image processing technology, and the text of the script text content can be obtained. According to the preset script format, the obtained text can be further arranged into target text data.

[0083] Optionally, the target text data can be extracted from the body image data, which can include: obtaining the text in the body image data by an OCR (Optical Character Recognition) technology, and performing natural paragraph semantic classification labeling on the recognized text. Optionally, the natural paragraph labeling categories of the natural paragraph semantic classification labeling can include six types of scene title, action line, character, dialogue, transition, and others, wherein the transition type can refer to text content that does not involve visual effect production, such as "Act I", "flash", and the like; and the other type can refer to script information that does not involve visual effect analysis, such as screenwriter, producer, version, modification date, and the like.

[0084] Optionally, the target text data can also be extracted from the body image data, which can include: detecting the dialogue format and performing accurate dialogue splitting in the recognized text. Since most scripts are not standard script formats, non-standard dialogue formats often occur, so it is necessary to recognize and analyze the text in the composite format of "character + action description + colon + dialogue content", and uniformly convert it into a standard script dialogue format, such as converting the format of "Xiaoming wakes up and cries and rushes forward: 'Mother! Mother…'" which mixes action description and dialogue, into the standard format "Xiaoming (wakes up, cries and rushes forward) [line break] Mother! Mother…". The accurate dialogue splitting can be achieved by calling a large language model such as Qwen3 (Qwen3), which can accurately separate the character information, action information, and dialogue content in the text, and recombine them into two independent objects.

[0085] Step 230, performing paragraph optimization processing on the target text data to obtain a script to be analyzed.

[0086] The paragraph optimization processing can be an operation of verifying and arranging the text paragraph structure integrity and rationality of the target text data.

[0087] The target text data presents the script content in a preset script format, but due to the limitations of script pagination and text extraction, there can be text fragmentation problems in the target text data. Therefore, based on the grammatical and semantic integrity, the paragraph optimization processing of the target text data can be performed to obtain a script to be analyzed. The script to be analyzed can be used as the basis data for the subsequent steps to complete the whole process of script analysis. Optionally, the script to be analyzed can also be provided to the front-end display interface for visual display as a script analysis result.

[0088] Optionally, the target text data is subjected to paragraph optimization processing to obtain a script to be analyzed, which can include: performing punctuation integrity detection on the target text data, detecting whether a previous paragraph in the target text data ends with a complete punctuation mark by identifying various ending punctuation marks such as Chinese and English periods, exclamation marks, question marks, ellipses, and semicolons, and identifying continuous paragraphs that need to be merged; can also include performing label consistency detection on the target text data to determine whether the type labels of adjacent paragraphs are consistent to ensure that the merged paragraphs belong to the same content type; can also include performing coherence evaluation on the target text data to evaluate the semantic coherence of adjacent paragraph contents and determine whether merging is needed through semantic analysis.

[0089] Optionally, the final obtained script to be analyzed can be a standardized file in JSON (JavaScript Object Notation) format.

[0090] Exemplarily, Figure 3 A flowchart of a script standardization process provided by Embodiment Two of the present application is shown in FIG. 2. As shown in the figure, the flowchart can be integrated in a document standardization module to achieve the standardization of the input original script through the above-mentioned sampling layout detection, natural paragraph splitting, and cross-page detection steps, and to obtain a standardized JSON output script to be analyzed. Figure 3

[0091] The above-mentioned implementation introduces a double-layer architecture of a target detection model and a VL (Vision-Language) model to achieve the format unification processing of scripts from different sources. The double-layer architecture is characterized in that the layout detection layer is responsible for quickly locating the text area that needs semantic understanding, and the semantic understanding layer is responsible for understanding the script content and labeling the category of the natural paragraph. This method can unify the formats of all scripts with different layouts and different file formats, providing convenience for front-end display and system analysis. In view of the problem that the traditional rule matching method cannot solve the script format and layout unification problem, the above-mentioned implementation has significant generality advantage and does not require maintenance cost. In actual tests, 7 different formats of scripts can be effectively parsed, and the accuracy is above 96%. Among them, the accuracy of OCR reaches 100%, and there is a small amount of error in the classification task, but this error will not affect the front-end display effect and the later script analysis processing.

[0092] Step 240, generating script structure data of the script to be analyzed.

