Quality evaluation method and device for plot text generation

Abstract

The invention relates to a plot text generation quality evaluation method and device, and the method comprises the steps: obtaining a machine plot text generated by a text generation model based on a target video, and obtaining a standard plot text of the target video; according to the same granularity splitting rule, splitting the machine plot text into a plurality of machine plot units, and splitting the standard plot text into a plurality of standard plot units; based on a matching result of each machine plot unit and the corresponding standard plot unit, calling a large language model to determine a final matching score of the machine plot text and the standard plot text; and evaluating the quality of the machine plot text based on the final matching score. According to the method, the plot text quality evaluation accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of text generation technology, and in particular to a method and apparatus for quality assessment of plot text generation. Background Technology

[0002] Against the backdrop of continuously growing demand for multimedia content consumption, the need for generating narrative text for long-form videos, as a core content format, is becoming increasingly prominent. Multimodal big data models, with their ability to integrate video footage, audio, and other information, have made significant progress in the field of long-form video narrative text generation. They can quickly convert hour-long videos into narrative description text, providing support for scenarios such as video retrieval and content summarization.

[0003] However, current quality assessment of long-form video narrative texts still faces key bottlenecks. Traditional assessment methods are designed for short texts, and when faced with long narrative texts containing complex scene transitions and character interactions, they struggle to fully cover the content details. Although some studies have attempted to apply Large Language Models (LLMs) to assessment, they have limitations in handling long documents. Their context windows cannot accommodate the complete information of long narrative texts, and they easily overlook key plot points in the middle paragraphs, such as core event turning points and important character actions. This results in a significant deviation between the assessment results and the actual video narrative, making it impossible to accurately assess the quality of the narrative text. Summary of the Invention

[0004] This application provides a method and apparatus for quality assessment of plot text generation to solve the problem of inaccurate plot text assessment.

[0005] Firstly, this application provides a method for quality assessment of plot text generation, the method comprising: Obtain the machine-generated narrative text based on the target video by the text generation model, and obtain the standard narrative text of the target video; Following the same granularity splitting rules, the machine plot text is split into multiple machine plot units, and the standard plot text is split into multiple standard plot units; Based on the matching results of each machine plot unit and the corresponding standard plot unit, the large language model is invoked to determine the final matching score between the machine plot text and the standard plot text; The quality of the machine-generated narrative text is evaluated based on the final matching score.

[0006] Optionally, the machine story text can be divided into multiple machine story units according to the same granularity splitting rules, including: Identify multiple core granularity words in the core granularity layer, multiple secondary granularity words in the secondary granularity layer, and multiple basic granularity words in the basic granularity layer; By traversing the machine's narrative text and marking the position of each core granularity word, core granularity units between two adjacent core granularity words are generated; By traversing the text in each core granularity unit and marking the position of each secondary granularity word, a secondary granularity unit between two adjacent secondary granularity words is generated; By traversing the text in each secondary granularity unit and marking the position of each basic granularity word, basic granularity units between two adjacent basic granularity words are generated, and each basic granularity unit is used as a split machine story unit.

[0007] Optionally, based on the matching results of each machine-generated storyline unit and its corresponding standard storyline unit, the final matching score between the machine-generated storyline text and the standard storyline text is determined by calling a large language model, including: Based on the standard story units matched by each machine story unit, multiple matching unit groups are formed; The large language model is invoked to evaluate each matching unit group on multiple evaluation dimensions, and the unit matching score of each matching unit group under each evaluation dimension is obtained. Calculate the unit matching score of all matching unit groups under each evaluation dimension to obtain the overall text evaluation score of the machine-generated plot text and the standard plot text under the evaluation dimension. The final matching score between the machine-generated plot text and the standard plot text is obtained by weighting the overall text evaluation scores.

[0008] Optionally, based on the standard story units matched by each machine story unit, multiple matching unit groups are formed, including: Determine the timing number of each machine story unit and each standard story unit, wherein the timing number includes the number of the core granularity unit, the number of the secondary granularity unit, and the number of the basic granularity unit. Based on the time sequence number of the machine plot unit, a retrieval-assisted enhancement technique is used to find multiple candidate standard plot units with the same core granularity unit number and the same secondary granularity unit number. The machine-generated story unit is matched with the candidate standard story unit by keywords to determine the target standard story unit with the highest keyword matching degree. The machine-based story unit is paired with the target standard story unit to form a matching unit group.

[0009] Optionally, the evaluation dimensions include content relevance, logical consistency, and detail richness. The unit matching scores for all matching unit groups under each evaluation dimension are calculated to obtain the overall text evaluation scores for the machine-generated narrative text and the standard narrative text under the evaluation dimensions, including: Under the content relevance dimension, the overall text evaluation score under the content relevance dimension is determined based on the sum of the content relevance scores of all matching unit groups and the number of the standard plot units; Under the logical consistency dimension, the overall text evaluation score under the logical consistency dimension is determined based on the sum of the logical consistency scores of all matching unit groups and the number of machine story units. Under the richness-of-details dimension, the overall text evaluation score under the richness-of-details dimension is determined based on the sum of the richness-of-details scores of all matching unit groups and the number of machine narrative units.

[0010] Optionally, after evaluating the quality of the machine-generated narrative text based on the final matching score, the method further includes: If the quality of the machine-generated narrative text is lower than a preset quality level, then the target evaluation dimension causing the anomaly is identified, and the anomaly type under the target evaluation dimension is determined. Determine the model internal parameters corresponding to the anomaly type and adjust the model internal parameters, wherein the model internal parameters corresponding to each anomaly type are not completely the same; The machine story text is regenerated based on the adjusted text generation model, and the newly generated machine story text is re-segmented and its matching score is determined until the quality of the machine story text is higher than the preset quality level.

[0011] Optionally, the core granularity layer represents the plot development, the secondary granularity layer represents the scene changes, and the basic granularity layer represents the character behavior changes.

