Script analysis system and method based on large language model

Abstract

The present application provides a script analysis system and method based on a large language model. In the system, the data processing module performs feature engineering, data integration and annotation on the script content data, script feedback data and production experience data to obtain a feature data set; the model training module fine-tunes the pre-trained large language model on the feature data set for the preset natural language understanding task to obtain a script analysis model; based on the supervised training of the given time series neural network in the feature data set, a script value prediction model is obtained; the report service module responds to the analysis request for the target script, respectively calls the script analysis model and the script value prediction model, obtains the analysis results and feedback data of the target script, and generates an analysis report of the target script. This application combines production experience, script content, feedback and other factors, uses a large language model to understand the deep plot of the script and accurately predict the value of the script, and realizes a comprehensive analysis of the script.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a script analysis system and method based on a large language model. Background Art

[0002] The script evaluation of film and television dramas mainly relies on the experience and judgment of industry experts, which is highly subjective and inefficient. With the development of artificial intelligence technology, it has become possible to objectively analyze scripts using natural language processing (NLP) and AI technology.

[0003] At present, the relevant technologies are largely lacking in the understanding of the deep content of the script and the integration of industry expertise. On the one hand, most of the existing technologies are limited to small model modeling and analysis, which have good effects in text classification, information retrieval, and summary generation, but cannot follow the natural language understanding logic to analyze the deep plot of the script. Only summarizing the content of the movie script in a labeled manner, without following the natural language understanding logic to analyze the deep plot of the script, will result in the analysis results being not accurate and unreliable. On the other hand, the existing technologies have a single perspective on the understanding and analysis of the script, only considering the audience demand side, without combining the supply-side production expertise and objective indicators. The analysis results are not comprehensive enough and have no practical reference value.

[0004] Therefore, there is an urgent need for a method that can understand the deep-level plot of the script and intelligently provide accurate and comprehensive data support for the evaluation and analysis of the script content of film and television dramas. Summary of the invention

[0005] In order to achieve the above objectives, the present application provides a script analysis system and method based on a large language model, which can understand the deep-level plot of the script and accurately predict the value of the script, thereby realizing a comprehensive analysis of the script.

[0006] In a first aspect, the present application provides a script analysis system based on a large language model, the system comprising:

[0007] A data processing module is used to perform feature engineering, data integration and labeling on the acquired multi-category sample data to obtain a feature data set, wherein the multi-category data includes script content data, script feedback data and production experience data, and the labeled labels include at least one of the script's emotional information, plot elements and plot structure;

[0008] A model training module is used to: fine-tune the pre-trained large language model on the feature data set for a preset natural language understanding task to obtain a script analysis model, wherein the script analysis model is used to output at least one of the emotional information, plot elements and plot structure of the script;

[0009] The script content data and the production experience data in the feature data set are used as input features, the script feedback data in the feature data set is used as a prediction result label, and a given time series neural network is trained to obtain a script value prediction model;

[0010] A reporting service module is used to: respond to an analysis request for a target script, respectively call the script analysis model and the script value prediction model, analyze the script content and production data of the target script, and obtain analysis results and feedback data of the target script, wherein the analysis results include at least one of emotional information, plot elements and plot structure; and generate an analysis report of the target script based on the analysis results and feedback data of the target script.

[0011] In a possible implementation manner, the reporting service module is further configured to:

[0012] In response to a comparison and modification request for the target script, converting the target script into structured script data according to a target structure template, wherein the structured script data describes various plot elements according to the target structure template;

[0013] Based on the structured script data, providing an interactive interface for modifying various plot elements of the target script;

[0014] Update the plot elements of the target script according to the modification operation obtained from the modification interaction interface;

[0015] Based on the analysis results and feedback data corresponding to the updated target script, a modification evaluation report of the target script is generated.

[0016] In a possible implementation manner, the data processing module is used to:

[0017] Feature engineering: using natural language processing technology to classify the sentiment tendency of the script feedback data; using a given topic modeling algorithm to extract the main topics or elements from the script feedback data; semantically understanding the scenes, characters, and dialogues in the script content data to identify at least one of the plot turning points, plot elements, and plot structure; converting the production experience data into quantitative features, which at least include: amount, cycle, and quantity;

[0018] Data integration: Integrate the processed script content data, script feedback data, and production experience data into a given database or data framework in a given data format;

[0019] Data labeling: Manually label some of the script content data, script feedback data, and production experience data. The labels used for labeling include: the quality level of the script content, the emotional tendency of the feedback data, and the discussion topic.

[0020] In a possible implementation manner, the temporal neural network is any one of a multilayer perceptron, a convolutional neural network, and a recurrent neural network; and the model training module is used for:

[0021] The plot turning points, plot elements, plot structure corresponding to the script content data and the quantitative features corresponding to the production experience data are fused as input features, the emotional tendency and / or discussion topic corresponding to the script feedback data are used as prediction result labels, and the time series neural network is trained by adjusting hyperparameters and cross-validation to obtain a script value prediction model.

[0022] In a possible implementation manner, the system further includes: a data acquisition module; the data acquisition module is used to:

[0023] Acquire a plurality of script content data from a given network database, website address and / or local database, wherein the script content data includes: one or more of a character description text, a dialogue text, and a narrative text;

[0024] Acquire multiple script feedback data from a given network database, website address and / or local database according to keywords corresponding to the script, wherein the script feedback data includes one or more of market value, evaluation score, evaluation reason and number of followers;

[0025] From a given network database, website address and / or local database, production experience data is obtained according to the keywords, and the production experience data includes: plot narrative structure theory, and one or more of production budget, film type, publicity investment, production company, director, screenwriter, starring, whether it is IP adaptation, and industry evaluation.

[0026] In a possible implementation manner, the feedback data output by the script value prediction model includes: total value prediction, value change trend prediction, and key factor prediction;

[0027] The reporting service module is used to: use the script value prediction model to predict the change trend of market value and the change range of total market value within a given scenario and / or period;

[0028] Determine the output value of the script value prediction model under different input feature combinations, determine the impact sensitivity of various input features according to the output values ​​corresponding to the different input feature combinations, and determine at least one key factor according to the impact sensitivity;

[0029] The input features include: production experience data and script content data; the more significant the change in output value caused by the numerical change of the input feature, the stronger the sensitivity of the input feature.

[0030] In one possible implementation, the pre-trained large language model is: a BERT model, a GPT series model, or a Transformer architecture model pre-trained on text data; the model training module is used to:

[0031] Inputting the feature data set into the large language model;

[0032] The results of the preset sentiment analysis task, plot element recognition task and plot structure recognition task are performed according to the large language model, and compared with the preset prompt template to obtain an error, wherein the prompt template indicates the semantic association between the model input and output;

[0033] The parameters of the large language model are iteratively adjusted according to the error until the error of the large language model in executing the sentiment analysis task, the plot element recognition task and the plot structure recognition task meets the preset target large language model, thereby obtaining the script analysis model.

[0034] In a possible implementation manner, the prompt template includes at least one of the following:

[0035] A. Plot structure analysis template: script content data as input, and multiple plot structure tags as prompt words;

[0036] B. Rhythm and tension analysis template: The script content data is used as input, and the emotional intensity label, emotional change label, plot advancement label, and plot conflict evaluation label of each scene in the script are used as prompt words;

[0037] C. Character Depth and Arc Analysis Template: Extract character features and character relationships from script content data, generate a time series in chronological order, and use the time series as input; the character's personality development trajectory label is used as a prompt word;

[0038] D. Relationship dynamics analysis template: Based on the character relationship map in the script content, the character relationship map is used as input; at least one relationship label among the character interaction frequency, emotional tendency, cooperation, and conflict extracted from the character relationship map is used as a prompt word;

[0039] E. Dialogue quality analysis template: real dialogue data is used as input, and character personality tags are used as prompt words;

[0040] F. Creativity and originality analysis template: the script content data to be analyzed is used as input, and the similarity between the input content and the scripts in the given script database is used as output;

[0041] G. Problem and suggestion analysis template: script content data is used as input, and the annotated logical error labels, plot incoherence labels, and modification suggestion text are used as prompt words;

[0042] H. Market and audience positioning analysis template: script content data and production data as input, audience group labels as prompts;

[0043] I. Technical and production considerations template: script content data and production data are used as input, and production difficulty labels and budget labels are used as prompts.