[0093] Optionally, generating the script structure data of the script to be analyzed can include: performing target structuring processing on the script to be analyzed to obtain structured scripts of each session; constructing a script knowledge graph according to the structured scripts to obtain the script structure data. ​

[0094] The structured script includes target entity data and entity relationship data in the scene. The target entity data can be data defining and describing each entity in each scene of the script, and the entity relationship data can be data describing the relationship between each entity and other entities. The structured script can be a script with a preset specific data structure composed of target entity data and entity relationship data in each scene. The target structured processing can be an operation of extracting the required data from the script data to be analyzed and arranging the structured script data in a preset specific format. The script knowledge graph can be a knowledge graph for presenting each entity and the association between entities in the script text content.

[0095] Through target structured processing, entity information in each scene can be obtained from the script data to be analyzed, so that the entities and the relationship between entities contained in each scene can be determined according to the entity information, and the related data can be arranged into a structured script. The structured script can provide a clear entity boundary for subsequent construction of a script knowledge graph, ensuring data consistency and analysis accuracy of the entire analysis process.

[0096] For example, the structured script of any scene can have the following structure:

[0097] class SceneElements(BaseModel):

[0098] scene_info: Scene #scene basic information (title, interior and exterior, time, etc.)

[0099] characters: List[Character] #character information and relationship

[0100] locations: List[Location] #location information

[0101] props: List[Prop] #props and ownership

[0102] vehicles: List[Vehicle] #vehicle information

[0103] animals: List[Animal] #animal information

[0104] event: List[Event] #event information

[0105] Optionally, a large language model such as Qwen3 can be used to perform target structured processing on the script to be analyzed according to the scene, and obtain the structured script of each scene.

[0106] According to the target entity data in the structured script, a target entity node to be created in the script knowledge graph can be determined, and according to the target entity relationship data in the structured script, an associated relationship to be established between the target entity nodes can be determined, so as to construct the script knowledge graph and obtain the script structure data. Optionally, the script knowledge graph can be provided to a front-end display interface and visually presented as a script analysis result.

[0107] Optionally, constructing the script knowledge graph according to the structured script to obtain the script structure data can include: creating a target entity node according to the target entity data; constructing the target entity node into a target hierarchical graph according to the entity relationship data; and constructing each target hierarchical graph into the script knowledge graph according to the associated relationship between each target entity node and each scene, to obtain the script structure data.

[0108] The target hierarchical graph can be a knowledge graph with a hierarchical structure.

[0109] According to the target entity data, an entity in the scene can be determined, each entity can be represented by a node in the script knowledge graph, and therefore the target entity node can be created according to the target entity data. The entity relationship data describes the relationship between each entity and other entities, and therefore the connection structure relationship between each target entity node in the script knowledge graph can be determined according to the entity relationship data, so as to construct the target entity node into a target hierarchical graph. Further, according to the association between the target entity nodes and the scenes, each target hierarchical graph can be merged by associating the nodes of the target hierarchical graphs, to construct the script knowledge graph.

[0110] Optionally, the target hierarchical graph can include a basic entity layer graph, a scene hierarchical layer graph, and an event analysis layer graph.

[0111] The same target hierarchical graph can be composed of target entity nodes of the same type, wherein the basic entity layer graph is composed of basic entities such as scenes, roles, props, vehicles, and animals, which do not have an internal hierarchical structure. The scene entity usually has an internal hierarchical structure, and in visual effect content, the scene-related visual effects are very important analysis content, and therefore the scene hierarchical layer graph can be formed separately. The event usually affects the state of the basic entity and the scene, and directly affects the visual effect requirements of the script, and therefore the event analysis layer graph can be established by analyzing the influence of any event on the state of the related entities in the previous and subsequent scenes, and establishing the causal relationship between the event and the state change of the entity.

[0112] Optionally, the three-layer knowledge graph including the basic entity layer graph, the scene hierarchical layer graph, and the event analysis layer graph can be constructed in parallel, and after the construction is completed, the three-layer graphs can be merged through the scene nodes and stored in a preset graph structure database, so as to be retrieved in subsequent analysis steps.