[0012] Secondly, this application provides a quality assessment device for plot text generation, the device comprising: The acquisition module is used to acquire machine-generated story text based on the target video by the text generation model, and to acquire the standard story text of the target video; The splitting module is used to split the machine plot text into multiple machine plot units according to the same granularity splitting rules, and to split the standard plot text into multiple standard plot units; The determination module is used to determine the final matching score between the machine plot text and the standard plot text by calling the large language model based on the matching result between each machine plot unit and the corresponding standard plot unit. An evaluation module is used to evaluate the quality of the machine-generated narrative text based on the final matching score.

[0013] Thirdly, this application provides an electronic device, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus.

[0014] Fourthly, this application also provides a computer storage medium storing computer-executable instructions for executing a quality assessment method for plot text generation as described in any of the preceding claims of this application.

[0015] Compared with the prior art, the technical solution provided in this application has the following advantages: First, the machine-generated narrative text and the standard narrative text are obtained. Then, the two types of text are split into corresponding units at a uniform granularity, transforming the complex long text evaluation into a structured unit-level evaluation. This avoids the problem that key information in the middle of long texts is easily overlooked, making the evaluation more focused on details and easier to operate. Then, the final matching score is determined based on the unit matching results, intuitively presenting the degree of fit between the machine-generated narrative text and the standard narrative text. Finally, the quality of the machine-generated narrative text is evaluated based on the final matching score. This application improves the accuracy of narrative text quality evaluation by decomposing the long text evaluation into multiple sub-units and by using structured analysis and the assistance of a large language model. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0019] Figure 1 A schematic diagram of a quality assessment system for plot text generation provided in this application embodiment; Figure 2 A flowchart illustrating a quality assessment method for plot text generation provided in this application embodiment; Figure 3 This application provides a schematic diagram of the overall process for quality assessment of plot text generation; Figure 4 A schematic diagram of the structure of a quality assessment device for plot text generation provided in this application embodiment; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0022] Optionally, in the embodiments of this application, the above-described method for quality assessment of plot text generation can be applied to, for example... Figure 1 The hardware environment shown consists of terminal 101 and server 103. Figure 1 As shown, server 103 is connected to terminal 101 via a network and can be used to provide services to the terminal or clients installed on the terminal. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The network mentioned above includes, but is not limited to, wide area network, metropolitan area network or local area network. Terminal 101 includes, but is not limited to, PC, mobile phone, tablet computer, etc.

[0023] The following will describe in detail a quality assessment method for plot text generation provided in this application embodiment, taking its application to a server as an example, with specific implementation details. Figure 2 As shown, the specific steps are as follows: Step 201: Obtain the machine-generated narrative text based on the target video from the text generation model, and obtain the standard narrative text of the target video; Step 202: Following the same granularity splitting rules, split the machine plot text into multiple machine plot units and split the standard plot text into multiple standard plot units; Step 203: Based on the matching results of each machine plot unit and the corresponding standard plot unit, call the large language model to determine the final matching score between the machine plot text and the standard plot text; Step 204: Evaluate the quality of the machine-generated narrative text based on the final matching score.

[0024] In step 201, the server first obtains two types of core text. One type is the machine-generated narrative text output by the text generation model after performing multimodal information parsing on the target long video. This type of text needs to fully cover the scenes, character interactions, and plot development of the long video, with a length of several thousand or tens of thousands of words depending on the video length. The other type is the standard narrative text, which is written by a professional human team and uploaded via the terminal. During the writing process, the human team corresponds to the video content segment by segment, accurately marking scene switching nodes, specific character behaviors, and key plot twists. It also undergoes multiple rounds of cross-validation and final review to ensure complete consistency with the actual video content. The server stores the standard narrative text as the benchmark text for subsequent quality assessment.

[0025] In step 202, the server decomposes the machine-generated storyline text and the standard storyline text according to a unified granularity splitting rule. The splitting rule is centered on the long-video storyline logic and can be divided from the following dimensions: scene-wise, dividing the text fragments corresponding to each independent scene into a unit; user-wise, dividing the text fragments corresponding to each user character's dialogue into a unit; and plot-wise, dividing the text fragments corresponding to the beginning, development, climax, and ending of the event into independent units. The server uses this rule to split the machine-generated storyline text into multiple machine-generated storyline units and the standard storyline text into multiple standard storyline units. Each unit contains key information from the corresponding video fragment, ensuring complete alignment of the two types of units in terms of information coverage and division dimensions.

[0026] In step 203, the server first invokes RAG (Retrieval-Augmented Generation) technology to retrieve the content with the highest semantic relevance to each machine plot unit from the standard plot units, establishing a correspondence between units. Then, it invokes a large language model to evaluate the matched unit pairs based on three dimensions: content relevance, logical consistency, and detail completeness. In terms of content relevance, the large model evaluates whether the machine plot unit accurately covers the core information of the corresponding standard plot unit. In terms of logical consistency, it evaluates whether the plot development of the machine plot unit remains coherent with the standard plot unit. In terms of detail completeness, it evaluates whether the machine plot unit contains character behavior details and scene details from the standard plot unit. Based on the matching results of each unit in the three dimensions, the server calculates the final matching score between the machine plot text and the standard plot text. The large language model can be Gemini; this application does not restrict the specific type of model.

[0027] In step 204, the server first retrieves the calculated final matching score and, in conjunction with a preset quality assessment level system, performs a tiered assessment of the machine-generated narrative text quality. This assessment level system is calibrated based on the actual application scenarios of long-video narrative texts (such as general narrative search and content summarization), and is specifically divided into three core levels: If the final matching score is within the first range, it is judged as excellent, indicating that the machine-generated narrative text is highly consistent with standard narrative texts in terms of content relevance, logical consistency, and detail completeness, and can be directly used in products, such as supporting users' precise narrative retrieval of long videos; if the final matching score is within the second range, it is judged as acceptable, indicating that the machine-generated narrative text basically covers the core narrative of the video, but there are a few missing details or logical connection flaws, requiring manual assistance to supplement and correct before use; if the final matching score is within the third range, it is judged as unacceptable, meaning that the machine-generated narrative text has obvious content deviations (such as missing key plots) or logical inconsistencies (such as reversing the plot order), and cannot meet basic application requirements.