[0044] In a possible implementation manner, the data processing module is further used to:

[0045] Data preprocessing: After data cleaning and word segmentation, the multi-category data is denoised and irrelevant words are deleted. Finally, the multi-category data is converted from text into a digital encoding form that can be understood by the model.

[0046] In a possible implementation manner, the reporting service module is used to:

[0047] In response to an analysis request for a target script, calling the script analysis model, analyzing the script content of the target script, and obtaining an analysis result of the target script;

[0048] Calling the script value prediction model to analyze the script content data and production data, and predicting feedback data of the target script;

[0049] Generate an analysis report of the target script based on the analysis results and feedback data of the target script.

[0050] In a possible implementation, the analysis report includes: script content analysis results, script value prediction results, and modification suggestions;

[0051] The script analysis results include at least one of: plot structure, rhythm and tension, character depth and arc, character relationship dynamics, dialogue quality, creativity and originality

[0052] The modification suggestions include: at least one of the following: logical loopholes in the script, incoherent plots, character development problems, and modification methods;

[0053] The script value prediction result includes: the total value, the value change trend and at least one key factor affecting the total value and the value change trend;

[0054] The script value prediction result also includes: target audience group and / or script production difficulty.

[0055] In a second aspect, a script analysis method based on a large language model is provided, the method comprising:

[0056] Performing feature engineering, data integration and labeling on the obtained multi-category sample data to obtain a feature data set, wherein the multi-category data includes: script content data, script feedback data and production experience data, and the labeled labels include at least one of the script's emotional information, plot elements and plot structure;

[0057] For a preset natural language understanding task, fine-tune the pre-trained large language model on the feature data set to obtain a script analysis model, wherein the script analysis model is used to output at least one of the emotional information, plot elements, and plot structure of the script;

[0058] The script content data and the production experience data in the feature data set are used as input features, the script feedback data in the feature data set is used as a prediction result label, and a given time series neural network is trained to obtain a script value prediction model;

[0059] In response to an analysis request for a target script, the script analysis model and the script value prediction model are called respectively to analyze the script content and production data of the target script to obtain analysis results and feedback data of the target script, wherein the analysis results include at least one of emotional information, plot elements and plot structure; and an analysis report of the target script is generated based on the analysis results and feedback data of the target script.

[0060] In a possible implementation, the method further includes steps related to more functions implemented by the system described in the first aspect.

[0061] In a third aspect, a computing device is provided, comprising a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement the script analysis method based on a large language model as provided in the first aspect.

[0062] In a fourth aspect, a computer-readable storage medium is provided, wherein at least one program is stored in the storage medium, and the at least one program is executed by a processor to implement the script analysis method based on a large language model as provided in the first aspect.

[0063] In a fifth aspect, a computer program product is provided, comprising a computer program / instructions, wherein when the computer program / instructions are executed by a processor, the script analysis method based on a large language model as provided in the first aspect is implemented.

[0064] The technical solution provided by this application includes at least the following technical effects:

[0065] This application widely collects relevant data on production experience, script content, feedback evaluation and other factors as training data sets, and fine-tunes the basic large model in combination with theoretical works on film script writing. It innovatively produces a film script analysis method that can automatically disassemble the script content, comprehensively evaluate multiple influencing factors, and internalize film production expertise. It uses a large language model to understand the deep-level plot of the script and accurately predict the value of the script, thereby achieving a comprehensive analysis of the script. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a schematic diagram of a script analysis system based on a large language model provided in an embodiment of the present application;

[0067] Figure 2 It is a schematic diagram of a script modification assistant provided in an embodiment of the present application;

[0068] Figure 3 It is a functional schematic diagram of a script analysis system based on a large language model provided in an embodiment of the present application;

[0069] Figure 4 It is a functional schematic diagram of another script analysis system based on a large language model provided in an embodiment of the present application;

[0070] Figure 5 It is a flowchart of a script analysis method based on a large language model provided in an embodiment of the present application;

[0071] Figure 6 It is a schematic diagram of the hardware structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] To further illustrate each embodiment, the present application provides drawings. These drawings are part of the disclosure of the present application, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, a person of ordinary skill in the art should be able to understand other possible implementations and the advantages of the present application. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components. The term "at least one" in the present application means one or more, and the term "multiple" in the present application means two or more.

[0073] In response to the defects in the existing technology, this application aims to build a script analysis technology for film and television works that can automatically disassemble the script content, comprehensively evaluate multiple influencing factors, and internalize the professional knowledge of film and television production, so as to provide accurate and comprehensive evaluation reports for film and television production, and guide script optimization and film and television production strategies.

[0074] The present application is now further described in conjunction with the accompanying drawings and specific implementation methods.

[0075] An embodiment of the present application provides a script analysis system based on a large language model. Figure 1 is a schematic diagram of a script analysis system based on a large language model provided in an embodiment of the present application, such as Figure 1 As shown, the system at least includes a data processing module, a model training module and a report service module. Figure 1 The technical solution of the embodiment of the present application is introduced.

[0076] (1) Data processing module

[0077] In the embodiment of the present application, the data processing module is used to perform feature engineering, data integration and annotation on the acquired multi-class sample data to obtain a feature data set. Optionally, the script can be a movie script, a TV drama script, a drama script or any other text that describes the story plot in natural language, and the present application does not make specific restrictions on this.

[0078] The multiple types of data include: script content data, script feedback data and production experience data. The annotated tags include at least one of the script's emotional information, plot elements and plot structure.

[0079] For example, taking movies as an example, script content data refers to movie script data. Script content data is used to provide rich learning materials for the large language model, so that the model can understand and master the structure, narrative method, character construction, dialogue style, plot development and other plot elements of the movie script through deep learning. The script content data contains rich dialogues and narrative texts, which helps the model learn the contextual meaning of natural language and improve its ability to understand complex language structures and emotional expressions.

[0080] In some possible implementations, the script content data includes the text content of the script, usually including character dialogues, scene descriptions, action instructions, etc. The script content data is crucial for understanding the theme, emotion, character relationships, etc. of the script.

[0081] The following is an example of script content data:

[0082] "Script title: "River of Time"

[0083] Scene 1: Ancient library

[0084] Roles: Zhang San (young scholar), Li Si (librarian)

[0085] Zhang San (flipping through an ancient book): "This book records a long-lost magic..."

[0086] Li Si (smiles): "Yes, that is the wisdom of our ancestors. But please use it with caution."

[0087] ...(More dialogue and scene description)"

[0088] Exemplarily, taking movies as an example, script feedback data includes movie review information. For any released movie, its movie review information can reflect the audience's preference for movies from a certain level. For example, movie review data may include: film type, production company, director, screenwriter, starring, whether it is an IP adaptation, release time, number of people who want to watch, reasons for good reviews, and other rich and specific information. Script feedback data is a true reflection of the value of historical scripts in the actual market, and can be used to train script value prediction models so that the script value prediction models can learn the inherent natural laws between script content and work production strategies and market value.

