Artificial intelligence scenario multi-user collaborative generation method and system

Abstract

The application belongs to the technical field of digital content creation operation and collaborative management, and particularly relates to an AI plot multi-user collaborative generation method and system, which comprises the following steps: obtaining initial data containing themes, roles and styles, extracting emotional tones and node conflict intensities through semantic analysis, generating a structured demand feature set, and realizing standardized control of creation demands. A plot script is generated based on the demand features, abnormal nodes are positioned through multi-dimensional coherence detection, role behavior parameters are adjusted to obtain a revised plot structure, and content quality is automatically controlled. Scene node user feedback is collected, a plot deviation index is calculated in combination with a timestamp and user participation, and the plot deviation index is compared with a trigger condition. When the condition is met, scene descriptions are first generated by fusing feedback semantic features, scene versions are then generated based on detail density, response delay and interaction frequency, demand features are adjusted in combination with change records, style matching degrees and collaboration trajectories, a creation chain version is determined, and closed-loop iteration and full-link traceable management are realized.

Description

Technical Field

[0001] This invention belongs to the field of digital content creation, operation and collaborative management technology, specifically involving an AI-powered multi-user collaborative generation method and system for storylines. Background Technology

[0002] With the large-scale development of the digital content industry, narrative content, as a core production element in interactive entertainment, film and television media, online education, and other fields, has evolved from independent production by a single entity to collaborative creation by multiple users. In commercial content production and operation scenarios, how to achieve standardized control, content quality supervision, and version iteration management of multi-entity collaborative creation through intelligent data processing has become a core technological requirement in the field of digital content operation and management, falling under the category of data processing technologies for production management and commercial operation.

[0003] Currently, AI-generated content technology is widely used in assisting in the creation of narrative content. However, most existing technical solutions are designed for single-user creation scenarios, and can only generate narrative content based on scattered text prompts. They lack a structured control mechanism for multi-user collaborative needs. The creative requests submitted by multiple users are mostly unstructured and loose text, which cannot be transformed into reusable and traceable standardized constraints. This leads to the narrative deviating from the core creative direction after multiple rounds of collaboration, significantly increasing the management and control costs of content operation.

[0004] Secondly, existing collaborative creation platforms mostly only provide basic collaborative functions such as text editing and commenting, lacking the ability to automatically detect and correct the coherence of plot content and the consistency of character settings. The logical connection between plot scenes and the verification of the rationality of character behavior rely heavily on manual review. In large-scale creation scenarios with multiple users participating in parallel, manual review is inefficient and lacks standardized criteria, making it difficult to meet the management needs of large-scale content production.

[0005] Furthermore, existing technologies lack quantifiable deviation judgment criteria and hierarchical integration mechanisms when processing feedback from multiple users. When feedback from multiple sources is scattered and conflicting, it is impossible to achieve orderly screening and weighted integration of opinions based on feedback sequence and user contribution. This easily leads to problems such as local opinions dominating the overall content and repeated revisions failing to reach a creative consensus. At the same time, existing solutions lack version tracking and dynamic adjustment mechanisms for the entire creative process, failing to transform consensus feedback into creative constraints to achieve closed-loop iteration and making it difficult to support refined operation and management throughout the entire lifecycle of story creation. Summary of the Invention

[0006] To address the above issues, this invention provides an AI-powered multi-user collaborative plot generation method and system, which at least solves the problem of how to ensure consistent requirements, logical coherence, and closed-loop updates based on feedback when multiple users are involved in collaborative production.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an AI-based multi-user collaborative generation method for storylines, comprising: Obtain initial data containing themes, roles, and styles; perform semantic parsing on the initial data to extract sentiment tone and node conflict intensity, thereby obtaining a set of demand features; A plot script is generated based on the set of required features. The plot script is then tested and the character behavior parameters are adjusted to obtain a revised plot structure. Obtain feedback data of scene nodes in the revised plot structure, and determine the deviation index based on the timestamp of the feedback data and the user participation level determined by the feedback data; The deviation index is compared with the management trigger condition; when the deviation index meets the management trigger condition, the first management and the second management are executed sequentially. The first management includes: extracting and fusing the feedback semantic features of the feedback data to obtain a scene description; The second management includes: generating a scene version based on the detail density of the scene description, the response latency and interaction frequency corresponding to the feedback data, adjusting the set of demand features based on the change records, style matching degree and collaboration trajectory characteristics of the scene version, and determining the creation chain version.

[0008] Preferably, the step of performing semantic parsing on the initial data to extract sentiment tone and node conflict intensity to obtain a set of demand features includes: Text features are extracted from the theme, the role, and the style in the initial data to obtain theme features, role features, and style features; The emotional tone is determined based on the thematic features and the stylistic features; The intensity of node conflict is determined based on the differences in role objectives and role relationships in the role characteristics. The set of requirement features is generated based on the theme features, the role features, the style features, the emotional tone, and the node conflict intensity.

[0009] Preferably, the node conflict intensity is used to indicate the difference in the characters' goals, the degree of conflict corresponding to the characters' relationships, and the degree of emotional change in the scene nodes; The set of requirements features is used to constrain the theme direction, character behavior, style expression, and conflict intensity in the plot script.

[0010] Preferably, the step of detecting and adjusting character behavior parameters in the plot script to obtain a revised plot structure includes: The scene sequence, character motivations, event causality, and scene transitions in the plot script are detected to obtain coherence detection results; Based on the coherence detection results, scene nodes with abnormal coherence are identified from the plot script; Based on the character settings and adjacent scene relationships corresponding to the scene nodes with abnormal continuity, the character behavior parameters are adjusted to obtain the corrected plot structure.

[0011] Preferably, the role behavior parameters include the intensity of the role's goal, action tendency, amplitude of emotional reaction, and degree of conflict participation; The adjustment of the character behavior parameters includes: Based on the character settings corresponding to the scene nodes with coherence anomalies and the relationships between adjacent scenes, determine the parameters to be adjusted from the character behavior parameters; Adjust the parameters to be adjusted so that the node conflict intensity corresponding to the scene node with the coherence anomaly matches the event causality between adjacent scene nodes.

[0012] Preferably, the step of obtaining feedback data of scene nodes in the revised plot structure, and determining the deviation index based on the timestamp of the feedback data and the user participation level determined by the feedback data, includes: Obtain the feedback data associated with the scene node in the revised plot structure, wherein the feedback data includes feedback text, the timestamp, and the user identifier; The feedback order is determined based on the timestamp; The user engagement level is determined based on the number of feedback responses and the number of feedback responses adopted, which are associated with the user identifier. The deviation index is determined based on the feedback text, the feedback order, and the user engagement.