[0028] In addition, the server generates a quality assessment report, which details the final matching score, the corresponding quality level, and the specific scores of each machine-generated story unit in terms of content, logic, and detail. It clearly identifies quality weaknesses; for example, a unit might have a low score in the detail dimension due to missing key character dialogue. This report provides precise guidance for subsequent parameter adjustments to the text generation model, helping technical personnel to optimize the model in a targeted manner.

[0029] In this application, machine-generated narrative text and standard narrative text are first obtained. Then, the two types of text are split into corresponding units at a uniform granularity, transforming the complex long text evaluation into a structured unit-level evaluation. This avoids the problem that key information in long texts is easily overlooked, making the evaluation more focused on details and easier to operate. Then, the final matching score is determined based on the unit matching results, intuitively presenting the degree of fit between the machine-generated narrative text and the standard narrative text. Finally, the quality of the machine-generated narrative text is evaluated based on the final matching score. This application improves the accuracy of narrative text quality evaluation by decomposing long text evaluation into multiple sub-units and by using structured analysis and the assistance of a large language model.

[0030] As an alternative implementation, in step 102, the machine plot text is divided into multiple machine plot units according to the same granularity splitting rules, including the following: Step S11: Determine multiple core granularity words in the core granularity layer, multiple secondary granularity words in the secondary granularity layer, and multiple basic granularity words in the basic granularity layer; Step S12: By traversing the machine's narrative text and marking the position of each core granularity word, core granularity units are generated between two adjacent core granularity words; Step S13: By traversing the text in each core granularity unit and marking the position of each secondary granularity word, generate secondary granularity units between two adjacent secondary granularity words; Step S14: By traversing the text in each secondary granularity unit and marking the position of each basic granularity word, a basic granularity unit between two adjacent basic granularity words is generated, and each basic granularity unit is used as a split machine story unit.

[0031] In step S11, the server clarifies the granularity word categories corresponding to the three granularity layers based on the logical hierarchy of the long video's plot. The core granularity layer revolves around plot description, with core granularity words selecting expressions that define plot stages, covering words that identify key nodes such as the occurrence, development, climax, and ending of events, ensuring a complete outline of the plot's progression. The secondary granularity layer focuses on scene description, with secondary granularity words primarily consisting of expressions related to location, time, and atmosphere. These words accurately distinguish different scene spaces, time stages, and environmental atmospheres within the plot, achieving a clear division of scene dimensions. The basic granularity layer describes character behavior, with basic granularity words selecting expressions related to character actions, dialogue, and expressions, accurately capturing the specific behavioral details of characters within the plot. The server combines manually preset plot element rules with text semantic recognition technology to filter granularity words covering the three descriptive dimensions, constructing a complete granularity word system.

[0032] In step S12, the server initiates a machine-based traversal program for the plot text, scanning the text content segment by segment. When a core granularity word (corresponding to a key node in the plot description) is identified, its position in the text is automatically marked. By continuously marking the positions of two adjacent core granularity words, the text segment between these two marked points is determined as a core granularity unit. Each core granularity unit corresponds to a complete plot stage in the story, covering the complete process of events from occurrence to progression within that stage. This ensures both the continuity of the plot and the initial breakdown of the overall plot framework, allowing the originally lengthy text to form an ordered set of core units according to the plot logic.

[0033] In step S13, for each generated core granularity unit, the server restarts the traversal process, focusing on identifying secondary granularity words (corresponding to key information in scene descriptions) in the text and marking their positions. Using two adjacent secondary granularity words as boundaries, the text within the core granularity unit is split into multiple secondary granularity units. Each secondary granularity unit corresponds to a specific scene in the plot, fully containing the location information, time background, and atmosphere of that scene. This allows the scene switching logic within the core plot stage to be clearly presented through the division of secondary units, achieving further decomposition from the plot framework to scene details.

[0034] In step S14, the server iterates through each sub-granularity unit again, identifying and marking the basic granularity words (corresponding to key content describing character behavior). Multiple basic granularity units are generated, using two adjacent basic granularity words as dividing boundaries. These basic granularity units are then used as the final machine-generated storyline units. Each basic granularity unit focuses on a specific character behavior, fully encompassing the character's actions, dialogue, and facial expressions, accurately capturing the details of character behavior in the storyline. This achieves a deep decomposition from the scene level to the details of character behavior, allowing the decomposed units to accurately correspond to micro-level behavioral fragments in the storyline. Each core granularity unit includes multiple sub-granularity units, and each sub-granularity unit includes multiple basic granularity units.

[0035] This application constructs a decomposition logic based on a three-tiered framework of plot description, scene description, and character behavior description. Through layered decomposition, it transforms lengthy and complex machine-generated narrative text into a clearly structured and information-focused set of units. On one hand, the decomposition process perfectly aligns with the natural compositional logic of long-form video narratives, from the macro-level plot framework to the meso-level scene divisions, and then to the micro-level details of character behavior, achieving precise coverage of all dimensions of narrative information and avoiding the problems of information ambiguity and unclear dimensions found in traditional decomposition methods. On the other hand, each level of unit corresponds to a clear narrative element, laying the foundation for accurate matching with standard narrative units. This ensures that the matching process can unfold from multiple dimensions such as plot, scene, and character behavior, thereby improving the comprehensiveness and accuracy of subsequent quality assessments and providing reliable structured text support for the overall evaluation process.