[0089] In some possible implementations, the script feedback data usually comes from audiences, readers or reviewers, and includes comments, suggestions or feelings about the script. Such data is very important for understanding the popularity of the script, possible problems and improvement directions.

[0090] Here is an example of script feedback data:

[0091] "Script name: "River of Time"

[0092] Audience feedback 1: The plot is compact, the characters are vivid, and Zhang San's growth process is especially touching.

[0093] Audience Feedback 2: Although the magic element is novel, it feels a bit abrupt and does not blend naturally with the overall plot.

[0094] Professional review feedback: The script structure is complete, but some dialogues are a bit stiff, and it is recommended to be improved. "

[0095] For example, taking movies as an example, production experience data includes professional knowledge in the field of movie production, such as the review and summary of highly acclaimed and high-grossing movies (including text and / or numerical values), as well as the communication content with well-known directors and screenwriters (including text form). Production experience data is used to provide the supply-side control of movie quality to model training in text form.

[0096] In some embodiments, the production experience data usually comes from all parties involved in the script production, including directors, screenwriters, actors, etc. Based on the professional knowledge and practical experience in the industry, it is helpful to introduce suggestions and improvements in the adaptation, shooting, performance, etc. of the script into the script analysis process. The production experience data has important reference value for optimizing the script content and improving the production quality.

[0097] Here is an example of making empirical data:

[0098] "Script name: "River of Time"

[0099] Director's Experience: The script is suitable for shooting with nostalgic tones and a slow pace to highlight the feeling of the passage of time.

[0100] Screenwriter's Notes: During the creative process, we borrowed elements from ancient magic stories, but we needed to be especially careful when integrating them into the modern context so as not to appear out of place.

[0101] Actor's suggestion: The character Zhang San's growth process needs more inner drama to show his emotional changes. "

[0102] In other embodiments, the production experience data also includes theoretical works on script writing, such as Sydfield's three-act structure, McKee's narrative principles, the cat rescue method, sequence theory, classic Hollywood narrative model, visual scripts, etc.

[0103] Specifically, the Sydfield Three-Act Structure divides the story into three main parts: the beginning, the middle, and the end. The beginning is used to introduce characters, set the background, and propose the theme and conflict of the story; the middle drives the story to a climax through a series of events and the development of conflicts; the ending resolves the conflict, reveals the theme, and brings emotional satisfaction to the audience. McKee's narrative principle requires that a good story should have a clear theme, rich characterization, a tight plot structure, and deep emotional expression. The "Save the Cat" method is a simple screenwriting technique designed to help screenwriters create eye-catching moments in the story. Sequence theory is a theory about story structure that emphasizes dividing the story into a series of closely connected sequences. Each sequence has a clear goal and conflict, and the story is driven forward by resolving the conflict.

[0104] The above-mentioned theoretical works on script writing can provide a theoretical framework and reference standards for script creation. Based on these theoretical frameworks, the big language model can evaluate the structure, plot development, character creation, etc. of the script, so that the big language model can have a deeper understanding of the internal logic, theme and intention of the script. Through these industry-specific production theoretical knowledge, the model can be guided to evaluate the script from a professional perspective, rather than just staying on the surface of text analysis. Furthermore, as a support, the script writing theory can assist the big language model to more accurately judge the quality of the script and provide more targeted suggestions. This can greatly improve the accuracy and reliability of the evaluation. In addition, the various principles, techniques and methods mentioned in the theoretical works on script writing can provide a wealth of evaluation dimensions and indicators for the big language model. For example, the structural integrity of the script can be evaluated based on the Sydfield three-act structure, and the narrative ability of the script can be evaluated based on the McKee narrative principle.

[0105] By inputting theoretical works on scriptwriting, the large language model can learn the general rules and principles of scriptwriting. This allows the model to maintain a high degree of versatility and transferability when evaluating scripts of different styles and themes.

[0106] In the embodiment of the present application, the data processing module provides a high-quality, structured data foundation for subsequent model training and analysis, so that the film script analysis system can more accurately evaluate the quality of the script, predict audience reactions, provide creative guidance, etc.

[0107] In one possible implementation, the steps performed by the data processing module include: feature engineering, data integration, and data labeling.

[0108] Exemplarily, feature engineering specifically includes:

[0109] S111. Use natural language processing technology to classify the sentiment tendency of the script feedback data; use a given topic modeling algorithm to extract the main topics or elements from the script feedback data.

[0110] Among them, sentiment analysis refers to the use of natural language processing technology to analyze emotional tendencies, such as positive, negative or neutral.

[0111] S112. Perform semantic understanding on scenes, characters, and dialogues in the script content data, and identify at least one of plot turning points, plot elements, and plot structure.

[0112] Specifically, topic modeling can apply topic models such as LDA to extract the main topics or elements of discussion from a large number of comments; the third is script structure analysis, parsing the scenes, characters, and dialogues in the script to identify key plot turning points.

[0113] S113. Convert the production experience data into quantitative features, which at least include: amount, period, and quantity.

[0114] Specifically, experience knowledge encoding can be used to transform production experience data into quantifiable features, such as budget (corresponding amount), shooting cycle, post-production difficulties, etc.

[0115] In one possible implementation, experiential knowledge coding is used to express abstract industry experiential knowledge in a standard form. Exemplarily, language coding, image coding, and the like can be used to achieve standardization of abstract production experience data. Language coding, for example, is to systematize and structure experiential knowledge through text and language description. Graphic coding, for example, is to use graphics, symbols, etc. to concisely express complex experiential knowledge, thereby enhancing readability and comprehensibility. When encoding experiential knowledge, tools such as knowledge bases and knowledge maps can be used to assist the encoding process, thereby converting production experience data into standardized data that can be processed by the model.

[0116] S12, data integration. Specifically includes: integrating the processed script content data, script feedback data, and production experience data into a given database or data framework according to a given data format.

[0117] Exemplarily, this step S12 includes the following steps:

[0118] S121. For script content data, script feedback data, and production experience data, determine the table structure of each type of data set in the database or data framework according to a given data format.

[0119] For example, for script content data, fields such as "script ID", "title", "author", "content", etc. may be required. For script feedback data, fields such as "feedback ID", "script ID", "user ID", "emotional tendency", "feedback content", etc. may be required. For production experience data, fields such as "project ID", "script ID", "production cycle", "cost", "producer", etc. may be required.

[0120] S122. Map the script content data, script feedback data, and production experience data to the fields of the corresponding database or data framework.

[0121] As needed, the script content data, script feedback data, and production experience data may need to be further converted, such as converting the date to a specific format, converting the text encoding to UTF-8, etc. Then, a database management tool or programming language (such as SQL, Python, etc.) is used to write a script to import the processed script content data, script feedback data, and production experience data into a given database or data framework.

[0122] Optionally, after the import is complete, you can use query statements or other tools to check whether the data has been imported correctly and whether the data type and format meet the requirements to ensure the accuracy and completeness of the data.

[0123] It should be noted that the user-related data involved in this application is obtained with the full authorization of the user, and the feedback data containing user personal information or sensitive information is protected with the full authorization of the user, in compliance with data protection regulations, to ensure the legality and compliance of the data. Depending on the specific database or data framework requirements, it may be necessary to use a specific data format or encoding method. This application ensures that the requirements of data protection regulations are followed when processing relevant data.

[0124] S13, data labeling. Specifically, it includes: manually labeling part of the script content data, script feedback data, and production experience data. The labels used for labeling include: the quality level of the script content, the emotional tendency of the feedback data, and the discussion topic.