[0013] Preferably, the deviation index is used to indicate the degree of deviation between the plot content corresponding to the scene node and the set of demand features and the feedback text, wherein the plot content is obtained by parsing the corrected plot structure; Determining the deviation index based on the feedback text, the feedback order, and the user engagement includes: Semantic matching is performed between the plot content corresponding to the scene node and the set of demand features to obtain a first deviation value; Semantic matching is performed between the plot content corresponding to the scene node and the feedback text to obtain a second deviation value; The deviation index is obtained by weighting the second deviation value based on the user engagement level and combining it with the first deviation value.

[0014] Preferably, the step of extracting and fusing the feedback semantic features of the feedback data to obtain a scene description includes: Semantic features are extracted from the feedback text to obtain the feedback semantic features; The feedback opinion category is determined based on the semantic similarity between the feedback semantic features; The category weight of the feedback category is determined based on the user engagement level; According to the category weight, the feedback semantic features corresponding to the feedback opinion category are fused with the plot content corresponding to the scene node to obtain the scene description.

[0015] Preferably, the step of generating a scene version based on the detail density of the scene description, the response latency and interaction frequency corresponding to the feedback data, and adjusting the set of demand features based on the scene version's change history, style matching degree, and collaboration trajectory characteristics to determine the creation chain version includes: The text units of character behavior, dialogue content, and scene environment in the scene description are parsed, and the detail density is determined based on the number of parsed text units. The response delay is defined as the time interval between receiving the feedback data and generating the scenario version. The number of feedback data generated for the scene node per unit time is defined as the interaction frequency. When the detail density meets the density triggering condition, the response delay meets the delay triggering condition, or the interaction frequency meets the frequency triggering condition, the cyclic triggering parameters are adjusted, and the scene version is generated according to the adjusted cyclic triggering parameters, wherein the cyclic triggering parameters include the feedback scan cycle and the scene version generation cycle; The style matching degree is determined based on the degree of matching between the scene version and the style in the initial data; The collaboration trajectory characteristics are determined based on the timestamp, user engagement, and change records of the feedback data; The set of required features is adjusted based on the change records, the style matching degree, and the collaboration trajectory characteristics to obtain an updated set of required features, and the version of the creation chain is determined based on the updated set of required features.

[0016] Secondly, the present invention provides an AI-powered multi-user collaborative plot generation system, comprising: The requirement processing and correction module is used to obtain initial data containing theme, character and style, perform semantic parsing on the initial data, extract emotional tone and node conflict intensity to obtain a requirement feature set; generate a plot script based on the requirement feature set, and test and adjust the character behavior parameters of the plot script to obtain a corrected plot structure. The condition determination module is used to obtain feedback data of scene nodes in the revised plot structure, determine the deviation index based on the timestamp of the feedback data and the user participation level determined by the feedback data, and compare the deviation index with the management trigger conditions. The first management execution module is used to perform first management when the deviation index meets the management triggering condition, extract and fuse the feedback semantic features of the feedback data to obtain a scenario description; The second management execution module is used to perform second management when the deviation index meets the management triggering condition, generate a scene version based on the detail density of the scene description, the response delay and interaction frequency corresponding to the feedback data, adjust the set of demand features based on the change records, style matching degree and collaboration trajectory characteristics of the scene version, and determine the creation chain version.

[0017] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: 1. By semantically analyzing the initial data of themes, characters, and styles, and extracting the emotional tone and node conflict intensity, the system realizes the structured transformation of user creative intentions into a set of demand features. Its role is to establish standardized creative demand control rules, so that the subsequent plot generation is subject to unified constraints of theme direction, character behavior, style expression, and conflict intensity, thereby reducing the demand control cost in multi-subject collaborative scenarios and realizing reusable and traceable management of creative demands.

[0018] 2. By generating plot scripts based on a set of demand characteristics and adjusting character behavior parameters after checking the coherence of the plot scripts, a closed-loop correction of scene order, character motivation, event causality, and scene transitions in the plot scripts is achieved. Its role is to build an automated content quality control mechanism, reduce abrupt character behavior and plot continuity issues, reduce the management costs of manual editing and review, and ensure consistent quality standards under large-scale creation.

[0019] 3. By acquiring feedback data from scene nodes and combining it with timestamps and user engagement to determine the deviation index, a quantitative judgment on the degree of deviation between collaborative feedback and plot content is achieved. Its purpose is to establish a quantifiable content iteration trigger standard, accurately distinguish between scene nodes that need to absorb feedback and nodes that can remain stable, improve the control accuracy of the collaborative creation process, and avoid management chaos caused by subjective decisions.

[0020] 4. By extracting and fusing feedback semantic features when the deviation index exceeds the deviation threshold, the orderly fusion of feedback from multiple people into scene description is achieved. This is to form a standardized multi-source opinion integration and management mechanism, avoid repeated modifications caused by scattered feedback, reduce collaborative operation losses, and improve the efficiency of creative decision-making.

[0021] 5. By generating scene versions based on detail density, response latency, and interaction frequency, and adjusting the set of requirement features in conjunction with change records, style matching degree, and collaboration trajectory characteristics, reverse correction between plot versions and requirement constraints is achieved. Its role is to build a closed-loop operation and management system for the entire creative lifecycle, precipitate user consensus into standardized control rules, and maintain theme stability, style consistency, and traceability of the creative chain in multiple rounds of collaboration. Attached Figure Description

[0022] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a block diagram of the module combination of the system of the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solution, the present invention will be described in detail below with reference to embodiments. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0024] In multi-user collaborative story creation, different users typically focus on different aspects such as theme development, character settings, plot conflicts, scene atmosphere, and language style. A single generated result needs to maintain consistency and traceability across multiple sources of feedback. While deep learning models can generate story content from input text, without structured analysis and version update mechanisms for multi-user feedback, the generated content is easily influenced by local opinions and deviates from the overall creative goal. Therefore, it is necessary to incorporate initial user settings, plot nodes, feedback data, and version changes into the same collaborative processing chain, enabling multi-user feedback to be identified, filtered, integrated, and used to influence the story generation process. Based on this, this invention constructs a processing mechanism for AI-driven multi-user collaborative story generation, encompassing requirement analysis, plot correction, feedback deviation judgment, semantic fusion, and creation chain updates.

[0025] like Figure 1 As shown, the AI-powered multi-user collaborative plot generation method includes: Obtain initial data containing themes, roles, and styles; perform semantic parsing on the initial data; extract sentiment tone and node conflict intensity; and obtain a set of demand features. The server receives initial data submitted by the creator, including theme text, character setting text, and style description text. The server performs format validation and field merging on the initial data, storing the theme, character, and style in their respective data fields, and removing duplicate punctuation, whitespace, and invalid input. The server then uses a semantic parsing model to perform text segmentation, entity recognition, semantic tagging, and sentence relationship analysis on the initial data. It extracts the emotional tone from the theme and style, and identifies character goals, identities, and relationships from the character settings. The server further determines the intensity of node conflict by combining differences in character goals and character relationships, and writes the emotional tone, node conflict intensity, theme constraints, character constraints, and style constraints into the same requirement feature set. This requirement feature set is stored in association with the current creation task identifier, serving as the input basis for subsequent plot script generation, character behavior control, and feedback deviation judgment.