[0036] In step 103, based on the matching results of each machine-generated story unit and its corresponding standard story unit, the large language model is invoked to determine the final matching score between the machine-generated story text and the standard story text, including the following: Step S21: Based on the standard story units matched by each machine story unit, form multiple matching unit groups; Step S22: Call the large language model to evaluate each matching unit group on multiple evaluation dimensions, and obtain the unit matching score of each matching unit group under each evaluation dimension; Step S23: Calculate the unit matching score of all matching unit groups under each evaluation dimension to obtain the overall text evaluation score of the machine-generated plot text and the standard plot text under the evaluation dimension; Step S24: Calculate the weighted scores of each overall text evaluation to obtain the final matching score between the machine-generated plot text and the standard plot text.

[0037] In step S21, the server assigns a time sequence number containing information from the core granularity layer, secondary granularity layer, and basic granularity layer to each of the split machine-generated storyline units and standard storyline units. Based on the time sequence number of the machine-generated storyline unit, candidate standard storyline units under the same core granularity layer and secondary granularity layer are selected. Then, the target standard storyline unit with the highest overlap with the machine-generated storyline unit is determined through keyword matching. Finally, the two are combined into a matching unit group. This process is based on the hierarchical logic of long-form video storylines, using time sequence numbers to lock in the matching range and using keyword matching to ensure unit correspondence, thus eliminating cross-level and cross-content mismatch interference for subsequent accurate evaluation.

[0038] In step S22, the server invokes the large language model to evaluate each matching unit group from three dimensions: content relevance, logical consistency, and richness of detail. Based on preset evaluation rules, the large language model determines whether the machine-generated story unit covers the core content of the standard story unit, whether the plot order is consistent, and whether key details are complete, thus outputting the unit matching score for each matching unit group in each dimension. This step leverages the powerful semantic understanding and logical analysis capabilities of the large language model to transform abstract quality judgments into quantifiable scores, avoiding subjective biases in manual evaluation and achieving standardization and automation of the evaluation process.

[0039] In step S23, for the content relevance dimension, the server sums the content relevance scores of all matching unit groups and divides them by the total number of standard plot units to obtain the overall text evaluation score for that dimension. For the logical consistency dimension, the overall score is calculated by dividing the sum of the logical consistency scores of all matching unit groups by the total number of machine plot units. For the detail richness relevance dimension, the overall score is obtained by summing the detail richness relevance scores based on the total number of machine plot units. This process transforms local quality performance into a dimensional overall result by statistically integrating unit-level scores, clearly presenting the global performance of machine plot text in each core quality dimension.

[0040] In step S24, the server assigns corresponding weights to the overall evaluation scores of content relevance, logical consistency, and rich detail based on the application scenario requirements of the long-form video narrative text. The final matching score between the machine-generated narrative text and the standard narrative text is obtained through a weighted summation formula of the overall scores and weights for each dimension. This step comprehensively considers the impact of different dimensions on text quality, ensuring that the final score objectively reflects the overall quality level of the machine-generated narrative text and provides a unified and reliable quantitative basis for subsequent quality assessment.

[0041] This application employs a logical chain of unit matching, dimensional scoring, overall integration, and weighted calculation. First, it establishes precise unit correspondences by matching chronological numbers with keywords, ensuring a reliable evaluation foundation. Then, it leverages a large language model to achieve multi-dimensional quantitative scoring, improving the objectivity and efficiency of the evaluation. Next, it integrates dimensional scores to identify quality weaknesses. Finally, it calculates a weighted comprehensive score to fully reflect the text quality. This overall process effectively addresses the problems of vague references, significant subjective bias, and incomplete quality measurement in traditional long-form video narrative text evaluation. It provides precise direction for subsequent text generation model optimization, ultimately helping to improve the quality of machine-generated narrative text, making it more closely aligned with the actual video content and meeting the application needs of scenarios such as narrative retrieval and content summarization.

[0042] As an optional implementation, in step S21, forming multiple matching unit groups based on the standard story units matched by each machine story unit includes the following: Step S211: Determine the timing number of each machine story unit and each standard story unit, wherein the timing number includes the number of the core granularity unit, the number of the secondary granularity unit, and the number of the basic granularity unit. Step S212: Based on the time sequence number of the machine plot unit, use retrieval-assisted enhancement technology to find multiple candidate standard plot units with the same core granularity unit number and the same secondary granularity unit number; Step S213: Match the machine story unit with candidate standard story units using keywords to determine the target standard story unit with the highest keyword matching degree; Step S214: Form a matching unit group by combining the machine story unit with the target standard story unit.

[0043] In step S211, the server constructs a unified time-series numbering rule and performs number allocation for the machine-generated storyline units and standard storyline units that were previously divided into three levels of granularity: core granularity (plot development), secondary granularity (scene transitions), and basic granularity (character behavior changes). The core granularity units are numbered strictly according to the plot progression order, based on key stages of the long video storyline such as the occurrence, development, climax, and ending of events, assigning a unique sequence number to each plot stage to ensure complete synchronization between the numbering and the plot. The secondary granularity units are numbered based on the scene switching logic within the core plot stages. Under the same core granularity number, secondary sequence numbers are assigned to each scene according to the order in which they appear, accurately identifying the scene's hierarchical position within the overall storyline. The basic granularity units are numbered according to the order in which character behaviors occur within a scene. Under the same secondary granularity number, basic sequence numbers are assigned according to the chronological order of character behavior, achieving precise positioning of character behavior segments. By using a three-level nested time sequence numbering system (core, secondary, and basic), the server establishes a unique plot position identifier for each unit, ensuring that the numbering dimension of the machine plot unit is fully aligned with that of the standard plot unit, providing a unified reference standard for subsequent matching processes.