[0125] Manually annotate some of the script content data, script feedback data, and production experience data to add meaningful labels to these data for subsequent analysis, screening, and search

[0126] For example, in the manual annotation of script content data, the quality level of the script content may be labeled as "excellent", "good", "average", "poor", etc., or a more detailed rating system may be used, such as 1-5 stars or 1-10 points. The quality level of the script content is used to evaluate the quality of the script in terms of creativity, plot, character creation, dialogue, etc.

[0127] The annotation process can be carried out by professional script reviewers or screenwriter teams who read and analyze the scripts and give corresponding labels or scores based on quality assessment criteria.

[0128] For example, in the manual annotation of script feedback data, the emotional tendency labels of the feedback data are, for example, "positive", "negative", "neutral", etc., and can also be further subdivided, such as "very like", "relatively satisfied", "somewhat disappointed", etc. The emotional tendency of the feedback data is used to understand the audience's emotional response to the script, such as like, hate, neutral, etc. The annotation process can be carried out by professional data analysts or sentiment analysis experts who read and understand the feedback text, judge its emotional tendency according to the text content, and give corresponding labels.

[0129] For example, the labels of discussion topics in script feedback data are: extract the main topics or keywords based on the feedback content, such as "plot", "character", "special effects", "soundtrack", etc. Discussion topics are used to understand the main topics or concerns discussed by the audience in the feedback. The annotation process extracts the text of each feedback, determines the main topics or keywords, and assigns corresponding labels to them. If the feedback content involves multiple topics, multiple labels can be assigned to it.

[0130] For example, in the manual labeling of production experience data, since production experience data usually contains relatively structured information (such as production cycle, cost, staffing, etc.), structured data is usually used directly without complex manual labeling. If there are specific analysis needs, such as evaluating production efficiency, cost-effectiveness, etc., the relevant data can be labeled or classified.

[0131] Optionally, the steps performed by the data processing module also include: data preprocessing. Specifically, it includes: data cleaning and word segmentation of multiple types of data, then denoising and irrelevant word deletion, and finally converting the multiple types of data from text into a digital encoding form that can be understood by the model. Data preprocessing is performed before performing feature engineering.

[0132] Exemplarily, data cleaning includes steps such as removing irrelevant characters, punctuation marks, stop words, and correcting spelling errors to standardize the text format. Word segmentation includes dividing the text into words or phrases, and then restoring the words to facilitate subsequent analysis; denoising includes identifying and removing spam comments, advertising information, and meaningless feedback from the text; text digital encoding, for example, uses the bag-of-words model, TF-IDF, word embedding, etc. to convert the data in multiple text forms.

[0133] (2) Model training module

[0134] (2-1) Script Analysis Model

[0135] In an embodiment of the present application, the model training module is used to: for a preset natural language understanding task, fine-tune a pre-trained large language model on a feature data set to obtain a script analysis model, and the script analysis model is used to output at least one of the emotional information, plot elements and plot structure of the script.

[0136] In one possible implementation, the pre-trained large language model is: a BERT model, a GPT series model, or a Transformer architecture model that has been pre-trained on text data. Such models have been pre-trained on large-scale text data and have good language understanding capabilities.

[0137] In one possible implementation, the process of training the large language model by the model training module includes:

[0138] Input the feature data set into the large language model; perform the preset sentiment analysis task, plot element recognition task, and plot structure recognition task based on the large language model, and compare the results with the preset prompt template to obtain the error, which indicates the semantic association between the model input and output. Iterate and adjust the parameters of the large language model according to the error until the error of the large language model performing the sentiment analysis task, plot element recognition task, and plot structure recognition task meets the preset target of the large language model, and obtain the script analysis model.

[0139] When fine-tuning the large language model in this application, it helps the model learn knowledge in specific fields on specific tasks, such as identifying laughs, tears, turning points, etc. in the script. While widely introducing movie feedback data, accurately introduce data on film production. Specifically, script feedback information collected on the Internet (such as film reviews), classic scripts (such as Farewell My Concubine, The Shawshank Redemption, Crouching Tiger, Hidden Dragon, Forrest Gump, etc.) and theoretical works on film script writing (such as Sydfield's three-act structure, McGee's narrative principles, cat rescue methods, sequence theory, classic Hollywood narrative models, visual scripts, etc.) are used to fine-tune the basic large model, so that it can not only grasp the preferences of the movie market, but also have professional knowledge of movies.

[0140] In one possible implementation, the prompt template includes at least one of the following AIs.

[0141] A. Plot structure analysis template: script content data is used as input, and multiple plot structure tags are used as prompt words.

[0142] Specifically, a script dataset with plot structure labels is used to construct a prompt template. Fine-tuning the large model based on the prompt template can enable the model to have sequence labeling capabilities, so that it can more accurately identify the different parts of the script (beginning, development, turning point, climax, and ending), and then evaluate whether it follows classic narrative patterns such as the three-act structure.

[0143] B. Rhythm and tension analysis template: The script content data is used as input, and the emotional intensity labels, emotional change labels, plot advancement labels, and plot conflict evaluation labels of each scene in the script are used as prompt words.

[0144] Specifically, the emotional intensity and changes of each scene in the script are annotated as an indicator of tension, followed by annotating key events and their contribution to the advancement of the plot, as well as the critics' or audiences' evaluation of the rationality and attractiveness of the script's conflict settings. This enables the model to have the ability to identify the overall rhythm.

[0145] C. Character Depth and Arc Analysis Template: The character features and relationships extracted from the script content data are used to generate a time series in chronological order, with the time series as input; the character's personality development trajectory label is used as a prompt word.

[0146] Specifically, characters and relationships are extracted from the movie script, and information such as roles, personality traits, and motivations are organized. Then, time series are used to annotate and track the changes of characters as the script progresses, and their character development trajectory (character arc) is analyzed. Fine-tuning the above prompt template can enable the model to have the ability to characterize and identify psychological changes.

[0147] D. Relationship dynamics analysis template: Based on the character relationship map in the script content, the character relationship map is used as input; at least one relationship label among the character interaction frequency, emotional tendency, cooperation, and conflict extracted from the character relationship map is used as a prompt word.

[0148] Specifically, by constructing a character relationship map, we organize information such as character interaction frequency, emotional tendencies, cooperation, and conflict. Then, we use time series to annotate and track the changes in character relationships as the script progresses, and obtain a prompt template. Training based on this dimensional template enables the model to have the ability to evaluate the dynamic changes in character relationships and their role in promoting the story.

[0149] E. Dialogue quality analysis template: real dialogue data is used as input, and character personality tags are used as prompt words.

[0150] Specifically, by comparing with a large amount of real dialogue data, the model's ability to assess the authenticity of the dialogue can be improved. Through semantic similarity analysis and personality consistency detection, it can be ensured that the dialogue is consistent with the character's personality and promotes the plot or deepens the character.

[0151] F. Creativity and originality analysis template: The script content data to be analyzed is taken as input, and the similarity between the input content and the scripts in the given script database is taken as output.

[0152] Specifically, the originality of the script is evaluated by comparing it with the scripts in the existing database. Secondly, the core creative points in the script can be identified using the movie theme, and the degree of innovation of these ideas can be evaluated through a novelty scoring system.

[0153] Optionally, the novelty scoring system is, for example, an evaluation based on user feedback, which collects users' subjective evaluations of the novelty of the content recommended by the system through user questionnaires, scoring, etc., thereby directly reflecting user experience and satisfaction. The novelty scoring system is, for example, an evaluation based on algorithms: using the above evaluation indicators, combined with machine learning, data mining and other technologies, the novelty of the content generated by the system is quantitatively evaluated to objectively and quickly evaluate the degree of innovation.

[0154] G. Problem and suggestion analysis template: script content data is used as input, and annotated logical error labels, plot incoherence labels, and modification suggestion text are used as prompt words.