[0026] Semantic parsing is performed on the initial data to extract the sentiment tone and node conflict intensity, resulting in a set of demand features. This includes: extracting text features from the theme, role, and style in the initial data to obtain theme features, role features, and style features; determining the sentiment tone based on the theme features and style features; determining the node conflict intensity based on the differences in role goals and role relationships in the role features; and generating a set of demand features based on the theme features, role features, style features, sentiment tone, and node conflict intensity.

[0027] In one embodiment, before the initial data enters the semantic parsing process, the server establishes structured fields for theme, role, and style to prevent input from different sources from mixing into the same text block and causing ambiguity in the parsed object. The theme field records the main plot, event scope, and narrative goal; the role field records the role name, identity attributes, goal requests, relationship descriptions, and behavioral boundaries; and the style field records language style, narrative mood, and scene atmosphere. The server performs a minimum integrity check on each field. If the theme field lacks a clear event scope, the user input is retained and marked as incomplete theme information; if the role field lacks a goal request, the role goal is set to a pending state; if the style field lacks a clear description, the style constraint is set to a default neutral expression. The pending state does not directly block the process but is recorded with low confidence in the requirement feature set, so that the control weight of this field is reduced when the plot script is generated.

[0028] Text feature extraction is accomplished using a combination of natural language processing models and rule-based vocabulary. The server extracts theme features from topic texts, including core events, narrative objects, temporal context, spatial context, and plot objectives. The server extracts character features from character texts, including character identity, character objectives, dependencies, conflicts, collaborations, and information asymmetry between characters. The server extracts style features from style texts, including narrative tone, emotional tendency, rhythmic tendency, and expressive constraints. For fields with multiple sentences, the server preserves the contextual relationship according to sentence order, avoiding the fragmentation of character objectives or style tendencies due to keyword-based extraction. When conflicting descriptions exist within the same field, the server determines primary and secondary labels based on user clarity, description location, and semantic strength; for example, if the same style field contains both "relaxed" and "tense," and the topic field is a suspenseful event, the server will use "tense" as the primary emotional label and "relaxed" as a secondary expressive label.

[0029] The emotional tone is jointly determined by thematic and stylistic features. The server matches the event risk, narrative goal, and scene type from thematic features with the emotional tendency from stylistic features to obtain candidate emotional tags. Candidate emotional tags can include commonly used tags in content generation systems in this field, such as tension, depression, lightheartedness, warmth, suspense, action, and absurdity. The server does not require all tags to be effective simultaneously; instead, it determines a primary emotional tone and at least one secondary emotional tone based on the semantic strength of the input text. Semantic strength is determined based on the number of emotional words, their position, the risk of the thematic event, and the clarity of the stylistic description. If thematic and stylistic features conflict, the server prioritizes retaining the tone explicitly specified by the user in the stylistic field and uses the conflicting emotional tendency in the thematic field as a reference for plot node changes to prevent subsequent plot script generation from deviating from the user's settings.

[0030] The intensity of node conflict is determined by the differences in role goals and role relationships within the role characteristics. The server compares whether there is resource competition, opposing positions, information concealment, task conflict, or emotional conflict between different role goals, and determines the source of conflict based on role relationships. When role goals are completely consistent and the role relationship is cooperative, the node conflict intensity is set to low level; when role goals are partially consistent but implemented differently, or when there are implicit contradictions in the role relationship, the node conflict intensity is set to medium level; when role goals are mutually exclusive, and the role relationship includes opposition, betrayal, misunderstanding, or coercive constraints, the node conflict intensity is set to high level. The boundaries of each level are jointly set by the sample script annotation results, manually configured rules, and historical generated task records. The server simultaneously stores the conflict level and the basis for forming the conflict level in the requirement feature set, enabling the subsequent plot script generation stage to clearly identify which scenes require strong conflict and which scenes only need to maintain narrative connection.

[0031] The node conflict intensity is used to indicate the differences in character objectives, the degree of conflict in character relationships, and the degree of emotional change in scene nodes; the set of demand features is used to constrain the theme direction, character behavior, style expression, and conflict intensity in the plot script.

[0032] In one embodiment, node conflict intensity is used not only to indicate whether conflict exists between roles, but also to limit the degree and direction of conflict development within scene nodes. When identifying differences in role goals, the server breaks down these differences into criteria such as whether goals are opposite, whether goals compete for the same resource, whether the order of goal completion is mutually exclusive, and whether the goal is hindered by the actions of other roles. When identifying role relationships, the server breaks down these relationships into types such as cooperation, dependence, competition, opposition, concealment, misunderstanding, and temporary alliance. The degree of emotional change is determined by emotional words, negative expressions, transitional expressions, and intensity adverbs in the role's text and style text, used to determine whether a role's emotions have changed significantly within the same scene node. Node conflict intensity is formed by the above criteria, not by a single keyword, avoiding the direct determination of high conflict solely based on the appearance of words like "argument" or "danger."

[0033] The server sets grading rules for node conflict intensity. Low grade indicates minor differences in character goals, no significant antagonism in character relationships, and relatively mild emotional changes; medium grade indicates competition or tension between characters, but they still share common tasks or have room for reconciliation; high grade indicates mutually exclusive character goals, antagonistic relationships, and emotional changes that directly impact the plot. Grading rules can be configured with different weights based on plot type. Suspense, adventure, and competitive plots assign higher weights to event risk and goal conflict; family, school, and slice-of-life plots assign higher weights to the degree of conflict in character relationships and the degree of emotional change. The configuration results are written to the task configuration file and associated with the set of requirements features. Subsequent scene node generation uses the same configuration to avoid inconsistencies in conflict level judgment standards across different creative tasks.

[0034] The set of requirement features is used to constrain the theme direction, character behavior, style expression, and conflict intensity in the plot script. The server writes theme features into the theme constraint field, character features into the character constraint field, style features and emotional tone into the style constraint field, and node conflict intensity into the conflict constraint field. The theme constraint field restricts the plot script from deviating from the main events and narrative goals; the character constraint field restricts character behavior to conform to the character's identity, goals, and relationships; the style constraint field restricts the plot language, scene atmosphere, and emotional expression; and the conflict constraint field restricts the degree of conflict development at different scene nodes. Each field retains a source field identifier, which records whether the feature comes from theme input, character input, or style input, facilitating the differentiation of the deviation object during subsequent deviation judgment.