[0044] In step S212, the server initiates a targeted filtering process in the standard plot unit database, using the time sequence number of the machine plot unit as the core search condition. This process first extracts the core granularity unit number and the secondary granularity unit number from the machine plot unit's time sequence number, using these two numbers as search keywords. Then, it invokes search-aiding enhancement technology to filter units in the standard plot unit database with completely identical numbers. Specifically, it retains only standard plot units that are in the same plot stage and scene space as the machine plot unit, directly excluding units that cross plot lines (different core numbers) or cross scenes (different secondary numbers). This filtering logic significantly narrows the matching range through hierarchical limitation, avoiding the waste of computational resources and low matching efficiency caused by full-scale retrieval. Simultaneously, it ensures that the candidate units and the machine plot unit's plot background are highly consistent, laying the foundation for accurate matching at the subsequent content level.

[0045] In step S213, the server initiates a keyword matching process based on the core plot information for both the machine-generated plot unit and the candidate standard plot units. First, text semantic parsing technology is used to extract core keywords reflecting character behavior, key events, and scene details from the machine-generated plot unit. Then, these keywords are compared one by one with the text content of each candidate standard plot unit, and the number of perfectly matching keywords is counted. The candidate standard plot unit with the most matching keywords is selected as the target standard plot unit. If there are cases where the number of matching keywords is the same, the position of the keywords in the text is further compared, and units with a more concentrated keyword distribution are prioritized and selected as the target standard plot unit. This process, through quantitative keyword matching, ensures optimal content-level fit between the standard plot unit and the machine-generated plot unit, avoiding the problem of consistent hierarchy but content discrepancies.

[0046] In step S214, the server structurally binds the machine-generated narrative units with the target standard narrative units, generating matching unit groups containing basic unit information, time sequence numbers, and keyword matching results. The basic unit information includes text content fragments of both types of units and the corresponding timeline range of the narrative; the time sequence numbers indicate the three-level numbers of both types of units to verify hierarchical consistency; the keyword matching results record the list of matched keywords and the matching score, providing a basis for subsequent evaluation. The server organizes and archives all matching unit groups in chronological order, forming a complete unit matching dataset, providing standardized reference samples for subsequent multi-dimensional evaluation using large language models.

[0047] In this application, a multi-dimensional technical improvement is achieved through a full-process design that includes time-series numbering for positioning, candidate range filtering, keyword matching, and unit group binding: First, the three-level time-series numbering solves the problem of ambiguous unit positions in traditional matching, ensuring that machine-generated story units and standard story units are fully aligned at the plot, scene, and behavior levels; Second, the candidate range is narrowed down first through search-assisted enhancement technology to reduce computational costs, and then content fit is improved through keyword quantitative matching, balancing matching efficiency and accuracy.

[0048] As an optional implementation, the evaluation dimensions include content relevance, logical consistency, and detail richness. In step S23, the unit matching score of all matching unit groups under each evaluation dimension is calculated to obtain the overall text evaluation score of the machine-generated plot text and the standard plot text under the evaluation dimensions, including the following: Step S231: Under the content relevance dimension, determine the overall text evaluation score under the content relevance dimension based on the sum of the content relevance scores of all matching unit groups and the number of standard plot units; Step S232: Under the logical consistency dimension, determine the overall text evaluation score under the logical consistency dimension based on the sum of the logical consistency scores of all matching unit groups and the number of machine story units; Step S233: Under the detail richness-related dimension, determine the overall text evaluation score under the detail richness-related dimension based on the sum of the detail richness-related scores of all matching unit groups and the number of machine narrative units.

[0049] In step S231, during the content relevance dimension evaluation, the large language model first retrieves the unit-level scores for all previously matched unit groups in this dimension. These unit-level scores are quantitative values ​​derived from whether the machine-generated narrative unit fully covers the core content of the standard narrative unit. For example, whether the machine-generated narrative unit contains key information such as key events, core characters, and core scenes from the standard narrative unit. The scores are all within the range of [0-1]. Subsequently, the large language model sums up the content relevance scores of all matched unit groups to obtain the total score for this dimension, denoted as S_content.

[0050] Considering that the core evaluation objective of the content relevance dimension is whether the machine-generated narrative text conforms to the core information framework of the standard narrative, and that the standard narrative units are authoritative benchmarks verified by humans, their quantity directly represents the total amount of core content in the standard narrative, the large language model uses the total number M of standard narrative units as the calculation base. The final result is obtained through the formula: Overall evaluation score for the content relevance dimension = S_content / M. This calculation logic ensures that the evaluation result accurately reflects the overall level of the machine-generated narrative text in terms of core content coverage of the standard narrative. For example, if S_content=16 and M=20, the overall evaluation score for the content relevance dimension is 0.8, intuitively demonstrating that the machine-generated text covers 80% of the core content of the standard narrative, avoiding evaluation bias caused by fluctuations in the number of machine-generated narrative units.

[0051] In this process, the user inputs the prompt word "Prompt" into the Large Language Model (LLM), which guides the model to evaluate the overall text score of the matching unit group across any dimension. For example, the prompt word could be: "Please evaluate the content relevance between the machine-generated story unit and the standard story unit based on the following criteria, and output a quantitative score of 0-1 (1 for a perfect match, 0 for a complete mismatch)."

[0052] In step S232, for the logical consistency dimension, the large language model first collects the unit matching scores of all matching unit groups in this dimension. These scores mainly measure the degree of fit between machine-generated story units and standard story units in terms of plot development order and causal coherence. For example, they assess whether machine-generated story units have issues such as plot reversal, causal breaks, or logical contradictions. A [0-1] scoring system is also used (1 point represents complete logical consistency, and 0 points represent severe logical inconsistency). Next, the large language model sums the logical consistency scores of all units to obtain the total score for this dimension, denoted as S_logic.

[0053] Since the evaluation focus of the logical consistency dimension is whether the machine-generated narrative text itself is logically coherent and conforms to the natural development of the plot, and the number of machine-generated narrative units N directly reflects the total number of logical fragments in the machine-generated text, the large language model uses the total number of machine-generated narrative units N as the calculation base, and calculates the final result using the formula: Logical Consistency Dimension Overall Evaluation Score = S_logic / N. For example, if S_logic = 14 and N = 20, the overall evaluation score for the logical consistency dimension is 0.7, clearly indicating that the average coherence of the machine-generated narrative text at the logical level is 70%, which objectively reflects the logical integrity of the machine-generated text itself, rather than simply relying on the logical framework of standard narrative text.