[0155] Specifically, by marking logical errors, plot inconsistencies and other problems raised by film critics or audiences, it can point out negative cases and modification suggestions from script writing books from the audience's perspective. Through these two aspects of data, the model can be fine-tuned to enable the model to have the ability to identify script problems and make suggestions.

[0156] H. Market and audience positioning analysis template: script content data and production data are used as input, and audience group labels are used as prompts.

[0157] Specifically, the potential target audience group is determined based on the script content (such as subject matter, characters, and plot). Then, audience data (such as social media and movie reviews) is used to further refine the target group's age, gender, interests, and other characteristics. Fine-tuning the large language model based on the prompt template of this dimension can enable the model to have the ability to profile the audience.

[0158] I. Technical and production considerations template: script content data and production data are used as input, and production difficulty labels and budget labels are used as prompts.

[0159] Specifically, the prompt template of this dimension is constructed by sorting out the difficulty of film production, budget requirements, etc. and using label classification. Production difficulty, such as special effect complexity, shooting location, costumes and props, can quantify the difficulty corresponding to different indicators. For example, special effect complexity corresponds to 1-5 level labels, the higher the level, the more complex; the more shooting locations and the farther the distance between the shooting locations, the higher the difficulty.

[0160] For example, by constructing a reasonable prompt template (Prompt), the large language model can output scientific evaluation conclusions from multiple key dimensions, thereby providing a basis for subsequent revision and improvement. For details of the prompt construction, please refer to Table 1.

[0161] Table 1

[0162]

[0163]

[0164] The model parameters of the large language model technology have reached a scale of hundreds of billions of parameters, which supports learning a large amount of historical data and can understand hundreds of pages of plot content. In addition, the large language model has strong reasoning ability and can simulate the logic of human thinking for barrier-free communication and dialogue, and deeply understand human thinking. By providing accurate prompt information, the model can be guided to understand and analyze the plot. Through multiple rounds of dialogue, the understanding ability can be corrected and gradually enhanced. It is expected to bring a significant improvement to the intelligent analysis of movie scripts.

[0165] (2-2) Script Value Prediction Model

[0166] In the embodiment of the present application, the model training module is used to: use the script content data and production experience data in the feature data set as input features, use the script feedback data in the feature data set as the prediction result label, train a given time series neural network, and obtain a script value prediction model;

[0167] In one possible implementation, the temporal neural network is any one of a multilayer perceptron (MLP), a convolutional neural network (CNN), and a recurrent neural network (RNN).

[0168] In one possible implementation, the model training module is used to:

[0169] The plot turning points, plot elements, plot structure corresponding to the script content data and the quantitative features corresponding to the production experience data are fused as input features. The emotional tendency and / or discussion topic corresponding to the script feedback data are used as prediction result labels. The time series neural network is trained by adjusting hyperparameters and cross-validation to obtain a script value prediction model.

[0170] For example, the laughs, tears, turning points, etc. extracted from the script content data, together with the quantitative features extracted from the production experience data (such as budget, director influence, cast, release date, etc.), constitute the input features. Overfitting can be prevented by adjusting hyperparameters, cross-validation and other techniques to ensure the generalization ability of the model.

[0171] In one possible implementation, the output of the script value prediction model includes: total value prediction, value change trend prediction, and key factor prediction. Value refers to market value; taking a movie as an example, its market value can be measured by one or more of box office, attendance, and topicality; taking a drama as an example, its market value can be measured by one or more of the number of performances, attendance, and the number of touring cities.

[0172] The key factors are selected from the input features of the script value prediction model, including production experience data and script content data. For the script value prediction model, the more significant the change in the output value caused by the numerical change of its input feature, the more critical the input feature is, and it can be considered as a key factor.

[0173] Taking the box office of a movie as an example, the box office prediction result is not just a simple number, it contains multiple dimensions and trends, aiming to provide a scientific basis for the success of the movie project. Through the script value prediction model provided by this application, the following contents can be predicted:

[0174] 1. Box office total prediction. The script value prediction model predicts the total box office revenue of a movie in a specific market or globally based on the input features. Specifically, it outputs possible interval values, for example, the expected box office is between XX billion and YY billion.

[0175] 2. Box office trend forecast. Specifically, predict the expected box office trend after the movie is released, including the first weekend box office, first week box office, total box office distribution, etc. This indicator helps to understand the market acceptance speed and durability of the movie.

[0176] 3. Sensitivity analysis. Using the control variable method, analyze the impact of key variables in the input features (such as budget, publicity investment, and changes in actor influence) on box office forecasts. Specifically, by gradually changing the value of a single input feature (variable), analyze the changes in the output value (total box office and / or box office change trend) to help decision makers understand the sensitivity of changes in various factors to the final result.

[0177] (3) Reporting service module

[0178] In the embodiment of the present application, the reporting service module is used to: respond to the analysis request for the target script, respectively call the script analysis model and the script value prediction model, analyze the script content and production data of the target script, and obtain the analysis result and feedback data of the target script, and the analysis result includes: at least one of emotional information, plot elements and plot structure; generate an analysis report of the target script according to the analysis result and feedback data of the target script. The functions of the reporting service module are introduced from several aspects below.

[0179] (3-1) Plot Analysis

[0180] The reporting service module is used to: respond to an analysis request for a target script, call the script analysis model, analyze the script content of the target script, and obtain the analysis result of the target script. In one possible implementation, the analysis report includes: script content analysis results, script value prediction results, and modification suggestions. Specifically, the script analysis results include: plot structure, rhythm tension, character depth and arc, character relationship dynamics, dialogue quality, creativity and originality.

[0181] (3-2) Value Analysis

[0182] Among them, the reporting service module is used to: call the script value prediction model to analyze the script content data and production data, and predict the feedback data of the target script.

[0183] In one possible implementation, the feedback data output by the script value prediction model includes: total value prediction, value change trend prediction, and key factor prediction.

[0184] Based on the above example, the reporting service module is used to:

[0185] Using the script value prediction model, predict the changing trend of the market value and the changing range of the total market value within a given scenario and / or period; determine the output value of the script value prediction model under different input feature combinations, determine the impact sensitivity of various input features according to the output values ​​corresponding to different input feature combinations, and determine at least one key factor according to the impact sensitivity;

[0186] Among them, the input features include: production experience data and script content data; the more significant the change in output value caused by the numerical change of the input feature, the stronger the sensitivity of the impact of the input feature.

[0187] (3-3) Report Integration

[0188] Among them, the reporting service module is used to generate an analysis report of the target script based on the analysis results and feedback data of the target script.

[0189] In one possible implementation, the analysis report includes: script content analysis results, script value prediction results, and modification suggestions. Specifically, the script analysis results include: plot structure, rhythm tension, character depth and arc, character relationship dynamics, dialogue quality, creativity and originality. Modification suggestions include: script logic loopholes, incoherent plots, character development problems, and modification methods. The script value prediction results include: total value, value change trend, and at least one key factor affecting the total value and value change trend.

[0190] In a possible implementation, the output of the reporting service module supports the following formats:

[0191] 1. Report format, which may include charts, text analysis results, specific suggestions, etc.;

[0192] 2. Display in the form of a chart on a display device (e.g., a dashboard);

[0193] 3. Display directly on the software interface. Optionally, the results can be output in the form of highlighted annotations and annotation suggestions in the target script being analyzed.

[0194] The above output methods can facilitate screenwriters, producers or other users to intuitively understand the analysis results, and optimize the script accordingly to improve the market potential and artistic value of the work.

[0195] Optionally, the script value prediction result also includes: target audience group and / or script production difficulty.