[0113] The above embodiment performs global analysis based on a knowledge graph. The knowledge graph is a knowledge representation method based on a graph structure, which can abstract all elements in the script into nodes and edges in the graph structure through a triple pattern of entity-relation-attribute, and realize correlation analysis within the entire script through a graph traversal algorithm. The addition of the knowledge graph allows automatic identification of the evolution process of the same entity in different scenes. For example, a monster character is in a complete state in the 5th scene, is damaged by an attack in the 6th scene, self-repairs in the 8th scene, and completely reorganizes into a new form in the 20th scene. Traditional methods will treat these as four independent production tasks, but based on the above embodiment, it can be understood that these are four different states of the same character, and the continuity and consistency of the form conversion need to be considered. When the script describes the transformation of a city from prosperity to ruin, the system can identify the need to produce three versions of “city”, “city (war version)”, and “city (ruin version)”, and understand the progressive relationship between them. This correlation analysis can help the production team to reasonably arrange the modeling order and resource allocation. The addition of the knowledge graph can also alleviate the loss of information at the intermediate position when the large language model analyzes long text. When the model analyzes a very long script, it is easy to be distracted by irrelevant content in the script and ignore important information in the middle. Therefore, it is required that the large language model only makes judgments on specific entities in the knowledge graph, which can quickly and accurately extract key information that needs visual effect processing.

[0114] Step 250, performing cross-scene state analysis on the target entity nodes in the script structure data to obtain visual effect identification data.

[0115] Optionally, the cross-scene state analysis on the target entity nodes in the script structure data to obtain visual effect identification data can include: obtaining target entity data of each target entity node in each scene in turn according to a time sequence of the scenes; identifying target visual effect states of each target entity node according to the target entity data, and generating visual effect identification data for each target visual effect state.

[0116] The time sequence of the scenes can be a time sequence in which the scenes appear in the script content. The target visual effect state can be a state in which each entity has a visual effect production requirement.

[0117] The target entity nodes associated with each session can be obtained in time sequence of the sessions, and target entity data corresponding to the target entity nodes can be obtained. The target entity data describes the state of the corresponding entity, and the entity state described by the target entity data in different sessions can be different, that is, the entity state changes between sessions, and different states of the entity correspondingly produce different visual effect production requirements. Therefore, the target entity data of each target entity node in each session can be obtained in time sequence of the sessions, the state changes of each entity with the session and time can be obtained, and the target visual effect state of each entity in each state can be identified, so that corresponding visual effect identification data can be generated in the script structure data.

[0118] For example, for each entity in the script knowledge graph, such as a character, a prop, a scene, etc., the occurrence of the entity in different sessions can be traversed in time sequence, the state description of the entity in each session can be analyzed by a large language model, and state change nodes can be automatically identified, for example, when it is detected that a certain character changes from a "complete" state to a "damaged" state, the two states can be marked as producing different visual effect production requirements.

[0119] Optionally, the visual effect identification data can include visual effect asset markers. The visual effect asset marker can be a marker recording the visual effect asset corresponding to the target visual effect state of the entity. The visual effect asset can refer to a specific entity object that needs to be created by computer graphics technology, for example, can include a CG (Computer Graphics) character, that is, a person, an organism, a monster, etc. that needs to be digitized modeled; a digital prop, that is, an item, a weapon, equipment, etc. that cannot be shot or needs special effects; a digital scene, that is, a building, an environment, a background, etc. that needs to be constructed by CG; a digital carrier, that is, a carrier such as an aircraft, a future vehicle, etc. that cannot be shot. The visual effect asset needs to be completed in the pre-modeling stage.

[0120] Optionally, the visual effect identification data can also include a state evolution chain of each entity. For multiple target visual effect states of the same entity, a state evolution chain can be constructed to record the complete change process of the entity from the initial state to the final state. The state evolution chain can provide a clear visual effect asset production path and avoid the discontinuity problem between states.

[0121] For example, Figure 4 A flowchart for generating visual effect identification data is provided for the second embodiment of the present application. As Figure 4 shown, the flowchart can be integrated in the knowledge graph construction module to implement, by structuring the text of the JSON format script to be analyzed, a basic graph layer, a scene hierarchical layer, and an event analysis layer can be constructed to build and store the graph, and obtain script structure data. Further, cross-scene association analysis can be performed based on the script structure data to obtain visual effect identification data.