[0054] In step S233, in the evaluation of the detail richness-related dimension, the large language model first extracts the unit scores for all matching unit groups in this dimension. These scores focus on the machine's ability to reproduce the micro-details of the standard plot unit, including the coverage of detailed elements such as character expressions, action details, scene atmosphere descriptions, key dialogue fragments, and prop features. The scoring range is [0-1] (1 point represents complete detail coverage, and 0 points represents no key details). Then, the large language model sums the detail richness-related scores of all units to obtain the total score for this dimension, denoted as S_detail.

[0055] Considering that the core of detail richness is the performance level of machine-generated narrative text in terms of detail output, the evaluation should focus on the actual detail coverage of the machine text. Therefore, the large language model also uses the total number of machine narrative units N as the calculation base, and obtains the final result through the formula: Overall evaluation score of detail richness-related dimensions = S_detail / N. For example, if S_detail=12 and N=20, the overall evaluation score of detail richness-related dimensions is 0.6, which intuitively reflects that the average level of detail reproduction of machine-generated narrative text is 60%, accurately reflecting the machine text's ability to capture micro-level information of the narrative, and providing a quantitative basis for subsequent judgments on whether the machine text is vivid and specific.

[0056] As an optional implementation, after evaluating the quality of the machine-generated narrative text based on the final matching score, the method further includes the following: Step S31: If the quality of the machine-generated narrative text is lower than the preset quality level, identify the target evaluation dimension that caused the anomaly and determine the anomaly type under the target evaluation dimension. Step S32: Determine the model internal parameters corresponding to the anomaly type and adjust the model internal parameters. The model internal parameters corresponding to each anomaly type are not exactly the same.

[0057] Step S33: Regenerate the machine story text based on the adjusted text generation model, and re-segment and determine the matching score of the newly generated machine story text until the quality of the machine story text is higher than the preset quality level.

[0058] In step S31, when the server determines that the quality of the machine-generated narrative text is lower than the preset quality level based on the final matching score (e.g., a final matching score < 0.6, corresponding to an unqualified level), the anomaly tracing analysis process is initiated first. The server retrieves the overall text evaluation scores for the three evaluation dimensions of content relevance, logical consistency, and rich detail from previous evaluations. By comparing the scores of each dimension with the corresponding sub-preset thresholds (e.g., sub-threshold 0.7 for content relevance, 0.65 for logical consistency, and 0.6 for rich detail), the dimensions with scores lower than the sub-thresholds are identified as the target evaluation dimensions causing the quality anomaly. For example, if the overall score for the content relevance dimension is 0.58 (lower than the sub-threshold 0.7), while the scores for other dimensions meet the standard, then the content relevance dimension is determined as the target evaluation dimension.

[0059] Subsequently, the server further analyzed the anomaly types under the target evaluation dimensions: For the content relevance dimension, by retrospectively examining the content relevance scores of each matching unit group, the anomaly types were identified as omissions of core events (e.g., multiple units did not cover key events in the standard plot) or deviations of core characters (e.g., the main character in a unit does not match the standard plot); for the logical consistency dimension, by checking the unit sequence and the causal relationship of the plot, the anomaly types were determined to be reversed plot order (e.g., the unit sequence conflicts with the standard plot sequence) or broken causal logic (e.g., the lack of reasonable causal connections between units); for the detail richness dimension, by comparing the detail coverage of units, the anomaly types were determined to be missing character behavior details (e.g., the key actions of characters in the standard plot were not described) or missing scene details (e.g., the scene feature information in the standard plot was not mentioned). Through the two-step analysis of dimension positioning and type subdivision, the core causes of quality anomalies were accurately identified.

[0060] In step S32, the server, based on the identified target evaluation dimension and anomaly type combination, calls a preset anomaly type and parameter mapping library. This mapping library is built based on a large amount of model training data and quality optimization cases, and clarifies the internal parameters of the model that need to be adjusted for each anomaly type. For example, if the anomaly type is the omission of core events under the content relevance dimension, the corresponding parameter to be adjusted is the weight coefficient of event keywords in the model's attention mechanism. Increasing this weight enhances the model's ability to capture core events in the plot. If the anomaly type is the reversed plot order under the logical consistency dimension, the corresponding adjustment is made to the model's context window length and the decay coefficient of the temporal modeling module. The window is expanded to accommodate a longer plot logic chain, and the decay coefficient is optimized to strengthen the plot temporal memory. If the anomaly type is the lack of character behavior details under the detail richness dimension, the model's generation temperature parameter and the trigger threshold for detail description generation are adjusted. Lowering the generation temperature reduces the randomness of details, and lowering the trigger threshold increases the probability of detail generation.

[0061] During parameter tuning, the server strictly adheres to the principle of differentiated tuning: the direction and magnitude of parameter adjustments differ for each anomaly type. For example, missing core events requires increasing the attention weight (adjustment magnitude +20%), while core role deviation requires optimizing the parameters of the role recognition module (adjustment magnitude +15%), avoiding performance fluctuations caused by indiscriminate parameter tuning. Simultaneously, the server records the specific values ​​before and after parameter adjustments (e.g., attention weight 0.3 before adjustment, 0.36 after adjustment), providing data support for subsequent parameter optimization effect tracking and iteration.

[0062] In step S33, after the parameters are adjusted, the server controls the adjusted text generation model to reread the multimodal data (images, audio, etc.) of the target video and regenerate the machine story text according to the original generation logic. Subsequently, the server performs unit splitting (splitting into machine story units according to the three-level granularity rule) and matching score determination (constructing matching unit groups, calculating multi-dimensional scores, and weighting to obtain the final matching score) on the newly generated machine story text, and re-evaluates its quality level.