[0035] After the requirement feature set is generated, the server performs a consistency check on the set. The check includes verifying whether there are irreconcilable conflicts between theme and style features, whether character objectives lack behavioral boundaries, and whether the intensity of node conflicts is significantly inconsistent with character relationships. If the check passes, the server writes the requirement feature set into the cache and persistent database of the story generation task. If the check fails, the server does not discard the user input directly, but marks conflicting fields as pending confirmation fields and retains more explicit input content according to pre-set priorities. For example, if the style field explicitly specifies "suspense and tension," and the theme field only describes "friends traveling," the server retains "suspense and tension" as the primary style constraint and "friends traveling" as the story background constraint. This approach ensures that the requirement feature set has stable input boundaries, allowing the subsequent story script generation stage to read consistent theme direction, character behavior constraints, style expression constraints, and conflict intensity constraints.

[0036] A plot script is generated based on the set of demand features. The plot script is then checked for coherence. Based on the check results, the character behavior parameters are adjusted to obtain a revised plot structure. The server reads the theme constraints, character constraints, style constraints, and conflict constraints from the requirement feature set, inputs these constraints into the plot generation model, and generates a plot script containing multiple scene nodes. Each scene node records the scene order, participating characters, character behaviors, dialogue content, event triggering conditions, and scene transition relationships. The server performs a coherence check on the plot script, checking scene order, character motivations, event causality, and scene transitions. When the check results indicate a coherence anomaly, the server locates the abnormal scene node, reads the corresponding character settings and the relationships between preceding and following scenes, and adjusts the character behavior parameters within the scope of not changing the core constraints in the requirement feature set. The adjusted scene node is rewritten into the plot script, forming a revised plot structure, which serves as the basis for subsequent feedback data acquisition and deviation judgment.

[0037] The script is subjected to a coherence check, and the character behavior parameters are adjusted based on the check results to obtain a revised plot structure. This includes: checking the scene order, character motivations, event causality, and scene transitions in the script to obtain coherence check results; identifying scene nodes with inconsistencies in the script based on the coherence check results; and adjusting the character behavior parameters based on the character settings and the relationship between the preceding and following scenes corresponding to the scene nodes with inconsistencies to obtain a revised plot structure.

[0038] In one embodiment, before the plot script enters the coherence detection process, the server first breaks down the plot script into a sequence of scene nodes and establishes a scene identifier for each scene node. The sequence of scene nodes is arranged according to the narrative order in the plot script, and each scene node includes at least the scene time, scene location, participating characters, character behaviors, character dialogues, event results, and transition descriptions. The server verifies whether the scene nodes still revolve around the main event based on the theme constraints in the requirement feature set, verifies whether the behaviors of the participating characters are consistent with their roles, goals, and relationships based on the character constraints, verifies whether the scene expression deviates significantly from the established emotional tone based on the style constraints, and verifies whether the intensity of conflict in the scene nodes is within the allowable range based on the conflict constraints.

[0039] Continuity checks include scene sequence checks, character motivation checks, event causality checks, and scene transition checks. Scene sequence checks determine if there are time reversals, premature events, or missing necessary events between scene nodes. Character motivation checks determine if character behavior is foreshadowed and consistent with the character's goals, identity, and relationships. Event causality checks determine if the current scene's outcome has a corresponding cause and if there are any unexplained sudden results. Scene transition checks determine if adjacent scenes have evidence of location changes, time changes, event continuity, or emotional progression. The server generates a check tag for each check, categorized as normal, suspected anomaly, and anomaly. Normal tags do not trigger corrections; suspected anomaly tags are added to the review list; and anomaly tags directly enter the correction process.

[0040] The continuity detection result consists of scene identifier, anomaly type, anomaly location, associated role, preceding scene node, and subsequent scene node. Anomaly types include sequence anomalies, motivational anomalies, causal anomalies, and transition anomalies. The server determines the correction direction based on the anomaly type. For sequence anomalies, priority is given to adjusting the order of scene nodes or adding transition descriptions; for motivational anomalies, priority is given to adjusting the character's goal intensity and action tendency in the character's behavior parameters; for causal anomalies, priority is given to supplementing the event trigger conditions or adjusting the event outcome; for transition anomalies, priority is given to adjusting the scene transition descriptions and the emotional progression. For suspected abnormal scene nodes in the review list, the server performs a secondary judgment based on the relationship between preceding and following scenes. If the secondary judgment still does not meet the continuity requirements, it is treated as an abnormal scene node; if the secondary judgment can be explained by character settings or style settings, the original scene node is retained and the judgment basis is recorded.

[0041] The criteria for coherence detection can be jointly set by sample script annotation results, manually configured rules, and historical generation task records. Sample script annotation results provide reference boundaries for common narrative sequences, character motivations, and event causality; manually configured rules constrain necessary scene relationships for specific plot types; and historical generation task records identify recurring anomaly types in similar tasks. The server binds and saves the detection results to the plot script, ensuring that each correction can be traced back to a specific scene node and anomaly type. This approach ensures that correcting the plot structure is not a complete rewrite of the plot script, but rather a localized correction around the coherence anomaly locations, avoiding disruption of thematic direction, character behavior constraints, and style expression constraints within the requirement feature set.

[0042] The role behavior parameters include the intensity of the role's goal, action tendency, emotional reaction amplitude, and degree of conflict participation; adjusting the role behavior parameters includes: identifying the parameters to be adjusted from the role behavior parameters; adjusting the parameters to be adjusted so that the node conflict intensity corresponding to the scene node with abnormal coherence matches the event causal relationship between adjacent scene nodes.

[0043] In one embodiment, the character behavior parameters include the character's goal intensity, action tendency, emotional reaction amplitude, and conflict participation level. The goal intensity represents the degree of action a character is willing to take to achieve a goal, determined by the server based on the character's goal requirements, urgency, and obstacles. Action tendency indicates the character's tendency to proactively advance, passively respond, avoid conflict, or seek cooperation in the current scenario. Emotional reaction amplitude represents the degree of emotional change a character experiences in the face of changing events, determined based on the character's setting, the emotional tone of the scene, and the relationship between preceding and subsequent scenes. Conflict participation level indicates the depth of a character's involvement in the conflict at the current scene node, determined based on character relationships, differences in character goals, and the node's conflict intensity.

[0044] Based on the character settings and the relationship between preceding and following scenes, the server determines the parameters to be adjusted from the character's behavioral parameters for scene nodes with continuity anomalies. If the anomaly originates from a character suddenly performing a high-risk action without supporting context, the server prioritizes determining the character's goal intensity or action tendency as the parameter to be adjusted. This is done by reducing the abruptness of the action or adding goal-related transitional behaviors to ensure the character's behavior connects seamlessly with the preceding scene. If the anomaly originates from excessive emotional changes in a character without event triggers, the server prioritizes determining the magnitude of the emotional reaction as the parameter to be adjusted, and adjusts the emotional changes based on the event pressure, relationship changes, and information revelation levels in preceding and following scenes. If the anomaly originates from a mismatch between the conflict intensity and adjacent scenes, the server prioritizes determining the level of conflict participation as the parameter to be adjusted, ensuring that the conflict intensity corresponding to the anomaly scene node matches the causal relationship between events in adjacent scene nodes.