[0122] The above embodiment realizes breakthrough of cross-scene content association analysis, can automatically find visual effect production requirements of different states of the same entity, and avoids problems of possible omission of visual effect production requirements in independent paragraph analysis of traditional methods. In verification of the five test scripts, the overall recognition accuracy of the above embodiment reaches 85%.

[0123] Step 260, performing visual effect requirement analysis on the script to be analyzed according to the script structure data and the visual effect identification data, to obtain a visual effect analysis result.

[0124] Optionally, the visual effect requirement analysis on the script to be analyzed according to the script structure data and the visual effect identification data, to obtain a visual effect analysis result, can include: determining target visual effect assets according to the script structure data and the visual effect identification data; establishing a paragraph association relationship of the target visual effect assets and the script to be analyzed; and generating a visual effect analysis result according to the target visual effect assets and the paragraph association relationship.

[0125] The target visual effect assets can include all visual effect asset production requirements generated in the script to be analyzed. The paragraph association relationship can be a corresponding relationship of each target visual effect asset and a paragraph position generated in the script to be analyzed.

[0126] According to the script structure data and the visual effect identification data, the production requirements of the visual effect assets generated by the states of the entities in each scene can be determined, so that the target visual effect assets of the script to be analyzed can be determined. The production requirements of each visual effect asset are generated by different scenes and different states of different entities in the script to be analyzed, and each scene and each state of each entity can appear in the script to be analyzed in time sequence. Therefore, by establishing the paragraph association relationship of the target visual effect assets and the script to be analyzed, the position and order of the production requirements of each target visual effect asset appearing in the script to be analyzed can be determined, so that the visual effect content associated with each visual effect asset that needs to be produced can be analyzed.

[0127] It should be noted that the visual effect content can specifically refer to dynamic effect processing applied to visual effect assets or live-action pictures, and can include particle effects such as fire, explosion, smoke, and water splashes; environmental effects such as weather changes, light changes, and atmospheric effects; physical simulation such as collision, fragmentation, fluid, and cloth; and synthetic effects such as matte, color adjustment, and picture extension. Visual effects can usually be produced in a post-synthesis stage after a pre-modeling stage of producing visual effect assets, and therefore can be analyzed separately from visual effect assets.

[0128] Optionally, the visual effect analysis result can specifically include specific production types of visual effects and associate related target visual effect assets. The specific production types can include but are not limited to animation shots, special effect shots, material synthesis, and scene extension.

[0129] Optionally, the establishing of the association relationship between the target visual effect asset and the paragraph of the script to be analyzed can include: obtaining a shooting script paragraph of the script to be analyzed; and establishing a paragraph association relationship between each target visual effect asset and each shooting script paragraph.

[0130] The shooting script paragraph can be each paragraph of independent shooting obtained according to semantic analysis of the script to be analyzed. According to the semantics of the script to be analyzed, the script content can be segmented from the dimensions of scene change, role change and narrative rhythm, and the continuous content is divided into a plurality of relatively independent shooting script paragraphs, so that the production requirements of the target visual effect asset in each shooting script paragraph can be analyzed, and a visual effect analysis result more in line with the actual engineering requirements can be obtained.

[0131] Optionally, after obtaining the shooting script paragraph of the script to be analyzed, the method can further include: calculating a predicted number of shots of each shooting script paragraph.

[0132] The predicted number of shots can be the number of shots calculated for each shooting script paragraph. Through the prediction of the number of shots of the script to be analyzed, a quantitative basis can be provided for subsequent production pricing, further enriching the script analysis result. However, the prior art still has limitations in the aspect of script breaking, so the script can be segmented according to the semantics of the script, and the number of shots of each segment can be predicted.

[0133] Optionally, the calculation of the predicted number of shots of each shooting script paragraph can include the calculation of the number of shots of a dialogue scene, a fight scene and a mixed scene. The number of shots of the dialogue scene is calculated as: number of shots = number of dialogue turns × 1.5 + number of established shots (1-2), the number of shots of the fight scene is calculated as: number of shots = number of action segments × (2-3) + number of close-up shots, and the mixed scene adopts a segmented calculation and weighted combination method, considering the transition shots and transition requirements, and the weight can be calculated by considering factors such as the number of roles, emotional intensity, scene complexity, action complexity, number of participants, and space scale. The specific value can be set and adjusted according to experience.