[0063] If the quality of the new machine-generated narrative text is still lower than the preset quality level, the server returns to step S31, re-identifies the target evaluation dimension and anomaly type (which may be an unresolved original anomaly type or a newly emerging anomaly type), and matches and adjusts the parameters again based on the new anomaly situation; if the quality of the new text is higher than the preset quality level, the server stops iterating, saves the currently adjusted model parameter configuration as the optimal parameter scheme, and records the anomaly type, parameter adjustment, and quality improvement data of this optimization, updates the anomaly type and parameter mapping library, and continuously improves the model optimization system.

[0064] In this application, on the one hand, by precisely identifying the target evaluation dimensions and anomaly types, the problem of blind parameter tuning in traditional optimization is avoided, allowing parameter adjustments to focus on the core causes of quality anomalies and improving model optimization efficiency. On the other hand, based on the anomaly type and the parameter mapping library, differentiated parameter adjustments ensure that each quality problem has a targeted solution, avoiding the randomness and repetition of parameter adjustments. At the same time, through iterative verification and mapping library updates, the accuracy and stability of model optimization are continuously improved. The overall process forms a closed loop of quality assessment, problem localization, parameter optimization, and quality reassessment, effectively driving the quality of machine-generated narrative text to continuously approach the preset level, ensuring that the model output can stably meet the quality requirements of scenarios such as long video narrative retrieval and content summarization.

[0065] This application provides a schematic diagram of the quality assessment process for plot text generation, such as... Figure 3 As shown, the process steps include: Step 3001: Obtain the machine story text and the standard story text.

[0066] Step 3002: Determine the granular terms corresponding to the three levels of granularity: core granularity, secondary granularity, and basic granularity.

[0067] Step 3003: Split the machine plot text into machine plot units according to the three-level granularity layer, and split the standard plot text into standard plot units.

[0068] Step 3004: Assign timing numbers to machine story units and standard story units according to the three-level granularity layer.

[0069] Step 3005: Select candidate standard plot units corresponding to machine plot units according to the same core granularity unit number and secondary granularity unit number.

[0070] Step 3006: Determine the target standard plot unit that matches the machine plot unit based on the keywords.

[0071] Step 3007: Form a matching unit group between machine story units and target standard story units.

[0072] Step 3008: Obtain the unit matching score for each matching unit group in terms of content relevance, logical consistency, and rich detail relevance.

[0073] Step 3009: Calculate the overall text evaluation score for each evaluation dimension.

[0074] Step 3010: Calculate the final matching score between the machine plot text and the standard plot text based on the weighted average of the overall text evaluation scores under each evaluation dimension.

[0075] Step 3011: Evaluate the quality of the machine-generated narrative text based on the final matching score.

[0076] Based on the same technical concept, this application provides a quality assessment device for plot text generation, such as... Figure 4 As shown, the device includes: The acquisition module 401 is used to acquire the machine-generated plot text based on the target video by the text generation model, and to acquire the standard plot text of the target video; The splitting module 402 is used to split machine plot text into multiple machine plot units and standard plot text into multiple standard plot units according to the same granularity splitting rules. The determination module 403 is used to determine the final matching score between the machine plot text and the standard plot text based on the matching result between each machine plot unit and the corresponding standard plot unit by calling the large language model. Evaluation module 404 is used to evaluate the quality of machine-generated narrative text based on the final matching score.

[0077] Optionally, the split module 402 is used for: Identify multiple core granularity words in the core granularity layer, multiple secondary granularity words in the secondary granularity layer, and multiple basic granularity words in the basic granularity layer; By traversing the machine's narrative text and marking the position of each core granularity word, core granularity units between two adjacent core granularity words are generated; By traversing the text in each core granularity unit and marking the position of each secondary granularity word, a secondary granularity unit between two adjacent secondary granularity words is generated; By traversing the text in each secondary granularity unit and marking the position of each basic granularity word, basic granularity units between two adjacent basic granularity words are generated, and each basic granularity unit is used as a split machine story unit.

[0078] Optionally, the determining module 403 is used for: Based on the standard story units matched by each machine story unit, multiple matching unit groups are formed; The large language model is invoked to evaluate each matching unit group on multiple evaluation dimensions, and the unit matching score of each matching unit group under each evaluation dimension is obtained. Calculate the unit matching score of all matching unit groups under each evaluation dimension to obtain the overall text evaluation score of machine-generated plot text and standard plot text under the evaluation dimension; The final matching score between the machine-generated plot text and the standard plot text is obtained by weighting the overall text evaluation scores.

[0079] Optionally, module 403 is specifically used for: Determine the timing number for each machine story unit and each standard story unit, where the timing number includes the number of the core granularity unit, the number of the secondary granularity unit, and the number of the basic granularity unit. Based on the time sequence number of the machine plot unit, a retrieval-assisted enhancement technique is used to find multiple candidate standard plot units with the same core granularity unit number and the same secondary granularity unit number. The machine-generated storyline unit is matched with candidate standard storyline units by keywords to determine the target standard storyline unit with the highest keyword matching degree. The machine-based story units are matched with the target standard story units to form a matching unit group.

[0080] Optionally, module 403 is specifically used for: Under the content relevance dimension, the overall text evaluation score is determined based on the sum of the content relevance scores of all matching unit groups and the number of standard plot units. Under the logical consistency dimension, the overall text evaluation score is determined based on the sum of the logical consistency scores of all matching unit groups and the number of machine-generated narrative units. Under the detail richness-related dimension, the overall text evaluation score is determined based on the sum of the detail richness-related scores of all matching unit groups and the number of machine narrative units.