[0045] Once the adjustment parameters are determined, the server only makes local adjustments to abnormal scene nodes and their necessary adjacent scene nodes. Local adjustments include modifying character behavior descriptions, supplementing character decision-making basis, adjusting emotional expression in dialogue, adding event triggering conditions, or compressing abrupt conflicts. The server does not change the already determined thematic constraints, basic character identities, and main style constraints in the requirement feature set. For abnormal scene nodes involving multiple characters, the server determines the adjustment order according to the character relationships. Characters who drive plot twists are prioritized for adjusting the intensity of their character goals and action tendencies, while supporting characters are prioritized for adjusting their level of conflict participation and the magnitude of their emotional reactions. If the differences in goals of multiple characters simultaneously lead to anomalies, the server determines whether to retain the antagonistic relationship based on the conflict level corresponding to the node conflict intensity; low-level conflicts primarily focus on supplementing motivations and mitigating transitions, medium-level conflicts primarily focus on clarifying goal differences and event triggers, and high-level conflicts primarily focus on supplementing the basis for conflict escalation and causal continuity.

[0046] After adjusting the character behavior parameters, the server regenerates the text content of the abnormal scene nodes and writes the regenerated text content back into the story script. The server then performs a coherence check on the rewritten scene nodes, including the adjusted scene node, the preceding scene node, and the following scene node. If the check result meets the coherence requirements, the server confirms the adjusted story script as the corrected story structure. If the check result still shows abnormalities, the server selects the parameters to be adjusted again based on the abnormality type and limits the number of repeated adjustments to prevent the same scene node from deviating from the initial character settings in multiple rounds of adjustments. If the requirement is still not met after reaching the limit of repeated adjustments, the server retains the version with the highest check score and marks that scene node as a scene node that needs to receive priority user feedback in the subsequent feedback data acquisition phase. This mark is output along with the corrected story structure, so that subsequent user feedback can be focused on the story positions where coherence is still controversial.

[0047] Obtain feedback data on scene nodes in the revised plot structure, and determine the deviation index based on the timestamp of the feedback data and the user engagement determined by the feedback data; The server reads scene nodes from the revised plot structure and retrieves corresponding feedback data based on the scene node identifiers. Feedback data includes feedback text, timestamps, and user identifiers. The server sorts the feedback data by timestamp to obtain the feedback order and counts the number of feedback responses and the number of responses adopted based on the user identifiers to determine user engagement. The server performs semantic matching between the plot content corresponding to each scene node and the set of requirement features and the feedback text, obtaining two deviation results to represent the plot content's deviation from the initial requirements and collaborative feedback. The server combines the feedback order and user engagement to determine a deviation index and associates the deviation index with the scene nodes for later use in determining whether feedback semantic feature fusion is necessary.

[0048] Obtain feedback data for scene nodes in the revised plot structure, and determine the deviation index based on the timestamps of the feedback data and the user engagement determined by the feedback data. This includes: obtaining feedback data associated with scene nodes in the revised plot structure, wherein the feedback data includes feedback text, timestamps, and user identifiers; determining the feedback order based on the timestamps; determining user engagement based on the number of feedback instances associated with the user identifiers and the number of feedback adoptions; and determining the deviation index based on the feedback text, feedback order, and user engagement.

[0049] In one embodiment, when the server obtains feedback data associated with scene nodes in the revised storyline structure, it establishes a feedback association table for each scene node. The feedback association table records the scene node identifier, feedback text, timestamp, and user identifier. The scene node identifier comes from the node division results in the revised storyline structure; the feedback text comes from the text content entered by collaborating users in the storyline editing interface, comment interface, or modification suggestion interface; the timestamp is written by the server when receiving feedback data; and the user identifier is generated from the login account, temporary collaboration identity, or role permission record of the collaboration platform. Before writing feedback data into the feedback association table, the server performs a validity check on the feedback text, including whether the feedback text is empty, whether it has a semantic relationship with the scene node, and whether it is a duplicate submission. Empty text and feedback data that cannot be matched with a scene node are not included in the deviation index determination; duplicate submissions of feedback data are merged and recorded according to the same user identifier and similar timestamps.

[0050] The server determines the feedback order based on timestamps. This feedback order represents the sequential relationship of multiple feedback data within the same scenario node. The server arranges the feedback data from earliest to latest timestamp and writes a sequence number in the feedback association table. If multiple feedback data have the same timestamp, the server generates a sequence number based on the data reception order and user identifier, ensuring that each valid feedback data has a unique sequential position. The feedback order does not solely determine the degree of deviation; rather, it is used to determine the time period in which feedback occurs within a set and whether the feedback consistently points to the same issue. For feedback texts with short intervals and similar semantics, the server groups them into the same feedback group; for feedback texts with longer intervals but changing semantic direction, the server retains different feedback groups to avoid mixing early opinions and subsequent corrections in the calculation.

[0051] User engagement is determined based on the number of feedback submissions and the number of feedback adoptions associated with a user ID. The number of feedback submissions represents the number of valid feedback data submitted by the same user ID in the current creation task, while the number of feedback adoptions represents the number of times the user ID's feedback data was written into the scene version or used as the basis for scene description fusion. The server can further adjust user engagement by combining the number of scene nodes the user participated in and the validity of the feedback text. The user engagement value can be set between zero and one, where zero indicates that the user ID's feedback data does not participate in the deviation index weighting, and one indicates that the user ID's feedback data has the highest participation weight. For newly added user IDs, the server can set a default engagement level, determined based on the member permissions of the collaborative task or the initial system configuration. For user IDs with a high number of feedback submissions but a low number of adoptions, the server reduces user engagement to prevent a large number of low-relevance feedback submissions from changing the deviation index. For user IDs with a low number of feedback submissions but a high number of adoptions, the server maintains a higher user engagement level, allowing high-quality feedback to be included in the deviation assessment.

[0052] The deviation index is used to indicate the degree of deviation between the plot content corresponding to a scene node and the set of demand features and feedback text. The deviation index is determined based on the feedback text, feedback order, and user participation, including: performing semantic matching between the plot content corresponding to the scene node and the set of demand features to obtain a first deviation value; performing semantic matching between the plot content corresponding to the scene node and the feedback text to obtain a second deviation value; weighting the second deviation value according to user participation and combining it with the first deviation value to obtain the deviation index.

[0053] In one embodiment, the deviation index is used to indicate the degree of deviation between the plot content corresponding to a scene node and the set of requirements features and feedback text. The server performs semantic matching between the plot content corresponding to the scene node and the set of requirements features to obtain a first deviation value. The first deviation value reflects the deviation of the plot content from the initial theme direction, character behavior constraints, style expression constraints, and conflict intensity constraints. The server performs semantic matching between the plot content corresponding to the scene node and the feedback text to obtain a second deviation value. The second deviation value reflects the deviation of the plot content from the feedback opinions of collaborating users. Semantic matching can be performed using text vector similarity, semantic tag consistency detection, or natural language inference models. The server selects one of these methods based on the existing model capabilities of the plot generation system, or converts the results of multiple methods into a deviation value within the same range.