[0134] Optionally, the visual effect demand analysis of the script to be analyzed according to the script structure data and the visual effect recognition data to obtain the visual effect analysis result can include: inputting the script structure data and the visual effect recognition data into a script analysis model to obtain a visual effect analysis result output by the script analysis model.

[0135] The script analysis model can be a pre-trained large language model, which can identify the target visual effect asset according to the input script structure data and visual effect recognition data, analyze the visual effect analysis result and output.

[0136] Optionally, before inputting the script structure data and visual effect identification data into the script analysis model and obtaining the visual effect analysis result output by the script analysis model, the method can further include: generating model training samples according to the visual effect judgment knowledge base; training the large language model according to the model training samples to obtain the script analysis model.

[0137] The large language model has strong semantic understanding ability and rich background knowledge of film and television production, and can accurately understand the script content and identify visual effect related elements. However, according to the prior art, it is found that the large language model has two levels of cognitive bias in film and television production: the first level is the basic concept understanding bias, even if the large language model has the concept of visual effect assets and visual effects, there will still be cognitive ambiguity in the task, and sound production will be included in the category of picture production; the second level is the production process understanding bias, the model will analyze all visual effect elements that need to be produced, but in the initial quotation stage, the production team pays more attention to the main visual effect assets rather than the detailed effects. Therefore, it cannot directly fit the actual production process.

[0138] To solve the above problems, a special visual effect judgment knowledge base can be constructed, which summarizes ten content classification systems that differ from the model's cognition and the industry's production cognition. Specifically, it can include natural climate, environmental atmosphere, characters, vehicles and props, sound effect description, non-visual narrative, shot movement, rhetorical description, animals, and environmental description. Each category in the visual effect judgment knowledge base contains detailed judgment rules, typical cases, and easily confused scene cases. For example, for character description text, it explains in detail which role changes require CG production and which can be realized by makeup technology.

[0139] Based on the visual effect judgment knowledge base, a large number of high-quality training samples can be automatically generated, including positive samples, which include scenes that require visual effect production; negative samples, which include scenes that are easily misjudged but do not actually require visual effects; and adversarial samples, which include easily confused boundary cases. The model training process can use reinforcement learning strategies to guide the model to learn the correct judgment standard through a reward mechanism. Specifically, it can include: giving a positive reward when the model correctly distinguishes between visual effects and sound effects, giving an additional reward when the model accurately identifies the main visual effect assets, and giving a negative reward when the model has conceptual confusion. After multiple rounds of iterative training, the model gradually internalizes the professional knowledge and process specifications of film and television production, and finally obtains a specially optimized script analysis model for subsequent online analysis processes.

[0140] Optionally, script structure data and visual effects recognition data can be input into the script analysis model to obtain the visual effects analysis results output by the script analysis model. Specifically, this may include: reading the visual effects recognition data of the current scene from the script knowledge graph based on the script analysis model, which may include a list of identified entities, such as characters, props, vehicles, animals, and other entities and their attribute information. Based on the script analysis model, judgments can be made only for specific entities in the script knowledge graph. Entity constraints effectively eliminate the interference of irrelevant text in the judgment process, ensuring that the model focuses on the identified entity objects, and ultimately outputting the target visual effects assets.

[0141] Optionally, inputting script structure data and visual effects recognition data into the script analysis model to obtain the visual effects analysis results output by the script analysis model may further include: intelligently segmenting the script scene content based on three dimensions—scene changes, character changes, and narrative rhythm—dividing continuous scene content into several relatively independent shooting segments; based on the segmentation results, the script analysis model predicts the number of shots in each segment based on preset shot number prediction rules. The preset shot number prediction rules may include formulas and rules for calculating the number of shots in dialogue scenes, action scenes, and mixed scenes.

[0142] Optionally, inputting script structure data and visual effects recognition data into the script analysis model and obtaining the visual effects analysis results output by the script analysis model may also include: based on the script analysis model, analyzing the content requirements for visual effects production according to the obtained target visual effects assets and shooting storyboard segments, including identifying the content that needs to produce visual effects in each shooting storyboard segment, determining the specific production type, and associating it with relevant target visual effects assets.