[0081] Optionally, the device is also used for: If the quality of the machine-generated narrative text is lower than the preset quality level, the target evaluation dimension that causes the anomaly is identified, and the anomaly type under the target evaluation dimension is determined. Determine the model's internal parameters corresponding to each anomaly type and adjust the model's internal parameters accordingly. Note that the model's internal parameters are not exactly the same for each anomaly type. The machine-generated story text is regenerated based on the adjusted text generation model, and the newly generated machine-generated story text is re-segmented and its matching score is determined until the quality of the machine-generated story text is higher than the preset quality level.

[0082] Optionally, the core granularity layer is the plot development, the secondary granularity layer is the scene changes, and the basic granularity layer is the character behavior changes.

[0083] like Figure 5 As shown, this application provides an electronic device including a processor 501, a communication interface 502, a memory 503, and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504.

[0084] Memory 503 is used to store computer programs.

[0085] In one embodiment of this application, the processor 501, when executing a program stored in the memory 503, implements a quality assessment method for plot text generation provided in any of the foregoing method embodiments.

[0086] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of a quality assessment method for plot text generation as provided in any of the foregoing method embodiments.

[0087] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0088] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0089] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0090] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for quality evaluation of scenario text generation, characterized in that, The method comprises: obtaining a machine plot text generated by a target video based on a text generation model, and obtaining a standard plot text of the target video; splitting the machine plot text into a plurality of machine plot units and splitting the standard plot text into a plurality of standard plot units according to the same granularity splitting rule; based on the matching result of each machine plot unit and the corresponding standard plot unit, calling a large language model to determine the final matching score of the machine plot text and the standard plot text; based on the final matching score, evaluating the quality of the machine plot text.

2. The method of claim 1, wherein, Splitting the machine plot text into a plurality of machine plot units according to the same granularity splitting rule comprises: determining a plurality of core granularity words of a core granularity layer, a plurality of secondary granularity words of a secondary granularity layer, and a plurality of basic granularity words of a basic granularity layer; by traversing the machine plot text and marking the position of each core granularity word, generating a core granularity unit between adjacent two core granularity words; by traversing the text in each core granularity unit and marking the position of each secondary granularity word, generating a secondary granularity unit between adjacent two secondary granularity words; by traversing the text in each secondary granularity unit and marking the position of each basic granularity word, generating a basic granularity unit between adjacent two basic granularity words, and taking each basic granularity unit as a split machine plot unit.

3. The method of claim 1, wherein, Based on the matching result of each machine plot unit and the corresponding standard plot unit, calling a large language model to determine the final matching score of the machine plot text and the standard plot text comprises: forming a plurality of matching unit groups according to the standard plot units matched by each machine plot unit; calling a large language model to evaluate each matching unit group in multiple evaluation dimensions to obtain a unit matching score of each matching unit group in each evaluation dimension; calculating the unit matching scores of all matching unit groups in each evaluation dimension to obtain an overall text evaluation score of the machine plot text and the standard plot text in the evaluation dimension; weighting calculating each overall text evaluation score to obtain the final matching score of the machine plot text and the standard plot text.

4. The method of claim 3, wherein, Forming a plurality of matching unit groups according to the standard plot units matched by each machine plot unit comprises: determining the time sequence number of each machine plot unit and each standard plot unit, wherein the time sequence number comprises the number of core granularity units, the number of secondary granularity units, and the number of basic granularity units; according to the time sequence number of the machine plot unit, using retrieval auxiliary enhancement technology to find a plurality of candidate standard plot units with the same core granularity unit number and the same secondary granularity unit number; performing keyword matching between the machine plot unit and the candidate standard plot unit to determine a target standard plot unit with the highest keyword matching degree; forming a matching unit group from the machine plot unit and the target standard plot unit.

5. The method of claim 3, wherein, The evaluation dimensions include a content correlation dimension, a logic consistency dimension, and a detail richness correlation dimension, and a unit matching score of all matching unit groups under each evaluation dimension is calculated to obtain an overall text evaluation score of the machine plot text and the standard plot text under the evaluation dimension, which includes: Under the content correlation dimension, an overall text evaluation score under the content correlation dimension is determined according to a sum value of content correlation scores of all matching unit groups and a number of the standard plot units; Under the logic consistency dimension, an overall text evaluation score under the logic consistency dimension is determined according to a sum value of logic consistency scores of all matching unit groups and a number of the machine plot units; Under the detail richness correlation dimension, an overall text evaluation score under the detail richness correlation dimension is determined according to a sum value of detail richness correlation scores of all matching unit groups and a number of the machine plot units.

6. The method of claim 1, wherein, After evaluating the quality of the machine plot text based on the final matching score, the method further includes: If the quality of the machine plot text is lower than a preset quality level, a target evaluation dimension causing an abnormality is identified, and an abnormality type under the target evaluation dimension is determined; Model internal parameters corresponding to the abnormality type are determined, and the model internal parameters are adjusted, wherein the model internal parameters corresponding to each abnormality type are not completely the same; The machine plot text is regenerated based on the text generation model after the parameters are adjusted, and the matching score of the newly generated machine plot text is determined again until the quality of the machine plot text is higher than the preset quality level.

7. The method of claim 2, wherein, The core granularity layer is a plot development, the secondary granularity layer is a scene transformation of the plot, and the basic granularity layer is a role behavior change of the plot.

8. A quality evaluation device of scenario text generation, characterized by, The device includes: An acquisition module configured to acquire a machine plot text generated by a text generation model based on a target video and acquire a standard plot text of the target video; A splitting module configured to split the machine plot text into a plurality of machine plot units and split the standard plot text into a plurality of standard plot units according to the same granularity splitting rule; A determination module configured to determine a final matching score of the machine plot text and the standard plot text based on a matching result of each machine plot unit and a corresponding standard plot unit by calling a large language model; An evaluation module configured to evaluate a quality of the machine plot text based on the final matching score.

9. An electronic device, comprising: The device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is configured to store a computer program; The processor is configured to execute the program stored in the memory to implement the method in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1-7. The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1-7.

Images