[0054] The deviation index can be determined as follows: in, The deviation index represents the degree of deviation between the plot content corresponding to a scene node and the combined deviation from the set of demand features and feedback text. The first deviation value indicates the degree of deviation between the plot content and the set of required features; The second deviation value indicates the degree of deviation between the plot content and the feedback text; User engagement is used to represent the weight of the user identifier corresponding to the feedback text in the current creation task. The initial demand constraint weight controls the proportion of influence of the initial demand on the deviation index. The above expression merges the initial demand deviation and the feedback text deviation into a comparable judgment data, outputting the deviation index. The server determines whether the current scene node enters the feedback semantic feature fusion process based on the comparison between the deviation index and the deviation threshold.

[0055] The weight of requirement constraints is determined by the creation task configuration. For narrative tasks that require strict adherence to theme and style, the server increases the weight of requirement constraints, making the set of requirement features have a greater impact on the deviation index. For narrative tasks primarily involving collaborative rewriting by multiple users, the server decreases the weight of requirement constraints, making the feedback text have a greater impact on the deviation index. The deviation threshold is set based on the narrative type, the importance of scene nodes, and historical generation records. Key scene nodes are used to carry major plot twists, and their deviation thresholds are set lower, allowing even slight deviations to trigger feedback semantic feature fusion. Ordinary transitional scene nodes are used to connect narrative content, and their deviation thresholds can be set higher to avoid frequent triggering of the fusion process. When saving the deviation index, the server simultaneously saves the first deviation value, the second deviation value, user participation, and the weight of requirement constraints, enabling the subsequent scene description generation stage to trace the source of deviation. If the first deviation value is high and the second deviation value is low, it indicates that the narrative content deviates from the initial requirements but is close to the feedback text; if the first deviation value is low and the second deviation value is high, it indicates that the narrative content meets the initial requirements but has not absorbed collaborative feedback. The server selects different fusion strategies based on different sources of deviation, providing clear input for subsequent feedback semantic feature fusion.

[0056] When the deviation index exceeds the deviation threshold, the feedback semantic features of the feedback data are extracted and fused to obtain a scene description; The server reads the deviation index associated with scene nodes and compares it with a deviation threshold. The deviation threshold, set based on the scene node's importance, plot type, and historical modification history, determines whether the current scene node needs to incorporate collaborative feedback. When the deviation index exceeds the threshold, the server reads the feedback text from the feedback data, extracts semantic features from the feedback text, and obtains feedback semantic features. The server categorizes feedback opinions according to the semantic similarity between these features and assigns category weights based on user engagement. These category weights determine the order and proportion of different feedback opinion categories entering the scene description. The server then merges the feedback semantic features with the corresponding plot content of the scene node according to the category weights, generates a scene description, and writes the scene description back to the corresponding scene node as the content basis for subsequent scene version generation.

[0057] When the deviation index exceeds the deviation threshold, the feedback semantic features of the feedback data are extracted and fused to obtain a scene description, including: extracting semantic features from the feedback text to obtain feedback semantic features; determining the feedback opinion category based on the semantic similarity between feedback semantic features; determining the category weight of the feedback opinion category based on user participation; and fusing the feedback semantic features corresponding to the feedback opinion category with the plot content corresponding to the scene node according to the category weight to obtain a scene description.

[0058] In one embodiment, the server initiates a feedback fusion process when the deviation index exceeds a deviation threshold. This process adds limitations on the semantic features of the feedback, the category of the feedback opinion, and the category weight. The deviation threshold is not a fixed text marker, but is set in relation to the importance of the scene node. Scene nodes that bear the main plot turning point, changes in character relationships, or emotional shifts use a lower deviation threshold, while ordinary narrative transition scene nodes use a higher deviation threshold. When comparing the deviation index and the deviation threshold, the server simultaneously reads the scene node identifier, plot content, feedback text, feedback order, and user engagement, avoiding directly rewriting the plot content based on a single piece of feedback text.

[0059] When the server extracts semantic features from the feedback text, it first segments the text into sentences, removes repetitive expressions and invalid interjections, and then identifies the object of modification, the direction of modification, and the intensity of feedback. The object of modification can be character behavior, character dialogue, scene atmosphere, plot rhythm, event causality, or stylistic expression. The direction of modification indicates whether the feedback text requests the addition, reduction, replacement, retention, or reinterpretation of a plot element. The intensity of feedback is determined based on the explicitness, negation, emotional words, and suggested actions within the feedback text. The server incorporates the object of modification, the direction of modification, and the intensity of feedback into the feedback semantic features, enabling these features to be used for subsequent classification and fusion, rather than simply being stored as ordinary text vectors.

[0060] The server determines the feedback category based on the semantic similarity between feedback semantic features. Semantic similarity can be determined by combining text vector similarity, keyword overlap, and consistency of the modified object. When feedback semantic features point to the same scene node, the same modified object, and the same modification direction, the server classifies them into the same feedback category. When feedback semantic features point to the same modified object but the modification direction is opposite, the server classifies them into different feedback categories and marks them as conflict categories. When feedback semantic features point to different modified objects, the server establishes categories according to character behavior, scene atmosphere, plot pacing, event causality, and stylistic expression. After the feedback categories are formed, the server records the number of feedback texts contained in each category, the user identifier of the feedback source, the average feedback time, and the feedback intensity for use in determining category weights.

[0061] Category weights are determined based on user engagement and adjusted by considering the number of feedback submissions within the feedback category, feedback adoption records, and conflict status. Feedback submitted by users with high engagement contributes significantly to their respective feedback categories. When multiple users express the same direction of modification, the server increases the category weight for that feedback category. Even if a feedback category has a small number of submissions, but originates from users with high historical adoption rates, the server retains the effective weight for that category. For conflicting categories, the server does not directly aggregate opposing opinions. Instead, it prioritizes feedback categories with higher weights and more consistent with the set of requirements, using those with lower weights as supplementary information or content awaiting confirmation. This avoids the simultaneous appearance of contradictory character behaviors or stylistic expressions in the scenario description.

[0062] The server merges the semantic features of feedback categories with the plot content of scene nodes according to category weights. The fusion process is based on the original plot content of the scene node, without directly replacing the entire scene node with feedback text. For character behavior feedback, the server adjusts the character action description and motivation explanation based on the feedback semantic features; for character dialogue feedback, the server rewrites the dialogue while maintaining character identity and style constraints; for scene atmosphere feedback, the server adjusts the environmental description and emotional words; for event causality feedback, the server supplements the event triggering conditions or corrects the event outcome. After fusion, the server checks whether the scene description still meets the requirements for theme direction, character behavior, style expression, and conflict intensity in the requirement feature set. If the check passes, the scene description is written back to the scene node; if the check fails, the server reduces the category weight of the feedback category that caused the deviation and re-executes the fusion until a scene description consistent with both the requirement feature set and the high-weight feedback categories is generated.