[0143] For example, Figure 5 This is a flowchart illustrating the training and operation of a script analysis model according to Embodiment 2 of the present invention. Figure 5 As shown, the process may include a reinforcement learning training phase, which can typically be executed offline. Based on a pre-built domain knowledge base, training samples are generated through a sample generator to perform reinforcement learning training on the LLM (Large Language Model) to obtain an optimized LLM. Figure 5The model workflow shown can be executed online in general and can be divided into a visual effect asset judgment stage, a shot number prediction stage, and a visual effect demand analysis stage. In the visual effect asset judgment stage, the entity list corresponding to the visual effect recognition data in the script knowledge graph and the divided scene text paragraphs in the script to be analyzed are input into the optimized LLM to obtain a target visual effect asset list. In the shot number prediction stage, the scene text paragraphs are input into the semantic segmenter in the model to obtain a shot number prediction result. In the visual effect demand analysis stage, the scene text paragraphs and the visual effect asset list obtained from the foregoing processes are comprehensively analyzed, the optimized LLM is input for asset correlation checking, and a visual effect demand list and a visual effect classification are generated.

[0144] The foregoing embodiment provides a targeted model training optimization method based on the functions of an existing large language model, corrects and optimizes the functional defects of the existing large language model, and uses the optimized model to automatically analyze the visual effect demand and segmentation planning of a script, thereby achieving efficient and comprehensive automatic analysis of the script.

[0145] Optionally, after the visual effect demand analysis of the script to be analyzed based on the script structure data and the visual effect recognition data is performed to obtain a visual effect analysis result, the following steps can be further included: outputting a target visual effect asset to a visual effect asset judgment interface; outputting a predicted shot number to a shot number prediction interface; and outputting a visual effect analysis result to a visual effect analysis interface.

[0146] The visual effect asset judgment interface, the shot number prediction interface, and the visual effect analysis interface can be preset independent service interfaces that can be flexibly called according to actual needs. Optionally, the visual effect asset judgment interface, the shot number prediction interface, and the visual effect analysis interface can be connected to a front-end display interface to provide the data output to the interface to the front-end display interface for visual presentation.

[0147] An exemplary visual effect asset judgment interface is shown in the following table. Figure 6 A flowchart of a script analysis method provided by the second embodiment of the present application is shown in the following figure. Figure 6 As shown, by performing document standardization processing, knowledge graph construction, and visual effect content analysis on the input script, the script analysis products obtained at each stage can be provided to a front-end display interface to form a visual script analysis report.

[0148] The script analysis method provided by the embodiment of the present application comprehensively improves the script and visual effect content analysis process in multiple dimensions. The complete script visual effect content analysis time is shortened from several workdays in the traditional manual method to about 50 minutes. Taking a script of 174 scenes and 50 pages as an example, the format unification takes 10 minutes and 54 seconds, the structuring takes 6 minutes and 42 seconds, the knowledge graph construction takes 20 minutes and 45 seconds, and the visual effect analysis takes 5 minutes and 30 seconds. The analysis efficiency is improved by 20 to 30 times. In addition, by providing correlation analysis on the entire script, reusable visual effect assets can be identified, and resource waste caused by repeated modeling can be avoided. At the same time, the knowledge graph technology provides visual display of the correlation information of the internal elements of the script, which provides an intuitive auxiliary tool for manual checking and reduces the item-by-item checking workload by about 80 to 90%.

[0149] The technical scheme of the embodiment generates script structure data of a script to be analyzed, performs cross-scene state analysis on target entity nodes in the script structure data, obtains visual effect identification data, and performs visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain a visual effect analysis result, thereby solving the problem of lack of document processing capability and scene correlation analysis capability in the prior art and realizing flexible and universal, accurate and efficient automatic script analysis.

[0150] Embodiment three

[0151] Figure 7 A structural schematic diagram of a script analysis device provided for the third embodiment of the present application is shown in Figure 7 The script analysis device includes a structure generation module 310, a visual effect identification module 320 and an analysis result module 330, wherein

[0152] The structure generation module 310 is configured to generate script structure data of a script to be analyzed.

[0153] The visual effect identification module 320 is configured to perform cross-scene state analysis on target entity nodes in the script structure data to obtain visual effect identification data.

[0154] The analysis result module 330 is configured to perform visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain a visual effect analysis result.