[0196] (3-4) Intelligent comparison modification

[0197] In a possible implementation, the reporting service module further supports a comparison and modification function. Specifically, the process of the reporting service module implementing the comparison and modification function includes:

[0198] In response to a comparison modification request for a target script, the target script is converted into structured script data according to a target structure template, and the structured script data describes various plot elements according to the target structure template. Based on the structured script data, a modification interaction interface for various plot elements of the target script is provided; the plot elements of the target script are updated according to the modification operation obtained from the modification interaction interface; and a modification evaluation report of the target script is generated according to the analysis results and feedback data corresponding to the updated target script.

[0199] Exemplarily, for the comparison modification request of the target script, a famous theory or a successful movie is selected as the target structure template for comparison to intelligently generate modification suggestions. Taking the target structure target from Blake Snyder's "Save the Cat: The Secret of Film Script Writing" as an example, the script is compared and analyzed with the famous film theory, and the template script is split into a structured script template proposed in "Save the Cat: The Secret of Film Script Writing", including 15 "beats" or key scenes. Secondly, each of the 15 "beats" or key scenes has a specific function; based on this, specific script adjustment suggestions are specified for the plot elements (such as plot design, character shaping, climax setting, etc.) of the target script under the structural template, and customized modification plans are output. Exemplarily, the script adjustment suggestions are, for example: strengthening the arc of a certain character, adjusting the plot to enhance the conflict, etc. By modifying the interactive interface, users can be allowed to try different modification plans directly in the system and view the analysis feedback in real time.

[0200] In one possible implementation, the above-mentioned comparison and modification function can be realized as an auxiliary tool, or an intelligent modification assistant, which aims to provide a quick tool for script writing, not only to improve the technical quality of the script, but also to promote the deepening and optimization of creativity, and ultimately help create more fascinating film scripts. Figure 2 is a schematic diagram of a script modification assistant provided in an embodiment of the present application. The system provided in an embodiment of the present application can be implemented as an AI tool for film and television scripts, which outputs various modification suggestions based on the basic information to be analyzed, such as Figure 2 As shown, the modification suggestions include: 1. Problems with the script and how to improve it 2. Relevant information and box office reports of similar scripts / projects 3. Production suggestions beyond the script content, feedback predictions for different production methods, etc.

[0201] This application provides a targeted comparison and modification function, which can make optimization suggestions for the shortcomings of the script based on standard structures such as script writing theory. For example, if the model finds that there are problems with the plot development of the script, it can provide suggestions for plot reconstruction based on theories such as the classic Hollywood narrative model.

[0202] Figure 3is a functional schematic diagram of a script analysis system based on a large language model provided in an embodiment of the present application, such as Figure 3 As shown, the data processing module extracts samples of big data related to high-quality scripts and box office, performs feature statistics on the related box office, and obtains script content data and script feedback data. Combined with other feature factors other than script content, that is, the source of production experience data (cast team, etc.), all are input into the model training module for feature learning and modeling. The report service module is responsible for taking the script to be analyzed and the production data of the script to be analyzed (funds, team, etc.) as the input of the model; the model (including the script analysis model and the script value prediction model) can output an analysis report, which can guide the modification of the script.

[0203] In a possible implementation, the system provided in the embodiment of the present application also includes: a data acquisition module.

[0204] (4) Data acquisition module

[0205] In one possible implementation, the data acquisition module is used to:

[0206] Acquire a plurality of script content data from a given network database, website address and / or local database, wherein the script content data includes: one or more of character description text, dialogue text and narrative text;

[0207] Obtaining a variety of script feedback data from a given network database, website address and / or local database according to keywords corresponding to the script, the script feedback data including: one or more of market value, evaluation score, evaluation reason, and number of followers;

[0208] From a given network database, website address and / or local database, production experience data is obtained according to keywords. The production experience data includes: plot narrative structure theory, and one or more of production budget, film type, publicity investment, production company, director, screenwriter, starring, whether it is IP adaptation, and industry evaluation.

[0209] Figure 4 is a functional schematic diagram of another script analysis system based on a large language model provided in an embodiment of the present application, such as Figure 4 As shown in the figure, in this system, the algorithm model used by the model training module includes a large language model for in-depth interpretation of the script and a neural network model for predicting the value of the script; the data processing steps performed by the data processing module include entity extraction, feature recognition and data labeling; the data collected by the data collection module includes film review information, film scripts and film production experience. The functions provided by the system's report service module include script analysis, box office prediction and modification assistant.

[0210] This application widely collects relevant data on production experience, script content, feedback and evaluation as training data sets, and fine-tunes the basic large model in combination with theoretical works on film script writing. It introduces the relevance of film and television scripts as an important factor in box office prediction, and automatically disassembles the script through a large language model, extracting features such as plot summary, turning points, laugh points, and tear points as the basis for prediction, and conducts box office analysis based on fundamental logic. The innovative concept optimizes the prediction model and improves the prediction accuracy.

[0211] Furthermore, this application uses production experience data such as theoretical works on script writing as one of the training samples of the big language model, which can provide theoretical support for the big language model when evaluating scripts, enhance deep understanding, improve evaluation accuracy and reliability, enrich evaluation dimensions and indicators, assist in generating optimization suggestions, enhance versatility and transferability, and promote self-learning and evolution.

[0212] An embodiment of the present application provides a script analysis method based on a large language model. Figure 5 This is a flowchart of a script analysis method based on a large language model provided in an embodiment of the present application, refer to Figure 5 The technical solution provided by this application includes the following steps S1 to S4. The implementation principles and specific implementation methods of the following steps S1 to S4 can all refer to the above introduction to the script analysis system based on the large language model. The following is mainly used to illustrate the possible sequence between the steps, and the principles are not repeated.

[0213] S1. Perform feature engineering, data integration and labeling on the acquired multi-category sample data to obtain a feature data set.

[0214] Among them, the multiple categories of data include: script content data, script feedback data and production experience data, and the annotated labels include at least one of the script's emotional information, plot elements and plot structure.

[0215] S11. Feature engineering.

[0216] S111. Use natural language processing technology to classify the sentiment tendency of the script feedback data; use a given topic modeling algorithm to extract the main topics or elements from the script feedback data.

[0217] S112. Perform semantic understanding on scenes, characters, and dialogues in the script content data, and identify at least one of plot turning points, plot elements, and plot structure.

[0218] S113, converting the production experience data into quantitative features, where the quantitative features at least include: amount, period, and quantity;

[0219] S12, data integration: Integrate the processed script content data, script feedback data, and production experience data into a given database or data framework according to a given data format;

[0220] S13. Data labeling: Manually label some of the script content data, script feedback data, and production experience data. The labels used for labeling include: the quality level of the script content, the emotional tendency of the feedback data, and the discussion topic.

[0221] In some possible implementations, before performing S111 feature engineering, data preprocessing is performed: data cleaning and word segmentation are performed on the multi-category data, followed by denoising and irrelevant word deletion, and finally the multi-category data is converted from text into a digital encoding form that can be understood by the model.

[0222] S2. For the preset natural language understanding task, fine-tune the pre-trained large language model on the feature dataset to obtain the script analysis model.

[0223] The script analysis model is used to output at least one of the script's emotional information, plot elements, and plot structure.

[0224] In one possible implementation, the pre-trained large language model is: a BERT model, a GPT series model, or a Transformer architecture model that has been pre-trained on text data.

[0225] In a possible implementation manner, this step S2 includes:

[0226] S21, inputting the feature data set into the large language model;

[0227] S22, performing preset sentiment analysis tasks, plot element recognition tasks, and plot structure recognition tasks according to the large language model, and comparing the results with preset prompt templates to obtain errors, where the prompt template indicates the semantic association between the model input and output;

[0228] S23. Iteratively adjust the parameters of the large language model according to the error until the error of the large language model in performing the sentiment analysis task, the plot element recognition task, and the plot structure recognition task meets the preset target of the large language model, thereby obtaining a script analysis model.