[0063] The scene version is generated based on the detail density of the scene description, the response latency and interaction frequency corresponding to the feedback data, and the set of requirements features is adjusted based on the scene version change record, style matching degree and collaboration trajectory characteristics to determine the creation chain version.

[0064] The server reads the scene description and divides the character behaviors, dialogue content, and scene environment in the scene description into text units to obtain the detail density. The server reads the reception time of the feedback data and the scene version generation time to determine the response latency, and counts the number of feedback data generated by the same scene node per unit time to determine the interaction frequency. The server compares the detail density, response latency, and interaction frequency with their corresponding thresholds. When any indicator meets the trigger condition, the server adjusts the loop trigger parameters and generates a scene version according to the adjusted feedback scanning cycle and scene version generation cycle. The server records the changes in the scene version relative to the original scene description, determines the degree of matching between the scene version and the style in the initial data, and combines the timestamps of the feedback data, user engagement, and change records to form a collaboration trajectory feature. The server adjusts the requirement feature set based on the change records, style matching degree, and collaboration trajectory features to obtain an updated requirement feature set, and writes the updated requirement feature set into the creation chain version.

[0065] A scene version is generated based on the detail density of the scene description, the response latency corresponding to the feedback data, and the interaction frequency. The requirement feature set is then adjusted based on the scene version's change history, style matching degree, and collaboration trajectory characteristics to determine the creation chain version. This includes: determining the detail density based on the number of text units corresponding to character behavior, dialogue content, and scene environment in the scene description; determining the response latency as the time interval between receiving feedback data and generating the scene version; determining the interaction frequency as the number of feedback data generated for scene nodes per unit time; adjusting the loop trigger parameters when the detail density is below a detail density threshold, the response latency exceeds a latency threshold, or the interaction frequency is below an interaction frequency threshold, and generating a scene version based on the adjusted loop trigger parameters, which include the feedback scanning cycle and the scene version generation cycle; determining the style matching degree based on the matching degree between the scene version and the style in the initial data; determining the collaboration trajectory characteristics based on the timestamps of the feedback data, user engagement, and change history; adjusting the requirement feature set based on the change history, style matching degree, and collaboration trajectory characteristics to obtain an updated requirement feature set, and determining the creation chain version based on the updated requirement feature set.

[0066] In one embodiment, when the server generates a scene version of the scene description, it adds a linkage constraint between detail density, response latency, interaction frequency, and loop triggering parameters. Detail density is used to represent the sufficiency of content in the scene description that can be understood and modified by collaborating users. The server uses character behavior, dialogue content, and scene environment as three text unit sources. Character behavior text units include actions, decisions, goal changes, and reasons for actions; dialogue content text units include character speech, dialogue partners, and dialogue emotions; and scene environment text units include location, time, environmental state, and atmosphere descriptions. The server counts the number of these text units and determines a detail density threshold based on the scene node type. Key transition scenes require a higher detail density threshold, while ordinary transition scenes can use a lower detail density threshold. When the detail density is lower than the detail density threshold, the server marks the scene node as a scene node that needs supplementary description and prioritizes supplementing the missing text units when generating the scene version.

[0067] Response latency represents the time consumed when feedback data enters the system and is written into the scene version. The server writes the reception time upon receiving feedback data and the version generation time upon completing scene version generation, using the time interval between the two as the response latency. The latency threshold is set based on the scale of the collaborative task, the number of participating users, and the server's processing capacity. Lower latency thresholds are used for multi-user real-time collaborative tasks, while higher latency thresholds are used for asynchronous creation tasks. When the response latency exceeds the threshold, the server shortens the feedback scanning cycle or increases the execution priority of the scene version generation cycle to prevent feedback data from remaining in a pending state for extended periods.

[0068] Interaction frequency represents the amount of feedback data generated for scene nodes per unit of time. The server counts the amount of feedback data according to the scene node identifier and calculates the interaction frequency according to a preset time window. The interaction frequency threshold is set based on the importance of the scene node and the historical feedback density. When the interaction frequency is lower than the threshold, the server determines that the current scene description may lack sufficient content for discussion, or that collaborating users are not paying enough attention to the scene node. Without changing the topic direction and role settings, the server extends the feedback scanning cycle or adds modifiable scene details, and generates a scene version according to the adjusted cyclic trigger parameters. The cyclic trigger parameters include the feedback scanning cycle and the scene version generation cycle. The feedback scanning cycle controls the frequency at which the server reads feedback data, and the scene version generation cycle controls the frequency at which the server writes the fusion result into the scene version.

[0069] After a scene version is generated, the server compares it with the original scene description to obtain a change log. The change log includes newly added text units, deleted text units, replaced text units, adopted feedback data, and unadopted feedback data. The server determines the style matching degree based on the degree of matching between the scene version and the style in the initial data. The matching objects include narrative tone, emotional tendency, and scene atmosphere. The server also determines collaboration trajectory features based on the timestamps of the feedback data, user engagement, and the change log. These collaboration trajectory features represent the path of the feedback data's impact on the scene version at different time periods. If the change log shows that a certain type of feedback is continuously adopted, and the style matching degree is not lower than a set range, the server writes the semantic features corresponding to that type of feedback into the requirement feature set. If the style matching degree is lower than the set range, the server reduces the weight of the feedback features that cause style deviation, while retaining the style constraints in the initial data. The updated requirement feature set and the scene version are written into the same creation chain version, ensuring that subsequent plot generation and feedback fusion can use consistent requirement constraints.

[0070] like Figure 2 As shown, an AI-powered multi-user collaborative plot generation system includes: The requirement processing and correction module is used to obtain initial data containing theme, character and style, perform semantic parsing on the initial data, extract emotional tone and node conflict intensity to obtain a requirement feature set; generate a plot script based on the requirement feature set, and test and adjust the character behavior parameters of the plot script to obtain a corrected plot structure. The condition determination module is used to obtain feedback data of scene nodes in the revised plot structure, determine the deviation index based on the timestamp of the feedback data and the user participation level determined by the feedback data, and compare the deviation index with the management trigger conditions. The first management execution module is used to perform first management when the deviation index meets the management triggering condition, extract and fuse the feedback semantic features of the feedback data to obtain a scenario description; The second management execution module is used to perform second management when the deviation index meets the management triggering condition, generate a scene version based on the detail density of the scene description, the response delay and interaction frequency corresponding to the feedback data, adjust the set of demand features based on the change records, style matching degree and collaboration trajectory characteristics of the scene version, and determine the creation chain version.