[0155] The technical scheme of the embodiment generates script structure data of a script to be analyzed, performs cross-scene state analysis on target entity nodes in the script structure data, obtains visual effect identification data, and performs visual effect demand analysis on the script to be analyzed according to the script structure data and the visual effect identification data to obtain a visual effect analysis result, thereby solving the problem of lack of document processing capability and scene correlation analysis capability in the prior art and realizing flexible and universal, accurate and efficient automatic script analysis.

[0156] Optionally, the script analysis device can further include a script processing module configured to perform text layout analysis on the original script to obtain text image data, extract target text data from the text image data according to a preset script format, and perform paragraph optimization processing on the target text data to obtain a script to be analyzed.

[0157] Optionally, the structure generation module 310 can include a structure unit configured to perform target structuring processing on the script to be analyzed to obtain structured scripts for each scene, wherein the structured scripts include target entity data and entity relationship data in the scene, and a graph construction unit configured to construct a script knowledge graph according to the structured scripts to obtain script structure data.

[0158] Optionally, the graph construction unit can be specifically configured to create target entity nodes according to the target entity data, construct the target entity nodes into target hierarchical graphs according to the entity relationship data, and construct the target hierarchical graphs into the script knowledge graph according to the association between each target entity node and each scene to obtain the script structure data.

[0159] Optionally, the visual effect recognition module 320 can include a data acquisition unit configured to acquire target entity data of each target entity node in each scene in sequence according to a time sequence of the scenes, and a state recognition unit configured to recognize target visual effect states of each target entity node according to the target entity data and generate visual effect recognition data for each target visual effect state.

[0160] Optionally, the analysis result module 330 can include an asset determination unit configured to determine target visual effect assets according to the script structure data and the visual effect recognition data, a paragraph association unit configured to establish an association between the target visual effect assets and paragraphs of the script to be analyzed, and a result generation unit configured to generate a visual effect analysis result according to the association between the target visual effect assets and the paragraphs.

[0161] Optionally, the analysis result module 330 can include a model data input unit configured to input the script structure data and the visual effect recognition data into a script analysis model, and a model output acquisition unit configured to acquire a visual effect analysis result output by the script analysis model.

[0162] The script analysis device can further include a model training module configured to generate a model training sample according to a visual effect judgment knowledge base, and train a large language model according to the model training sample to obtain a script analysis model.

[0163] The script analysis device provided in the embodiments of the present application can perform the script analysis method provided in any of the embodiments of the present application, and has the corresponding functional modules and beneficial effects of the execution method.

[0164] Embodiment Four

[0165] Figure 8 A structural schematic diagram of a script analysis device provided in Embodiment Four of the present application is shown in the figure, which includes a processor 510, a memory 520, an input device 530 and an output device 540; the number of processors 510 in the script analysis device can be one or more, and one processor 510 is taken as an example in the figure; the processor 510, the memory 520, the input device 530 and the output device 540 in the script analysis device can be connected through a bus or other means, and the connection through a bus is taken as an example in the figure. Figure 8 Figure 5 Figure 5

[0166] The memory 520 is a kind of computer readable storage medium, which can be used to store software programs, computer executable programs and modules, such as program instructions / modules (for example, the structure generation module 310, the visual effect identification module 320 and the analysis result module 330 in the script analysis device) corresponding to the script analysis method in the embodiment of the present application. The processor 510 executes the software programs, instructions and modules stored in the memory 520, thereby performing various functional applications and data processing of the script analysis device, that is, realizing the script analysis method described above.

[0167] The memory 520 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application program required by a function; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory 520 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some examples, the memory 520 can further include a memory remotely arranged with respect to the processor 510, which can be connected to the script analysis device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0168] The input device 530 can be used to receive input digital or character information, and generate key signal input related to user settings and function control of the script analysis device. The output device 540 can include a display device such as a display screen.

[0169] Embodiment Five

[0170] Embodiment Five of the present application also provides a storage medium containing computer executable instructions, which are used to execute a script analysis method when executed by a computer processor, including:

[0171] Generating script structure data of a script to be analyzed;

[0172] ​​​Perform cross-episode state analysis on the target entity node in the script structure data to obtain visual effect identification data;

[0173] Perform visual effect demand analysis on the to-be-analyzed script according to the script structure data and the visual effect identification data to obtain a visual effect analysis result.

[0174] Of course, the storage medium provided by the embodiment of the present application includes computer executable instructions, which are not limited to the method operations described above, and can also perform related operations in the script analysis method provided by any embodiment of the present application.