[0229] In one possible implementation, the prompt template includes at least one of the following:

[0230] A. Plot structure analysis template: script content data as input, and multiple plot structure tags as prompt words;

[0231] B. Rhythm and tension analysis template: The script content data is used as input, and the emotional intensity label, emotional change label, plot advancement label, and plot conflict evaluation label of each scene in the script are used as prompt words;

[0232] C. Character Depth and Arc Analysis Template: The character characteristics and character relationships extracted from the script content data are used to generate a time series in chronological order, with the time series as input; the character's personality development trajectory label is used as a prompt word;

[0233] D. Relationship dynamics analysis template: Based on the character relationship map in the script content, the character relationship map is used as input; at least one relationship label among the character interaction frequency, emotional tendency, cooperation, and conflict extracted from the character relationship map is used as a prompt word;

[0234] E. Dialogue quality analysis template: real dialogue data is used as input, and character personality tags are used as prompt words;

[0235] F. Creativity and originality analysis template: the script content data to be analyzed is used as input, and the similarity between the input content and the scripts in the given script database is used as output;

[0236] G. Problem and suggestion analysis template: script content data is used as input, and the annotated logical error labels, plot incoherence labels, and modification suggestion text are used as prompt words;

[0237] H. Market and audience positioning analysis template: script content data and production data as input, audience group labels as prompts;

[0238] I. Technical and production considerations template: script content data and production data are used as input, and production difficulty labels and budget labels are used as prompts.

[0239] S3. Take the script content data and production experience data in the feature data set as input features, take the script feedback data in the feature data set as the prediction result label, train the given time series neural network, and obtain the script value prediction model.

[0240] In one possible implementation, the temporal neural network is any one of a multilayer perceptron, a convolutional neural network, and a recurrent neural network.

[0241] In one possible implementation, this step S3 includes: fusing the plot turning points, plot elements, plot structure corresponding to the script content data and the quantitative features corresponding to the production experience data as input features, using the emotional tendencies and / or discussion topics corresponding to the script feedback data as prediction result labels, and training the time series neural network by adjusting hyperparameters and cross-validation to obtain a script value prediction model.

[0242] In one possible implementation, the output of the script value prediction model includes: total value prediction, value change trend prediction, and key factor prediction.

[0243] S4. In response to the analysis request for the target script, the script analysis model and the script value prediction model are called respectively to analyze the script content and production data of the target script to obtain the analysis results and feedback data of the target script; based on the analysis results and feedback data of the target script, an analysis report of the target script is generated.

[0244] The analysis results include at least one of emotional information, plot elements, and plot structure;

[0245] In one possible implementation, a script value prediction model is used to predict the changing trend of market value and the changing range of total market value within a given scenario and / or period;

[0246] Determine the output value of the script value prediction model under different input feature combinations, determine the impact sensitivity of various input features according to the output values ​​corresponding to the different input feature combinations, and determine at least one key factor according to the impact sensitivity;

[0247] Among them, the input features include: production experience data and script content data; the more significant the change in output value caused by the numerical change of the input feature, the stronger the sensitivity of the impact of the input feature.

[0248] In a possible implementation, a comparison modification function is also supported. Specifically, in response to a comparison modification request for a target script, the target script is converted into structured script data according to a target structure template, and the structured script data describes various plot elements according to the target structure template; based on the structured script data, a modification interaction interface for various plot elements of the target script is provided; the plot elements of the target script are updated according to the modification operation obtained from the modification interaction interface; and a modification evaluation report of the target script is generated according to the analysis results and feedback data corresponding to the updated target script.

[0249] In one possible implementation, the process of collecting script content data, script feedback data, and production experience data includes: obtaining a variety of script content data from a given network database, website address, and / or local database, the script content data including: one or more of character description text, dialogue text, and narrative text. From a given network database, website address, and / or local database, according to the keywords corresponding to the script, a variety of script feedback data are obtained, the script feedback data including: one or more of market value, evaluation score, evaluation reason, and number of followers. From a given network database, website address, and / or local database, according to keywords, production experience data are obtained, the production experience data including: plot narrative structure theory, as well as one or more of production budget, film type, publicity investment, production company, director, screenwriter, starring, whether it is IP adaptation, and industry evaluation.

[0250] This application widely collects relevant data on production experience, script content, feedback evaluation and other factors as training data sets, and fine-tunes the basic large model in combination with theoretical works on film script writing. It innovatively produces a film script analysis method that can automatically disassemble the script content, comprehensively evaluate multiple influencing factors, and internalize film production expertise. It uses a large language model to understand the deep-level plot of the script and accurately predict the value of the script, thereby achieving a comprehensive analysis of the script.

[0251] Furthermore, we use the experience of film and television production to fine-tune the large language model and train a special large model for the film and television field. On the one hand, the experience data of film and television production is very valuable, and it takes a long time to accumulate and has a certain threshold. After learning, the large language model can give full play to the value of experience data and serve more film and television workers in an intelligent way; on the other hand, it injects scientific and technological power into the film and television industry and improves the efficiency of film and television production.

[0252] Furthermore, this application uses production experience data such as theoretical works on script writing as one of the training samples of the big language model, which can provide theoretical support for the big language model when evaluating scripts, enhance deep understanding, improve evaluation accuracy and reliability, enrich evaluation dimensions and indicators, assist in generating optimization suggestions, enhance versatility and transferability, and promote self-learning and evolution.

[0253] The script analysis method based on a large language model provided in the present application can be executed by a computing device, which may be a server, a server cluster consisting of multiple physical servers, or a distributed file system, or a cloud server cluster that provides basic cloud computing services such as cloud storage and cloud services, cloud databases, cloud computing, cloud functions, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), big data and artificial intelligence platforms, etc. The present application does not limit this.

[0254] Figure 6 is a schematic diagram of the hardware structure of a computing device provided in an embodiment of the present application, such as Figure 6 As shown, the computing device includes a processor 601, a memory 602, a bus 603, and a computer program stored in the memory 602 and executable on the processor 601. The processor 601 includes one or more processing cores. The memory 602 is connected to the processor 601 via the bus 603. The memory 602 is used to store program instructions. When the processor executes the computer program, all or part of the steps in the above method embodiments provided in the present application are implemented.

[0255] Further, as an executable solution, the computing device may be a computer unit, which may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer unit may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the composition structure of the computer unit is merely an example of a computer unit and does not constitute a limitation on the computer unit, and may include more or fewer components than the above, or a combination of certain components, or different components. For example, the computer unit may also include input and output devices, network access devices, buses, etc., which are not limited in the embodiments of the present application.

[0256] Further, as an executable solution, the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, and uses various interfaces and lines to connect various parts of the entire computer unit.

[0257] The memory can be used to store computer programs and / or modules. The processor realizes various functions of the computer unit by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory 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 required for a function; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0258] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements all or part of the steps of the script analysis method based on the large language model in the embodiment of the present application.

[0259] The present application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements all or part of the steps of the method for script analysis based on a large language model in an embodiment of the present application.

[0260] If the module / unit integrated in the above-mentioned computing unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable media may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory) and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0261] Although the present application has been specifically shown and described in conjunction with the preferred embodiments, those skilled in the art should understand that various changes may be made to the present application in form and details without departing from the spirit and scope of the present application as defined by the appended claims, all of which are within the scope of protection of the present application.