[0071] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the present invention. These examples are merely for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or variations without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, variations, or combinations, or the direct application of the concept and technical solution of the present invention to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. An AI-powered multi-user collaborative plot generation method, characterized in that, include: Obtain initial data containing themes, roles, and styles; perform semantic parsing on the initial data to extract sentiment tone and node conflict intensity, thereby obtaining a set of demand features; A plot script is generated based on the set of required features. The plot script is then tested and the character behavior parameters are adjusted to obtain a revised plot structure. Obtain feedback data of scene nodes in the revised plot structure, and determine the deviation index based on the timestamp of the feedback data and the user participation level determined by the feedback data; Compare the deviation index with the management trigger conditions; When the deviation index meets the management trigger condition, the first management and the second management are executed sequentially. The first management includes: extracting and fusing the feedback semantic features of the feedback data to obtain a scene description; The second management includes: generating a scene version based on the detail density of the scene description, the response latency and interaction frequency corresponding to the feedback data, adjusting the set of demand features based on the change records, style matching degree and collaboration trajectory characteristics of the scene version, and determining the creation chain version.

2. The AI-based multi-user collaborative plot generation method according to claim 1, characterized in that, The initial data is semantically parsed to extract sentiment tone and node conflict intensity, resulting in a set of demand features, including: Text features are extracted from the theme, the role, and the style in the initial data to obtain theme features, role features, and style features; The emotional tone is determined based on the thematic features and the stylistic features; The intensity of node conflict is determined based on the differences in role objectives and role relationships in the role characteristics. The set of requirement features is generated based on the theme features, the role features, the style features, the emotional tone, and the node conflict intensity.

3. The AI-based multi-user collaborative plot generation method according to claim 2, characterized in that, The node conflict intensity is used to indicate the difference in the target of the characters in the scene node, the degree of conflict and the degree of emotional change corresponding to the relationship between the characters; The set of requirements features is used to constrain the theme direction, character behavior, style expression, and conflict intensity in the plot script.

4. The AI-based multi-user collaborative plot generation method according to claim 1, characterized in that, The process of detecting and adjusting character behavior parameters in the plot script to obtain a revised plot structure includes: The scene sequence, character motivations, event causality, and scene transitions in the plot script are detected to obtain coherence detection results; Based on the coherence detection results, scene nodes with abnormal coherence are identified from the plot script; Based on the character settings and adjacent scene relationships corresponding to the scene nodes with abnormal continuity, the character behavior parameters are adjusted to obtain the corrected plot structure.

5. The AI-based multi-user collaborative plot generation method according to claim 4, characterized in that, The role behavior parameters include the intensity of the role's goals, action tendencies, the magnitude of emotional reactions, and the degree of conflict involvement. The adjustment of the character behavior parameters includes: Based on the character settings corresponding to the scene nodes with coherence anomalies and the relationships between adjacent scenes, determine the parameters to be adjusted from the character behavior parameters; Adjust the parameters to be adjusted so that the node conflict intensity corresponding to the scene node with the coherence anomaly matches the event causality between adjacent scene nodes.

6. The AI-based multi-user collaborative plot generation method according to claim 1, characterized in that, The step of obtaining feedback data for scene nodes in the revised plot structure, and determining the deviation index based on the timestamp of the feedback data and the user engagement level determined by the feedback data, includes: Obtain the feedback data associated with the scene node in the revised plot structure, wherein the feedback data includes feedback text, the timestamp, and the user identifier; The feedback order is determined based on the timestamp; The user engagement level is determined based on the number of feedback responses and the number of feedback responses adopted, which are associated with the user identifier. The deviation index is determined based on the feedback text, the feedback order, and the user engagement.

7. The AI-based multi-user collaborative plot generation method according to claim 6, characterized in that, The deviation index is used to indicate the degree of deviation between the plot content corresponding to the scene node and the set of demand features and the feedback text, wherein the plot content is obtained by parsing the corrected plot structure. Determining the deviation index based on the feedback text, the feedback order, and the user engagement includes: Semantic matching is performed between the plot content corresponding to the scene node and the set of demand features to obtain a first deviation value; Semantic matching is performed between the plot content corresponding to the scene node and the feedback text to obtain a second deviation value; The deviation index is obtained by weighting the second deviation value based on the user engagement level and combining it with the first deviation value.

8. The AI-based multi-user collaborative plot generation method according to claim 7, characterized in that, The step of extracting and fusing the feedback semantic features of the feedback data to obtain a scene description includes: Semantic features are extracted from the feedback text to obtain the feedback semantic features; The feedback opinion category is determined based on the semantic similarity between the feedback semantic features; The category weight of the feedback category is determined based on the user engagement level; According to the category weight, the feedback semantic features corresponding to the feedback opinion category are fused with the plot content corresponding to the scene node to obtain the scene description.

9. The AI-based multi-user collaborative plot generation method according to claim 1, characterized in that, The process involves generating a scene version based on the detail density of the scene description, the response latency corresponding to the feedback data, and the interaction frequency, and adjusting the set of requirement features based on the scene version's change history, style matching degree, and collaboration trajectory characteristics to determine the creation chain version, including: The text units of character behavior, dialogue content, and scene environment in the scene description are parsed, and the detail density is determined based on the number of parsed text units. The response delay is defined as the time interval between receiving the feedback data and generating the scenario version. The number of feedback data generated for the scene node per unit time is defined as the interaction frequency. When the detail density meets the density triggering condition, the response delay meets the delay triggering condition, or the interaction frequency meets the frequency triggering condition, the cyclic triggering parameters are adjusted, and the scene version is generated according to the adjusted cyclic triggering parameters, wherein the cyclic triggering parameters include the feedback scan cycle and the scene version generation cycle; The style matching degree is determined based on the degree of matching between the scene version and the style in the initial data; The collaboration trajectory characteristics are determined based on the timestamp, user engagement, and change records of the feedback data; The set of required features is adjusted based on the change records, the style matching degree, and the collaboration trajectory characteristics to obtain an updated set of required features, and the version of the creation chain is determined based on the updated set of required features.

10. An AI-powered multi-user collaborative plot generation system, characterized in that, include: The requirement processing and correction module is used to obtain initial data containing themes, roles and styles, perform semantic parsing on the initial data, extract the sentiment tone and node conflict intensity, and obtain a set of requirement features. A plot script is generated based on the set of required features. The plot script is then tested and the character behavior parameters are adjusted to obtain a revised plot structure. The condition determination module is used to obtain feedback data of scene nodes in the corrected plot structure, and determine the deviation index based on the timestamp of the feedback data and the user participation level determined by the feedback data. And compare the deviation index with the management trigger conditions; The first management execution module is used to perform first management when the deviation index meets the management triggering condition, extract and fuse the feedback semantic features of the feedback data to obtain a scenario description; The second management execution module is used to perform second management when the deviation index meets the management triggering condition, generate a scene version based on the detail density of the scene description, the response delay and interaction frequency corresponding to the feedback data, adjust the set of demand features based on the change records, style matching degree and collaboration trajectory characteristics of the scene version, and determine the creation chain version.

Images