[0175] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., including a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in various embodiments of the present application.

[0176] It is worth noting that in the above embodiment of the script analysis device, each unit and module included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy mutual distinction, and do not limit the protection scope of the present application.

[0177] Although the present application has been described in detail in the foregoing general description, specific embodiments and experiments, some modifications or improvements can be made on the basis of the present application, which is obvious to those skilled in the art. Therefore, these modifications or improvements made on the basis of not deviating from the spirit of the present application are all within the scope of protection claimed by the present application.

Claims

1. A script analysis method, characterized in that, include: Generate script structure data for the script to be analyzed; Perform cross-scene state analysis on the target entity nodes in the script structure data to obtain visual effects recognition data; Based on the script structure data and the visual effects recognition data, a visual effects requirement analysis is performed on the script to be analyzed to obtain visual effects analysis results; the visual effects analysis results include the production type of visual effects and are associated with target visual effects assets; The types of visual effects produced include animated shots, special effects shots, material compositing, and scene extensions; The script structure data for generating the script to be analyzed includes: The script to be analyzed is subjected to target structuring processing to obtain the structured script for each scene; wherein, the structured script includes target entity data and entity relationship data in the scene; A script knowledge graph is constructed based on the structured script to obtain the script structure data.

2. The method according to claim 1, characterized in that, Before generating the script structure data for the script to be analyzed, the following is also included: Text layout analysis was performed on the original script to obtain the main text image data; According to the preset script format, extract the target text data from the main text image data; The target text data is subjected to paragraph optimization processing to obtain the script to be analyzed.

3. The method according to claim 1, characterized in that, The step of constructing a script knowledge graph based on the structured script to obtain the script structure data includes: Create the target entity node based on the target entity data; Based on the entity relationship data, the target entity nodes are constructed into a target hierarchical graph; Based on the association between each target entity node and each scene, the target hierarchical graph is constructed into the script knowledge graph to obtain the script structure data.

4. The method according to claim 1, characterized in that, The cross-scene state analysis of the target entity nodes in the script structure data to obtain visual effects recognition data includes: According to the time sequence of the sessions, obtain the target entity data of each target entity node in each session in sequence; The target visual state of each target entity node is identified based on the target entity data, and the visual recognition data is generated for each target visual state.

5. The method according to claim 1, characterized in that, The step of performing visual effects requirement analysis on the script to be analyzed based on the script structure data and the visual effects recognition data to obtain visual effects analysis results includes: The target visual effects asset is determined based on the script structure data and the visual effects recognition data; Establish the paragraph association relationship between the target visual assets and the script to be analyzed; The visual effects analysis results are generated based on the relationship between the target visual effects asset and the paragraph.

6. The method according to claim 1, characterized in that, The step of performing visual effects requirement analysis on the script to be analyzed based on the script structure data and the visual effects recognition data to obtain visual effects analysis results includes: The script structure data and the visual effects recognition data are input into the script analysis model; Obtain the visual effects analysis results output by the script analysis model; Before performing visual effects requirement analysis on the script to be analyzed based on the script structure data and the visual effects recognition data to obtain the visual effects analysis results, the method further includes: Model training samples are generated based on a visual effects judgment knowledge base; The script analysis model is obtained by training the large language model using the training samples of the model.

7. A script analysis device, characterized in that, include: The structure generation module is used to generate script structure data for the script to be analyzed. The visual effects recognition module is used to perform cross-scene state analysis on the target entity nodes in the script structure data to obtain visual effects recognition data. The analysis results module is used to perform visual effects requirement analysis on the script to be analyzed based on the script structure data and the visual effects recognition data, and obtain visual effects analysis results; the visual effects analysis results include the production type of visual effects and associate them with target visual effects assets; The types of visual effects produced include animated shots, special effects shots, material compositing, and scene extensions; The structure generation module includes: The structured unit is used to perform target structuring processing on the script to be analyzed to obtain the structured script for each session; wherein, the structured script includes target entity data and entity relationship data in the session; The graph construction unit is used to construct a script knowledge graph based on the structured script to obtain the script structure data.

8. A script analysis device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the script analysis method as described in any one of claims 1-6.

9. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the script analysis method as described in any one of claims 1-6.

Images