Claims

1. A script analysis system based on a large language model, characterized in that: The system comprises: A data processing module is used to perform feature engineering, data integration and labeling on the acquired multi-class sample data to obtain a feature data set, wherein the multi-class sample data includes script content data, script feedback data and production experience data, and the labeled labels include at least one of the script's emotional information, plot elements and plot structure; A model training module is used to: fine-tune the pre-trained large language model on the feature data set for a preset natural language understanding task to obtain a script analysis model, wherein the script analysis model is used to output at least one of the emotional information, plot elements and plot structure of the script; The script content data and the production experience data in the feature data set are used as input features, the script feedback data in the feature data set is used as a prediction result label, and a given time series neural network is trained to obtain a script value prediction model; The reporting service module is used to: in response to an analysis request for a target script, respectively call the script analysis model and the script value prediction model, analyze the script content and production data of the target script, and obtain the analysis result and feedback data of the target script, wherein the analysis result includes at least one of emotional information, plot elements and plot structure; generate an analysis report of the target script according to the analysis result and feedback data of the target script; wherein the feedback data output by the script value prediction model includes total value prediction, value change trend prediction and key factor prediction; The reporting service module is used to: use the script value prediction model to predict the change trend of market value and the change range of total market value within a given scenario and / or period; Determine the output value of the script value prediction model under different input feature combinations, determine the impact sensitivity of various input features according to the output values ​​corresponding to the different input feature combinations, and determine at least one key factor according to the impact sensitivity; The input features include: production experience data and script content data; the more significant the change in output value caused by the numerical change of the input feature, the stronger the sensitivity of the input feature.

2. The script analysis system based on a large language model according to claim 1 is characterized in that: The reporting service module is also used for: In response to a comparison and modification request for the target script, converting the target script into structured script data according to a target structure template, wherein the structured script data describes various plot elements according to the target structure template; Based on the structured script data, providing an interactive interface for modifying various plot elements of the target script; Update the plot elements of the target script according to the modification operation obtained from the modification interaction interface; Based on the analysis results and feedback data corresponding to the updated target script, a modification evaluation report of the target script is generated.

3. The script analysis system based on a large language model according to claim 1 is characterized in that: The data processing module is used to: Feature engineering: using natural language processing technology to classify the sentiment tendency of the script feedback data; using a given topic modeling algorithm to extract the main topics or elements from the script feedback data; semantically understanding the scenes, characters, and dialogues in the script content data to identify at least one of the plot turning points, plot elements, and plot structure; Converting the production experience data into quantitative features, wherein the quantitative features at least include: amount, cycle, and quantity; Data integration: Integrate the processed script content data, script feedback data, and production experience data into a given database or data framework in a given data format; Data labeling: Manually label some of the script content data, script feedback data, and production experience data. The labels used for labeling include: the quality level of the script content, the emotional tendency of the feedback data, and the discussion topic.

4. The script analysis system based on a large language model according to claim 3 is characterized in that: The time series neural network is any one of a multilayer perceptron, a convolutional neural network, and a recurrent neural network; the model training module is used for: The plot turning points, plot elements, plot structure corresponding to the script content data and the quantitative features corresponding to the production experience data are fused as input features, the emotional tendency and / or discussion topic corresponding to the script feedback data are used as prediction result labels, and the time series neural network is trained by adjusting hyperparameters and cross-validation to obtain a script value prediction model.

5. The script analysis system based on a large language model according to claim 1 is characterized in that: The system further comprises: a data acquisition module; the data acquisition module is used to: Acquire a plurality of script content data from a given network database, website address and / or local database, wherein the script content data includes: one or more of a character description text, a dialogue text, and a narrative text; Acquire multiple script feedback data from a given network database, website address and / or local database according to keywords corresponding to the script, wherein the script feedback data includes one or more of market value, evaluation score, evaluation reason and number of followers; From a given network database, website address and / or local database, production experience data is obtained according to the keywords, and the production experience data includes: plot narrative structure theory, and one or more of production budget, film type, publicity investment, production company, director, screenwriter, starring, whether it is IP adaptation, and industry evaluation.

6. The script analysis system based on a large language model according to claim 1, characterized in that: The pre-trained large language model is: a BERT model, a GPT series model or a Transformer architecture model pre-trained on text data; the model training module is used to: Inputting the feature data set into the large language model; Performing the preset sentiment analysis task, plot element recognition task, and plot structure recognition task output by the large language model, and comparing the results with the preset prompt template to obtain an error, wherein the prompt template indicates the semantic association between the model input and output; The parameters of the large language model are iteratively adjusted according to the error until the error of the large language model in executing the sentiment analysis task, the plot element recognition task and the plot structure recognition task meets the preset target large language model, thereby obtaining the script analysis model.

7. The script analysis system based on a large language model according to claim 6 is characterized in that: The prompt template includes at least one of the following: A. Plot structure analysis template: script content data as input, and multiple plot structure tags as prompt words; B. Rhythm and tension analysis template: The script content data is used as input, and the emotional intensity label, emotional change label, plot advancement label, and plot conflict evaluation label of each scene in the script are used as prompt words; C. Character Depth and Arc Analysis Template: Extract character features and character relationships from script content data, generate a time series in chronological order, and use the time series as input; the character's personality development trajectory label is used as a prompt word; D. Relationship dynamics analysis template: based on the character relationship map in the script content, the character relationship map is used as input; At least one relationship label among the character interaction frequency, emotional tendency, cooperation, and conflict extracted from the character relationship map is used as a prompt word; E. Dialogue quality analysis template: real dialogue data is used as input, and character personality tags are used as prompt words; F. Creativity and originality analysis template: the script content data to be analyzed is used as input, and the similarity between the input content and the scripts in the given script database is used as output; G. Problem and suggestion analysis template: script content data is used as input, and the annotated logical error labels, plot incoherence labels, and modification suggestion text are used as prompt words; H. Market and audience positioning analysis template: script content data and production data as input, audience group labels as prompts; I. Technical and production considerations template: script content data and production data are used as input, and production difficulty labels and budget labels are used as prompts.

8. The script analysis system based on a large language model according to claim 1, characterized in that: The data processing module is further used for: Data preprocessing: After data cleaning and word segmentation, the multi-class sample data are subjected to denoising and irrelevant word deletion, and finally the multi-class sample data are converted from text into a digital encoding form that can be understood by the model.

9. A script analysis method based on a large language model, characterized in that: The method comprises: Performing feature engineering, data integration and labeling on the acquired multi-category sample data to obtain a feature data set, wherein the multi-category sample data includes: script content data, script feedback data and production experience data, and the labeled labels include at least one of the script's emotional information, plot elements and plot structure; For a preset natural language understanding task, fine-tune the pre-trained large language model on the feature data set to obtain a script analysis model, wherein the script analysis model is used to output at least one of the emotional information, plot elements, and plot structure of the script; The script content data and the production experience data in the feature data set are used as input features, the script feedback data in the feature data set is used as a prediction result label, and a given time series neural network is trained to obtain a script value prediction model; In response to an analysis request for a target script, the script analysis model and the script value prediction model are respectively called to analyze the script content and production data of the target script to obtain analysis results and feedback data of the target script, wherein the analysis results include at least one of emotional information, plot elements and plot structure; an analysis report of the target script is generated according to the analysis results and feedback data of the target script; wherein the feedback data output by the script value prediction model includes total value prediction, value change trend prediction and key factor prediction; A reporting service module is used to: use the script value prediction model to predict the changing trend of market value and the changing range of total market value within a given scenario and / or period; Determine the output value of the script value prediction model under different input feature combinations, determine the impact sensitivity of various input features according to the output values ​​corresponding to the different input feature combinations, and determine at least one key factor according to the impact sensitivity; The input features include: production experience data and script content data; the more significant the change in output value caused by the numerical change of the input feature, the stronger the sensitivity of the input feature.

Images